Wikipedia:Village pump (technical)

 Policy Technical Proposals Idea lab WMF Miscellaneous 

• Table of contents
• First discussion
• Jump to bottom
• New topic
• All village pumps

The technical section of the village pump is used to discuss technical issues about Wikipedia. Bug reports and feature requests should be made in Phabricator (see how to report a bug). Bugs with security implications should be reported differently (see how to report security bugs).

If you want to report a JavaScript error, please follow this guideline. Questions about MediaWiki in general should be posted at the MediaWiki support desk. Discussions are automatically archived after remaining inactive for 5 days.

« Archives, 207, 208, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227

Early Explorations Into Semantic Search: Phase 0

Hi everyone,

We’re sharing an early look at an exploration into improving how people search on Wikipedia in order to gather your input. The goal is to help readers find the information they’re looking for more easily on Wikipedia itself, without needing to rely on external search engines.

One focus of this work is semantic search, a type of search that looks at the meaning of a query, not just the exact words typed, to help people find information resources. Today, Wikipedia search relies almost entirely on keyword matching, which works well when readers know the exact article they want, but less well when they have a question or are exploring a topic and the answer is inside an article without a keyword title match.

This post outlines why we’re exploring this, what our early research shows, and what kind of small, early experiment we’re considering.

Why are we working on this?

Many readers do not start their searches on Wikipedia. Instead, they often use external search engines or AI-powered tools, which then direct them to Wikipedia – or sometimes provide answers based on Wikipedia content without sending readers to the site at all.

If Wikipedia search does not meet modern expectations, especially for question-based or exploratory searches, readers are less likely to begin or continue their journeys here and instead rely on platforms where information isn’t made by humans, and is less reliable, neutral, and complete.

In short, improving search is one way to help Wikipedia readers find and enjoy what they read on our platform.

What has our preliminary research shown?

Our early research checked whether this problem is real and whether improving search could meaningfully help readers. Our findings suggest that it could.

1. About 98% of Wikipedia reading sessions originate outside Wikipedia search.

• The small group who use internal search are much more likely to be editors than casual readers. Most readers move between articles by returning to external search engines, even when links exist within Wikipedia itself.

2. Roughly 80–95% of on-wiki search sessions use autocomplete suggestions.

• The preference for autocomplete suggestions – those that appear as someone types – shows that small improvements to speed can have a large impact on success.

3. Between 4–7% of Wikipedia search queries are phrased as questions, but these queries are less likely to succeed.

• While this is a minority of searches, it shows that some readers attempt it and that many others likely avoid it because they’ve learned it doesn’t work.

What stage is this project in?

This work is currently in Phase 0, sharing early ideas, learning from research, and gathering community input.

What idea are we testing?

We’re exploring whether a hybrid search experience, one that combines keyword search with semantic search, could help readers find information more easily. The hybrid search would use machine learning, similar to how search engines rank and surface results today, to better match readers’ queries with relevant existing articles and sections.

Semantic search performs better for questions and exploratory searches, while keyword search works better for very specific or name-based queries. In early prototypes, combining the two approaches produced more useful results than either one alone.

Importantly, this exploration does not involve generating new answers or rewriting Wikipedia content. The goal is to better match readers’ queries to existing, editor-created articles and sections. Any experiment in this area would be small, limited, and designed to test whether this approach provides real value to readers.

What is the timeline?

Right now, we’re focused on discussing the problem space and sharing the findings of the report with you all. We especially want to understand if this problem space is worth learning more about. We are also trying to better understand if a simple Minimal Viable Product could be technically feasible.

Should there be alignment around further exploration, a possible next step would be a tightly constrained, time-bound A/B test with a limited group of readers, potentially beginning in February, to help answer open questions surfaced together.

What input are we looking for from you?

We’d especially appreciate your thoughts on the following:

1. What are your overall reactions to this exploration and the research behind it?
2. Are there risks or concerns you think we should be paying closer attention to?
3. What signals or outcomes would matter most in deciding whether a hybrid search approach is worth pursuing further?

For more details, including links to our research and early mockups, please see the project page.

Thank you! EBlackorby-WMF (talk) 16:09, 6 January 2026 (UTC)

Google is not a good role model here, given the amount of negative media coverage they have received in the last few years about how they are currently destroying their flagship product, in part by messing with people's search queries. Bringing this up as something Google "excels at" is actually baffling given how constant and how inescapable the negative commentary has been. (But I guess these are just "Internet pundits" and the feature isn't for them.)

Also, based on the project page, this seems to go well beyond "improving search." The mockups that claim to be "semantic search" appear to actually illustrate a "Because you liked..." recommendation feature, and a "suggested questions" widget, which surfaces AI-generated questions for no apparent reason. (For Q&A use cases, AI-generated questions can potentially introduce factual or representational bias. -- you don't say!)

(Neither here nor there: I'm pretty sure that the project page was at least edited, if not fully generated, with AI.) Gnomingstuff (talk) 04:41, 7 January 2026 (UTC)

@Gnomingstuff Google is good at semantic interpretation (and many other things as well). It's not as good at giving the right answer as it USED to, but that has more to do with the overall pool of garbage that they have to sort through, not keeping up with development of their own platform's core functionality, as well as actively worsening their quality because they want to keep you coming back to their website.

On a scale of 1 through 10, they are however still at 7. If people want to focus on Google messing up the 7-10 part, then they are ignoring that WE are at 0 and not even close to their 7.

this seems to go well beyond "improving search."

I think that depends on if your definition of search is that of someone used to old style search (a specific word matching technology), or the actual meaning for most people in the world (finding what they are looking for). —TheDJ (talk • contribs) 14:50, 7 January 2026 (UTC)

The post here seems to only discuss "old style" search: We’re exploring whether a hybrid search experience, one that combines keyword search with semantic search, could help readers find information more easily.. What the Phase 0 proposal page seems to actually be proposing are "since you liked..." and AI-generated questions widgets, with proof-of-concepts already built out.

I also guess I don't see the problem with people getting to Wikipedia pages via Google rather than internal search or blue links. They still get to the page either way, the only difference is squeezing out 2 extra pageviews." Gnomingstuff (talk) 20:21, 7 January 2026 (UTC)

I feel WMF developers should be informed somehow that any LLM-generated text placed within the Wikipedia UI is unlikely to receive a positive community reaction (either due to anti-AI sentiment or cautious acceptance but feeling Wikipedia's human-written nature is what gives it a purpose distinct from ChatGPT etc.). Otherwise they will continue to spend effort on stuff that is unlikely to be accepted by the editing community.

On the core idea of semantic search, I somewhat disagree. Yesterday I was searching for some New Zealand-related topics. I typed NZ {phrase}, but it didn't come up with the correct article; although it works often enough for me to instinctively do it, stuff like this still fails for me a reasonable portion of the time. Having a search system that handles this sort of thing would be a better way of handling this rather than creating redirects for every conceivable topic.  novov talk edits 09:28, 8 January 2026 (UTC)

"with proof-of-concepts already built out" proof of concepts get built all the time for a variety of reasons. to spark discussion, to visualize ideas etc etc. Stop telling people what to do. —TheDJ (talk • contribs) 12:00, 8 January 2026 (UTC)

...I didn't tell anyone what to do? Pointing out the contents of a page is not telling anyone what to do and I have no idea how you are twisting "proof of concepts are built out" into "I order you to do XYZ."

...also, the entire point of this topic was to ask for feedback, in part to help decide what to do Gnomingstuff (talk) 14:31, 8 January 2026 (UTC)

Gnomingstuff, I'm pretty sure that the project page was at least edited, if not fully generated, with AI. - No, I'm pretty sure it isn't. WP:AISIGNS is not a good metric for finding AI-gen pages outside of encyclopedic articles. A lot developers write with bolded phrases as signposts and bulleted lists since it makes it easier to skim documents. with proof-of-concepts already built out I can personally tell you that those prototypes could be built at speed, over a evening if required. (in fact, if I were to speculate, [1] (which is their other design) took more time to create than the AI question screenshot).

That being said,

EBlackorby-WMF, wrt What signals or outcomes would matter most in deciding whether a hybrid search approach is worth pursuing further? I do want to amplify one specific criticism by @Gnomingstuff and @Mir Novov, I do think The robot icon and AI label failed to clarify what was machine- vs human-generated and instead increased confusion. is a alarming finding that should have been a guard-rail, we cannot have Although the questions were AI-generated, participants often assumed questions were crowdsourced or editor-curated be a thing. The (enwiki-) community cares deeply about making sure folks understand that the encyclopedia is human-generated and organic and having some mixing of AI-gen content/attribution of human-gen content to AI will likely be poorly recieved by the community.

Moving to my personal feedback, I really really like "Concept 1" (the "because you read" feature). It's something I've been missing a fair bit. I really dislike the Q/A interfaces (mostly due to the reasons that folks in the thread have outlined). I'm personally ambivalent to the ask.toolforge.org prototype, though I wonder if instead to prompting the question, the interface could silently engage the reader to page snippets? (For example, iff a user types in "Company that owns biggest browser", y'all could silently switch from using CirriusSearch to surfacing Google#Google Chrome or similar). Sohom (talk) 14:19, 8 January 2026 (UTC)

I specifically had in mind stuff like "underscoring the need for transparent provenance labeling" and "align with Wikimedia’s infrastructure and privacy standards." As I said it's neither here nor there, this kind of document at any workplace is more likely than not to be AI-assisted and there's no rule against it. Gnomingstuff (talk) 14:33, 8 January 2026 (UTC)

I wonder if the AIs learned to write that way from the kind of techno-corporate bureaucratese that certain types of people have been writing for a long time? Anomie⚔ 00:06, 9 January 2026 (UTC)

(Probably, which is why I suspect no AI usage, that language is a bit corporate-speaky, but I've seen it used in earnest were folks were not using GPT) Sohom (talk) 10:51, 9 January 2026 (UTC)

@Sohom Datta and @Mir Novov, I appreciate your notes and acknowledge your concern around perceived mixing of AI- and human-generated content, as it's one we take very seriously as well. User testing showed that the bot labeling was confusing and would have to be completely rethought if Q&A explorations are found to be worth continuing. Right now it's clear that the current iteration of Q&A is not in production- or even an experiment-ready state.

However, there does seem to be interest in improvements to the search to support semantic-style queries. If this continues to be the case, we would return with a proposal for an experiment around search improvements and establish guardrails and goals together. Similarly, if there is another concept for improving wayfinding (that wouldn’t lead to confusing AI- with human-generated content) we’d return again to discuss before proceeding.

Interesting idea on silently engaging the reader to page snippets! Would this look like directly taking the user to the relevant heading instead of the top of the article? Do you have any other suggestions for areas for us to look into for information-finding? EBlackorby-WMF (talk) 20:47, 9 January 2026 (UTC)

@EBlackorby-WMF It would definitely be interesting if the search could send you to a specific heading in a article or even use tech like scroll-to-text-fragments to highlight areas in a article relevant to the user's queries. Sohom (talk) 19:30, 13 January 2026 (UTC)

@Gnomingstuff, thanks for your feedback. It helped us realize we uploaded a still of the design concept instead of a GIF, fixed here, which may have led to confusion about this work focusing on “Because You Read”-style suggestions.

While Google is definitely not a role model for this work, it is the most popular global search engine for people to discover content from Wikipedia, thus making it an important tool for understanding modern reader expectations. We’re trying to see how we can support the kind of queries users are used to making.

As for people coming to Wikipedia from Google, readers are now often simply reading the previews provided by Google Knowledge Panels and AI summaries (or their LLM of choice), rather than visiting the articles themselves. This tendency could harm their understanding of both the article content and the Wikipedia project, as well as prevent them from potentially becoming editors.

Do you have any other thoughts on ways to improve findability of article information for readers? EBlackorby-WMF (talk) 20:39, 9 January 2026 (UTC)

• Those of us who (sadly) recall the early days of DWIM know that this can be a very ambitious project. Please accept that you will find it hard to get the "super goal" of fully understanding what a user wants, but that you can achieve a good deal by even simple and clever means. So please have 3 or 4 goals of increasing complexity and do the simplest one first. A trivial example is keyword similarities such as fast/quick. Of course, before all else you need to improve the page Semantic search! It is in hopeless shape. Last talk comment was in 2019. Looking back, in the late 1980's I met a young investment banker who was somewhat obsessed with funding companies to do semantic search. Then around 2007 or so I happened to see him again, he had become a venture capitalist, and was trying to fund a company to do... guess what, semantic search. The LLMS do some such things but that should be your 3rd project, I suggest. So please do proceed, but step by step. Thanks

Yesterday, all my dreams... (talk) 22:02, 18 January 2026 (UTC)

1. It's an important R&D subject. One main problem users have is finding tutorials, help pages, policies and guidelines using the search. That is for example because these don't show in autocomplete and the default search results. Please also look into this wish proposal for something parallel to the conventional search as it could greatly help newcomers find the relevant tutorial and guideline pages (or other meta pages and articles/sections too).
2. Rather than just semantic search things please also work on and consider things like better integration and awareness-raising about other things users could use to quickly find what they're looking for such as the deepcategory search operator along with awareness and skills in knowing & finding categories. Another concern is that too little attention is paid and priority given to what the real-world users of Wikipedia in real-world practice needs or would like to have.
3. Maybe average reads per day and/or time spent on site mainspace articles increasing or reader surveys improving. Difficult question but basically whether it's a real-world improvement to readers in practice.

Prototyperspective (talk) 22:15, 20 January 2026 (UTC)

Really interesting point about helping newcomer editors find what they need for help. I'll share that back with the team.

We're definitely thinking about ways to make good use of the deepcategory search operator, there's such a wealth of organized information in there that readers could benefit more from.

Average reads and time spent are good ideas to track usefulness.

Do you have any thoughts about potential design considerations to help readers/newcomer editors know the sorts of things they could ask in a search with semantic capability? EBlackorby-WMF (talk) 21:35, 27 January 2026 (UTC)

Thanks! Btw, I'm thinking of maybe making a separate wish that is more specific in some ways (e.g. the application area such as only for helping newcomers instead of the myriad of listed ways this could be used) and broader in other ways (e.g. not specifying WikiChat in title).

Agree. Various things could be made possible/accessible that's based on it such as making it easier to filter an existing search or having the semantic search make use of it (e.g. searching in deepcat:"Films" if the user is clearly searching for a film) but so far I've mainly just explored its practical use on Commons. One could also show a button like "Search help and meta pages" near the top that searches again for just these namespaces because all the namespaces things are unknown and confusing to readers – e.g. they don't know these pages are not by default included in the search results. Also of note is that help pages are often very long with so much info that is helpful for various specific applications that readers landing there would have a hard time finding it or discovering it – this is what the wish is about to a large part but it's probably a bit too long or not written enough for people to see the large potential there and probable need for this (+other approaches addressing this).

Design considerations: when entering into the search bar show one stylized dropdown item like "Ask Wikimedia" which if selected/clicked makes the search use semantic search. At the top show the info that as a short/quick way to use this enter ? or q followed by the question in natural language. To raise awareness about this search mode one could consider changing the default label from "Search Wikimedia" to "Search or ask Wikimedia". Then there's of course also various ways to tell newcomers that they can search for guidance/helpful info relating to their editing. In part related to that, one could make short videos that explain in bite-sized pieces how to edit Wikipedia and start them off by entering a question into the search bar like "q how to make a video start at a certain timestamp" which then would show the relevant wiki help page section that explains how to do it. One could also show a search input box for just help, policy, meta pages etc in the editing window (or a wikilink to a help hub page which contains this searchbox in the center). Prototyperspective (talk) 16:15, 28 January 2026 (UTC)

Notified: Help talk:Searching, Help talk:Navigation, Help talk:Introduction to navigating Wikipedia/1 and mw:Help talk:CirrusSearch. Sdkb‑WMF ^talk 19:56, 11 January 2026 (UTC)

Working on a better search system is a good idea, question based search has been a my staple on Google for 20 years and now I use LLMs. Take the question how is gps accurate enough for surveying (how can the same system that says I am sitting my neighbour's living room be used with any accuracy). Wikipedia has the answer in Surveying, in the History section, 20th century subsection. Or we have an article: Real-time kinematic positioning, that is too complicated to answer the question efficiently. Neither Surveying or Real-time kinematic positioning appear in the top 20 results of a Wikipedia search (Toronto Marathon does though :-P). I tried answering this question more than 15 years ago on the Reference desk (I can't find it in my contribs now, kudos to anyone that can).

When someone has a question, they don't want a search to tell them "the answer may be in this 7,000 word article, have a look". Or they ask a LLM "write me 7,000 word passage that overviews the topic related to my question".

Wikipedia does topic overviews. Wikipedian's can make it easier to find information by placing anchors in article introductions to relevant sections and making articles readable for the general population. And they can participate in a human-written q&a system.

Side note: recently I used Google to search for perth contours (to make a map for a Wikipedia article). The entire screen on the first results page was dedicated to skin contouring clinics around Australia. When I scrolled down after the two results wanting me to pay, the CC BY-4.0 contours were found by searching the gov website that was 4th result. I am more than happy to write an answer to that question for the next person so they can avoid my terrible experience, naturally that is outside the scope of Wikipedia. Google search is vulnerable at the moment.

To answer to your initial questions:

1. I am not surprised that 98% of visits originate outside Wikipedia search (search engines offer better results), and people using Wikipedia search have learnt that it is only good for auto-completing a topic title.
2. Risks: Wikipedia provides articles that overview a topic and the current search box is effective at getting you to an article. If people want to find a Wikipedia article this behaviour with auto-complete is great. Unfortunately it is not good for everything else.
3. I think a decent search should be pursued regardless (to be honest I want a Google competitor that doesn't try to sell me stuff or generate erroneous answer), I am not sure what other options apart from hybrid there are. If hybrid can get me to the answer my question above faster, that is great.

Commander Keane (talk) 10:07, 12 January 2026 (UTC)

@Commander Keane, Thanks for this feedback! Yes, we think for now, hybrid search is the logical next step to explore. Keyword search definitely has its uses so we want to ensure we are building upon our existing capabilities. We’ll share more details on the initial experiment soon.

To your second point on risks, do you have any ideas for wayfinding within an article? EBlackorby-WMF (talk) 18:25, 13 January 2026 (UTC)

1. "What are your overall reactions to this exploration and the research behind it?" → Chat-style search is going to be a necessary part of Wikipedia as that becomes more and more how people expect to engage with the internet. Younger kids shout questions to Alexa or Siri without ever surfing the web. Early Wikipedia used a model that barely made use of search at all; this just feels like a logical step.
2. "Are there risks or concerns you think we should be paying closer attention to?" → A reader should always know upfront if any writing or content was generated by AI.
3. "What signals or outcomes would matter most in deciding whether a hybrid search approach is worth pursuing further?" → Do the changes result in people using Wikipedia's native search more or less? As search engines and AI software become increasingly eager to quote or even plagiarize Wikipedia, we need local solutions because without readers actually seeing the articles, none of them will ever become editors.

I liked the highlighting in the prototype. Cat is also a good example as it's 7853 words long and much of the specific cat information is spread across other articles. For example, I was recently trying to remember what to make sure to avoid when getting floor cleaner to use around cats. The answer (phenols) is on Wikipedia but located at Disinfectant § Phenolics.

On the search results example, the results copied over from cat have their citations omitted:

Cats have excellent night vision and can see at one sixth the light level required for human vision. This is partly the result of cat eyes having a tapetum lucidum, which reflects any light that passes through the retina back into the eye, thereby increasing the eye's sensitivity to dim light. Large pupils are an adaptation to dim light. The domestic cat has slit pupils, which allow it to focus bright light without chromatic aberration. [...]

Have you all tested ways to show the citations in the search?

Cats have excellent night vision and can see at one sixth the light level required for human vision. This is partly the result of cat eyes having a tapetum lucidum, which reflects any light that passes through the retina back into the eye, thereby increasing the eye's sensitivity to dim light. Large pupils are an adaptation to dim light. The domestic cat has slit pupils, which allow it to focus bright light without chromatic aberration. [...]

References

Good luck with the project, Rjjⁱⁱⁱ (talk) 14:25, 13 January 2026 (UTC)

@Rjjiii, appreciate your thoughtful comment. To your third point, our hope is that the changes would result in more readers using Wikipedia’s native search, so that is a key area we will be testing here.

Your note about citations within surfaced results is really helpful. The linked example is a very early mockup and does not reflect any sort of final design choice, so citations will be carefully considered. In addition to highlights, is there anything else you would be interested in seeing in terms of surfacing in-article results? EBlackorby-WMF (talk) 18:28, 13 January 2026 (UTC)

@EBlackorby-WMF: It makes sense to get feedback early on. One note regarding citations is that a study on how folks read Wikipedia showed that people were checking the citations in medical articles (via the popup) even when they were not going to the linked website.[2] I bring it up because I think it would affect how any testing is done. Few people follow citation links, but apparently, in context-dependent situations, some people are checking the citation itself to evaluate the type of source. Regarding "surfacing in-article results", I hesitate to suggest anything here because an important thing make clear to the reader is what has been added by their search plus the AI tool. Highlighting has long been used by search engines, so that's intuitive. For other elements, I suspect you would need to do some kind of test where you ask readers afterward to see if they were confused or mistaken. Rjjⁱⁱⁱ (talk) 18:57, 13 January 2026 (UTC)

Yes, I believe hybrid search is worth exploring. As an editor I use site full-text search to find related and duplicate material, and I'd like to be able to run longer and more natural-language queries and still get meaningful results. I'd also like to see better snippets on search results pages. It's interesting to see this question because a work teammate and I recently implemented hybrid semantic search to improve information retrieval for users of an internal reference website. For that website's purpose, AI-generated text is not desirable/appropriate, but users needed better relevance rankings for search results. We were using PostgreSQL full-text keyword search, and we added pgvector and embeddings generated with a commercial general-purpose foundation model, using reciprocal rank fusion to blend the results. (Switching to Elasticsearch was out of budget/scope.) To figure out whether hybrid search was worthwhile for us, I made a sample list of recent real queries and realistic queries, and we did a lot of qualitative testing to figure out whether the new approach returned similar or better results for the majority of sample queries, rarely worse results. We used a prototype where we could fiddle with the parameters, similar to https://semantic-search.wmcloud.org/. I'm no expert in any of this, but I learned a few things that may be interesting:

1. For my site, semantic search works better for longer and more complex queries, and keyword search works better for single-word and quoted queries. So for one-word and quoted queries, we just run keyword search, and for queries longer than ~15 words, we just run semantic search.
2. Snippet quality and presentation have a big impact on making search results meaningful and usable; people want to see a quick preview of their keywords in context in the document. We kept a relatively traditional search results page, but started offering multiple snippets and longer snippets from each document.
3. It wasn't easy to figure out a good chunking strategy for our documents, but chunking made a big difference for quality of ranking and snippets, both for traditional keyword search and semantic search.
4. We've had difficulty getting really good semantic search ranking with our very DIY approach. Some keyword queries that did well in traditional search got worse with early versions of hybrid search because the semantic search results were not helping. So we limited the impact of the semantic element by limiting the distance threshold. Tested a lot of queries to figure out the best threshold for us. We found that the right threshold lets semantic search shine when keyword search is struggling to produce any relevant results, without muddying up search keyword search results when they're quite relevant.
5. On the search results page, we changed the search input box a little bit - made it a multi-line entry box, with new labeling, to encourage longer queries and more natural-language queries.

Dreamyshade (talk) 06:52, 14 January 2026 (UTC)

• Coming late to this but it seems worth commenting on. First, broadly, I'd agree a hybrid search has major upsides if implemented well. For me, Wikipedia is already faster and easier than finding (accurate) information on google, and I would like that to be true for as many people as possible. It feels especially important as search engines seem likely to generate less and less traffic for us, even as they use our content for AI-powered answers. To skip ahead, the most major risks are fairly self evident: don't make the search worse for current users. In gauging success, I think something like clickthrough rate is much more important than session length or "depth", because it's well known most sessions are very short and that should be taken as a sign of Wikipedia's usefulness to people in the everyday, not as an opportunity for more "stickiness" or "user engagement".

With that out of the way, I wanted to express some mild puzzlement at the motivation and logic behind this project. You ask Why do readers find it easier to locate Wikipedia information elsewhere than on Wikipedia itself? It's a worthy question, but while I haven't been able to comb through all the research in detail, I haven't seen anything yet that points directly toward semantic search. It's certainly true that major search engines have semantic search, which certain users probably find convenient, but there are many other differences those platforms have that could plausibly make readers prefer them. For example, if I happen to open wikipedia on my mobile browser instead of the app, the local search bar becomes inaccessible once I scroll down to read the article. I can access duckduckgo with a tiny swipe at any time, making it far faster. I don't know that this is a surmountable technical problem, but I was just a bit surprised that alternate explanations or solutions to these discrepancies didn't appear to have been considered. Similarly, the degree to which this functionality is in-demand seems rather uncertain: apparently roughly 5% of queries use natural language. To me, that indicates the vast majority of current users probably don't need semantic search, and you should be very careful not to ruin things for them with this experiment. To be sure, 5% is not nothing, but it makes me wonder if this is really mission-critical, or if someone just noticed that google has semantic search and we don't.

Again, I'm really not opposed to exploring this, the justification just seems a little weak, and when we have critical tasks starving for attention and resources, I think it's fair to ask these kinds of questions. —Rutebega (talk) 08:08, 25 January 2026 (UTC)

@Rutebega Similarly, the degree to which this functionality is in-demand seems rather uncertain: Wikimedia currently does not have a "good" way to figure what the readers want and whether building a system is going to have the readers use it. Given that we have a decline in readership specifically because AI engines keeping eating our impressions, I (personally) feel like exploring possibilities to engage readers is more important, since if they go away, we might as well leave as well.

If you have alternate ideas, of things that will help readers engage with Wikipedia further feel free to share them, or if there are critical tasks starving for attention and resources that you want fixed feel free to share them at the annual plan talk page. Sohom (talk) 12:29, 25 January 2026 (UTC)

@Sohom Datta, I wouldn't at all dispute that exploring possibilities to engage readers is more important now. I had hoped it was clear from my comment that I do support this exploration. Eliza asked for overall reactions, and I was responding to what I see as a potential opportunity for more explorations that could address the same root problems. I also wanted to give Eliza and the project team a chance to explain more about their work and rationale, since there can be a big knowledge gap and at times a sort of language barrier when it comes to communications about these kinds of initiatives, at least for someone like me with no real background in MediaWiki.

I did not want to veer off-topic by bringing up other ideas for improvements, as I don't think this discussion was soliciting those. Since you asked though, I was reminded of a very petty grievance I have that nonetheless could be investigated as related to this topic: I have always been totally mystified that Vector 2022 moved the search bar to the upper left corner of the UI, far from all the other controls. I continued using Vector 2010 for several years mostly because of this, but support for dark mode and other features eventually convinced me to put up with it. I have to imagine some UX research was done at the time, but I have a hard time imagining why anyone would prefer it this way, and it's not aligned with other sites I regularly use, where search is almost always top right or centered. Obviously this does not qualify as an issue of critical importance, but I think it's worth paying attention to how our existing UI design impacts some of the user behavior we are trying to modify, and I may try to raise that appropriately regarding the annual plan. Thanks for engaging, as always. —Rutebega (talk) 19:22, 25 January 2026 (UTC)

@Rutebega That is not a bad shout to be honest! I do think making the search bar more prominent by putting it in the middle is a good idea! Sohom (talk) 20:04, 25 January 2026 (UTC)

@Rutebega Thanks for your detailed and thoughtful comment. Your question about how we arrived at semantic search as the answer to this problem is a good one! The simplest response is that based on the qualitative research we’ve done, we landed on semantic search as one possible solution to this problem among many, and one that seemed the most reasonable starting point. There are a variety of other improvements that could and should be made to the search experience on Wikipedia as well. In particular, improving the UI has a lot of potential (your sticky search bar idea is intriguing!).

Some highlights from the literature review that might interest you:

• User studies have shown that readers have specific expectations about Wikipedia’s search functionality from their experience with other platforms and which are not met by our current search. We believe at least part of this is due to lack of semantic search capabilities on our sites.
• A major pain point relayed by all language wiki readers during the course of this research was the inability to properly find information on Wikipedia due to lack of natural language search, as well as having to put in multiple search prompts before getting to desired information.
• Research about discovery needs of in-depth readers highlights that readers prefer external search engines over WP search because the former offers advanced search capabilities (e.g. semantic search via natural language queries).

One relevant quote from our Readers Foundational Research: “Wikipedia search forces you to almost have the exact wording that is inside of the Wikipedia pages… You can't just ask a question like you do on Google and get the Wikipedia searches that are relevant to that question.”

Basically, we think the decline in direct traffic paired with people using external sources for meaning-based queries to navigate to our content are strong signals that we are underperforming in this area. That said, the only way to determine if semantic capability can meaningfully move quantitative numbers is in an experiment to test it with Wikipedia traffic which is what we hope to learn in this exploration.

Thanks again for your feedback. Do you have any other ideas on potential improvements to the problem for us to look into? EBlackorby-WMF (talk) 23:27, 28 January 2026 (UTC)

Google is a terrible model for a search engine; from the perspective of the user, they have been destroying the search engine for decades to please their customers. "Remember that you are not the customer, but the product." A good design goal for a user-oriented search engine is to avoid false positives. Google used to have syntax for "this must be in the candidate" and "this must not be in the candidate", but they removed the support in order to increase their hit counts. I urge that any new search engine provide an easy way to specify

1. Must include this string as is: no changes in spacing, no changes word order, no substitution of synonyms
2. Must not include this string
3. Nice to have: Must include a string matching this regex
4. Nice to have: Narrow the results of the previous stringsearch

The scoring used to train any LLM used for the search should prioritize accuracy over hit count. -- Shmuel (Seymour J.) Metz Username:Chatul (talk) 10:01, 25 January 2026 (UTC)

Thanks @Chatul. The notes around specifications and scoring are really helpful, I'm passing along to our engineers for consideration. Are there any other guardrails you'd recommend keeping in mind through this work? EBlackorby-WMF (talk) 21:43, 27 January 2026 (UTC)

• Helping people find information on Wikipedia is a good aim. Using machine learning methods of natural language processing and semantic search is a reasonable way to approach that problem. But I would highlight, as a crucial design goal, that Wikipedia should allow people to search with natural-language queries without making them feel like they are using an LLM chatbot. That bar of "feel like" means avoiding even aesthetic details that might remind people of chatbots. Those little chat popup interfaces on every website are right out. I like bringing readers to a highlighted section of an article, rather than ever providing a response in text (even a response that is a quote from the article). There are probably many ways to achieve a useful interface that still reassures people that they are on Wikipedia, and not the rest of the internet, so they are getting human-written answers. ~ le 🌸 valyn (talk) 23:54, 26 January 2026 (UTC)

  • look at how archive.org uses metadata for search; e.g.:(Category:Foo)
  • how would I find pages with these properties: American + photography + Italian name + people + fashion

  Piñanana (talk) 12:44, 3 February 2026 (UTC)

  Like so: deepcategory:"American people" deepcategory:"Photography" deepcategory:"Italian-language names" deepcategory:"Fashion".
  I previously left a comment about the deepcategory search operator and potential/needs for improvements regarding it; categorization can also be improved (exa mpl es). Prototyperspective (talk) 13:06, 3 February 2026 (UTC)

  Agree on categories I think the research team have to specify lots of scenarios similar to what you have mentioned. Categories on wikipedia serve little purpose because there are so many unions, all maintained by off wiki batch processes. The major user of categories is wikidata. Is there any evidence that readers navigate by category?

  There are a few UX issue.

  - As mentioned by others, It risks creating an AI layer over the Wikipedians work (see simple summary)

  - makes Wikipedia risk feeling like google AI (answers your spelling mistakes) or grok (guesses what you meant) or Facebook (spooky guesses what you meant based on your searches)

  - Wikipedia UX does not reward experience, exploration, or allow preference (except some skin options).

  - Experience - More advanced search methods are not visible on mobile, and can only be chosen after a search on desktop. SQL/quarry, special search, key words, and regex are invisible.

  - Exploration - Readers expect the left pane to have tools, seperated under clear headings with expandable sub headings. There are two search options visible on the UI (search bar and left pane - portal). a search heading with the options underneath would be far more interesting, and allow links visualisation tools of navigation.

  - Preference - If this is rolled out as the default, then there will be community and reader pushback. ASD needs preference and control of their space. (sed Vector 2022). I am suggesting an editor preference profile on village pump ideas Wakelamp (talk) d[@-@]b 13:30, 3 February 2026 (UTC)

  are operators and parameters conflated ?

  • at Help:Searching#Spaces : "operators, such as intitle: and incategory:"
  • later at Help:Searching#Parameters : "The main parameters are namespace:, intitle:, insource:, incategory:, and prefix: (namespace as used here isn't literal – use the name of the actual namespace desired)."

  is deepcategory a search operator/parameter like insource ?

  examples:

  • for example, find pages that link to filmreference.com ( task: remove Reftags to filmreference.com )

  Special:Search/insource:www.filmreference.com

  • for example, find pages that link to imdb.com ( task: remove Reftags to imdb.com )

  https://en.wikipedia.org/w/index.php?search=insource%253A%22cite%20web%7Curl%253Dhttp%253A%252F%252Fwww.imdb.com%22

  Wikipedia Search: insource:"cite web|url=http://www.imdb.com"

  (I cant make: [[:Special:Search/insource:"cite web|url=http://www.imdb.com"]]

  Piñanana (talk) 15:08, 3 February 2026 (UTC)

  Categories serve lots of purposes such as finding articles and creating lists. They could be used more and this would require some technical improvements – for example a cat-exploring module or integrating them better into into the search filter options. Disagree that the major user of categories is wikidata, that's just one way these are useful. Of course do readers use categories, the question is how many / could more be done to make more people aware of these and how they could use these.

  What's proposed here would not be "creating an AI layer" and isn't like the simple summary. I wonder if you actually read the thread starting comment. Moreover, even if it would be you would need to explain what you mean by that and more importantly where you see why problems. Regarding Wikipedia feeling like this or that, maybe that's exactly what people want (/need). Again more assumptions. But there also the search would show results just like before but also enable this for additional results (or via a new separate search mode(?)). Also all AIs I tested manage to deal with your typos and show you things based on a typos-corrected query.

  Wikipedia UX does not reward experience, exploration sounds interesting but what do you mean? Nevertheless, seems offtopic here(?) SQL/quarry, special search, key words, and regex are invisible. you describe a reason for why semantic search could/would be useful. Regex,SQL/quarry is not sth people can just easily enter; people can't just write some SQL and regex when searching sth on the go on mobile or even in general. Semantic search can make use of such things in simple terms (make them accessible to real people in the real world en masse). If this is rolled out as the default the plan isn't to replace the normal default search. I'd oppose that too.

  @Piñanana: your question is a bit unclear to me but I'll try to answer it; re is deepcategory a search operator/parameter like insource ? as said earlier, it's a search operator (a powerful too-unknown underused one). It's not much like insource if that's what you were asking. I cant make: [[:Special:Se… See Template:Search link. examples: I don't know how these relate to the topic; that's use-cases for the insource search operator. Prototyperspective (talk) 15:31, 3 February 2026 (UTC)

  @Commander Keane ~2026-77304-4 (talk) 16:26, 4 February 2026 (UTC)

  The title is "Early Explorations Into Semantic Search: Phase 0", so I included all the search methods for comparison, as readers might prefer an advanced search similar to the Internet archive and journals. A hybrid search method with a single search input is best for looking at a new subject, while an advanced search is much better at detail.

  • Category Using 2021 server data, Large Scale Characterisation of how readers browse Wikipedia found that only 0.39% (Desktop) and 0.2% (laptop), Readers include editors. And Categories are not visible on the Android App by default. There are 2.4 million categories.
  • "What's proposed here would not be "creating an AI layer" and isn't like the simple summary. I wonder if you actually read the thread starting comment. " .. : I did, but the researchers are suggesting a combination of semantic (which today means LLM AI) and lexcia lexical. A hybrid search method with a single search input is best for looking at a new subject, while an advanced search is much better at detail. Editors and Dev forget, that the current advanced is specified as a default in preferences for logged in editors , while other users have to "access advanced search on Wikipedia, type any query into the search bar, press enter, and then click the "Advanced" option on the search results page to filter by namespace"
  • " SQL/quarry, special search, key words, and regex are invisible" see [Help:Searching]] as are the[lexical and advanced search options on preferences
  • Exploration - rewarding reader through discoverability of features [3] goes hand in hand with search. (greyed out menu items was how I discovered hidden features in systems) Wikipedia stickiness is based on going down rabbit holes. This suggestion is aimed at making the experience easier which will mean less clicks, while we should be aiming the various search options more visible to differentiate us from external search.
  • Glad we agree on preferences :-) But if it's on for all readers, then I am not certain what happens when they become editors
  • Lastly, the worst case for me is that this is brought in using wikimedia software rather than using OpenSearch or similar. New features are liked by Mediawiki's customers, but having everything done in house increases up maintenance dev and increases the risk (bugs, security) because it is not used elsewhere Wakelamp (talk) d[@-@]b 07:33, 10 February 2026 (UTC)

Watchlist labels

Hello. Suddenly we've something new. Whe you check your watchlist, a box comes up saying 'create & asign labels'. GoodDay (talk) 15:19, 2 February 2026 (UTC)

Yes, confusing. Doug Weller talk 16:24, 2 February 2026 (UTC)

Adding div.mw-special-watchlistlabels-onboarding { display:none; } in a new line to your common.css page, e.g. this edit, will make it go away. Writ Keeper ⚇♔ 16:28, 2 February 2026 (UTC)

Is there a page that explains labels? I clicked past it because I was looking for a specific thing this morning. Now I don't know what they are or how to use them. I know I didn't ask for it, but they don't care if we ask for "features," so my opinion on that doesn't matter. --Onorem (talk) 16:39, 2 February 2026 (UTC)

There is a help page at mw:Help:Watchlist_labels. William Avery (talk) 17:08, 2 February 2026 (UTC)

Indeed, if you visit the new tab, that help page is linked at upper right. --Redrose64 🌹 (talk) 20:20, 2 February 2026 (UTC)

There's really no need to hide it with CSS. It's just an informational popup that shows up only once. – SD0001 (talk) 20:38, 2 February 2026 (UTC)

First the orange bar notification is taken from us & now this new feature. Who's behind these changes, that nobody asked for? GoodDay (talk) 16:55, 2 February 2026 (UTC)

I can't answer that, but I like this feature. I plan to use it to divide up my watchlist into groups -- articles I'm watching because I worked on them; pages related to the bot I operate; FAC/GAN related pages; and everything else. I didn't ask for this but now I see I think it'll be very handy. Mike Christie (talk - contribs - library) 17:19, 2 February 2026 (UTC)

Multiple watchlists is something that various editors have requested for a long time; see m:Community Wishlist Survey 2017/Watchlists/Multiple watchlists, for instance. isaacl (talk) 17:32, 2 February 2026 (UTC)

without community consent? Features like this one are why we have WP:CONEXEMPT. Please stop complaining here unless you actually have a bug to report. Izno (talk) 17:41, 2 February 2026 (UTC)

I’ve had multiple watchlists for a long time, I forget which script. Doug Weller talk 18:32, 2 February 2026 (UTC)

Make it go away, the feature is useless for me, with more than 10k articles I'm not going to add labels now. fifteen years ago, perhaps. - Walter Ego 18:36, 2 February 2026 (UTC)

Make'em consensus-required. These sudden changes are annoying. GoodDay (talk) 19:15, 2 February 2026 (UTC)

Not having any changes is also annoying. People whining is annoying, people not writing articles is annoying, ppl getting shot in the streets is annoying, deep dish pizza is annoying, my niece is annoying. And that freaking piece of gravel in my shoe right now. I just choose to give priority to just a few and ignore the rest. —TheDJ (talk • contribs) 20:28, 2 February 2026 (UTC)

Hey! Deep dish pizza is amazing! ~2025-38536-45 (talk) 18:11, 4 February 2026 (UTC)

I wonder what Wiktionary's rules are about jargon. wikt:en:ANSI standard pizza is a redlink there. WhatamIdoing (talk) 22:53, 5 February 2026 (UTC)

I don't know much about Wiktionary's rules either so I might be wrong but I'm pretty sure the relevant rule is wikt:en:WT:SOP. If a term has a meaning that cannot be inferred from the individual words that compose it then it is included. Warudo (talk) 00:48, 7 February 2026 (UTC)

I for one would be against that, as nothing would ever get done.  novov talk edits 21:23, 2 February 2026 (UTC)

@GoodDay There have been ~200 code changes today (Feb 2). Asking for feedback from the community on all 200 of them in not scaleable in any way shape or form. This is not to mention that asking for community feedback on whether to fix production outages/revert bad code or even (say) translate a part of the language is not beneficial to the community or the developer base.

For what it's worth, this feature is something that the community has been wanting since 2008. There have been entries for it in the 2017community wishlist where it has seen community support and even on the new wishlist the community has shown support for WMF to work on this issue. I understand that you feel frustrated that you haven't been consulted and that your workflow has been impacted (in which case you are free to leave feedback on things that can be fixed) but I don't think the criticisim that the Foundation is adding things without community consensus holds water in this case. Sohom (talk) 23:49, 2 February 2026 (UTC)

We've already had "two" surprises, this new year. I hope we don't have anymore via sudden changes. GoodDay (talk) 23:54, 2 February 2026 (UTC)

I guarantee you that there will be more. Set your expectations accordingly. Wikipedia is always changing, as are most things in life. – Jonesey95 (talk) 00:16, 3 February 2026 (UTC)

All that's really changed is that your watchlist has one extra tab compared to how it was yesterday. You can safely ignore this new tab. The popup tells you about it. To make the popup go away: click Next and then Got it. Job done. --Redrose64 🌹 (talk) 19:23, 2 February 2026 (UTC)

If this is a change "that nobody asked for" then why does WP:PERRENIAL#Multiple watchlists exist? People here on the English Wikipedia have been asking for this since 2008. I for one am already using it. Warudo (talk) 20:28, 2 February 2026 (UTC)

Warudo, you have forgotten: requests for visible software changes don't count unless "I" asked for them. WhatamIdoing (talk) 22:56, 5 February 2026 (UTC)

Note that if watchlist filtering doesn't seem to be available and you want to enable it, go to Preferences / Watchlist / Advanced options and uncheck the "Use non-JavaScript interface" option. MANdARAX • XAЯAbИAM 21:51, 2 February 2026 (UTC)

Paging Samwalton9 (WMF) who has provided some updates at Wikipedia:Help desk § Watchlist labels. ClaudineChionh (she/her · talk · email · global) 23:24, 2 February 2026 (UTC)

Additional thoughts: I was aware of this feature being developed though I can't recall where I first saw it, and I certainly didn't know it was going to go live yesterday. I want to suggest that in future, WMF developers put some thought into how new features or other highly visible changes are introduced to this Wikipedia. For example, post a notice on WP:VPT or WP:VPW, with at least 48 hours' notice, ideally at least one week's notice (even power users have days off), outlining the following (one or two sentences per dot point may suffice):

• We're going to roll out (new feature) on (date)
• Why we're doing this
• What this will look like / How this might impact your current workflow
• Link to an up-to-date documentation page on MediaWiki/MetaWiki
• Link to local documentation page that will need to be updated

As someone who IRL has had to introduce new processes into communities of experienced and opinionated people, I get that not every new feature needs community consensus but I'm not surprised there has been pushback here. ClaudineChionh (she/her · talk · email · global) 00:33, 3 February 2026 (UTC)

@ClaudineChionh My personal reading is that folks would have been here regardless of whether this was announced a week ago in Tech News (which tbf is the thing that was missing). Most of the complaints are about a three clicks dismissable tutorial on how the watchlist works. Personally I really don't see how developers will "show" most editors this feature exists given that most non-technical users probably don't read Tech/News or even VPT for that matter. Sohom (talk) 13:42, 3 February 2026 (UTC)

Two clicks, actually. --Redrose64 🌹 (talk) 20:56, 3 February 2026 (UTC)

Oh I get it, there will always be some complaints whenever anything changes, but there's a part of me that wants to have a notice or documentation to point at and say "we were told about this weeks ago". In this case, seeing that it's a new watchlist feature, would a watchlist notice have been appropriate? ClaudineChionh (she/her · talk · email · global) 22:02, 3 February 2026 (UTC)

@ClaudineChionh, I thought watchlist notices were some kind of enwiki-specific thing, given that they need a gadget to make them function. Though looking at it now it just loads the extension, so maybe not. Qwerfjkltalk 22:11, 3 February 2026 (UTC)

@Samwalton9 (WMF) Claudine's bullet points above are spot-on. David10244 (talk) 05:17, 7 February 2026 (UTC)

• Yes, just make it go away until community consensus can be obtained for what seems like a useless box and a major intrusion for tens of thousands of users everytime they look at their watchlist (and no, most won't know to "Keep on clicking until it goes away"). WMF make-work project? Randy Kryn (talk) 00:45, 3 February 2026 (UTC)

  The only thing that's different from my Watchlist view is the addition of a "Manage labels" link next to the "Edit watchlist" link. Besides that new and nonintrusive link, I don't see that anything else has changed...? Some1 (talk) 04:40, 3 February 2026 (UTC)

  He's complaining about the popup box introducing the feature. As Redrose64 said, clicking "next" and "got it" will make it go away. But maybe the devs can add a "dismiss" button on the first popup if some folks find it that annoying. – SD0001 (talk) 06:55, 3 February 2026 (UTC)

I've updated the section header to be more useful, since as pointed out, this was something that many people have been asking for for two decades. Legoktm (talk) 04:28, 3 February 2026 (UTC)

Please remember to leave an anchor when you do this so that incoming links continue to work. I have added one to this section. – Jonesey95 (talk) 06:33, 3 February 2026 (UTC)

The big issues are lack of an agreed business change/introduction process for UX, and lack of acknowledgment that Wikipedians prefer to have a preference. On the idea forum, I am suggesting that there be an editor preference profile that would allow editors to mark themselves as early adopters, or new editors, .... It would allow you to have different setups for different purposes, and this watchlist issue would not have affected those editors marked essential only/wait 12 months. Wakelamp(wakelamp) 11:54, 3 February 2026 (UTC)

I don't see why there's all the drama. It's not as if they took away a useful feature, or have forced us to use a new feature that makes everything more difficult. Both of these have happened before, with no means to go back to the old way, but that's certainly not the case here. --Redrose64 🌹 (talk) 20:56, 3 February 2026 (UTC)

@Wakelamp, the "early adopted" thing already exists (and has existed since late 2013). Go to Special:Preferences#mw-prefsection-betafeatures and tick the box for "automatically enable". WhatamIdoing (talk) 22:58, 5 February 2026 (UTC)

The "manage labels" dropdown is dropped down every time I go to my watchlist, cluttering my view and covering up the watchlist options. I do not want this feature. Is there some way to make it go away? —David Eppstein (talk) 18:26, 3 February 2026 (UTC)

@David Eppstein, the second response in this thread is some css to make it go away. Qwerfjkltalk 21:05, 3 February 2026 (UTC)

Thanks! Although I found Redrose's instructions on how to make the popup go away adequate enough. —David Eppstein (talk) 21:20, 3 February 2026 (UTC)

Hi all - I'm the product manager who has been supporting this Community Tech project, thanks for sharing all your thoughts and feedback on this feature. Apologies for the confusion around the onboarding popup - we wanted to let folks know that this new feature was available and provide a brief orientation to introduce how it works, but I can totally see how it's annoying that the popup comes back if you don't make it to the end and click 'Got It'. The plan is to remove this popup again in the future, we just wanted to let existing active editors know about the new functionality.

I wanted to let you know that CommTech is continuing to work on this feature, and up next plan to allow you to assign labels while adding a page to your watchlist via the Watch star (T415173), while saving a page (T415251, T415633), and via the API (T416291). There are also plans to update the Help link in the top right of Watchlist pages to point to more useful pages (T416293), given the discussion about documentation here and elsewhere. Other feature requests, like searching by title on Special:EditWatchlist are being discussed and considered.

If you have any feedback, bug reports, or feature requests for the Watchlist Labels feature itself, I'd love to hear what you have to say! Samwalton9 (WMF) (talk) 11:54, 4 February 2026 (UTC)

Wikipedia has suddenly started opening the labels tab when I open my watchlist. Can someoone please advise how to stop this. Dudley Miles (talk) 14:53, 3 February 2026 (UTC)

@Dudley Miles When you see the popup, if you advance through it by clicking Next, then click the 'Got It' button at the end, it will not be displayed again. Samwalton9 (WMF) (talk) 15:01, 3 February 2026 (UTC)

Thanks for your help Samwalton9 (WMF). Dudley Miles (talk) 09:20, 4 February 2026 (UTC)

Hybrid Search: Phase 1 - Early A/B Experiment on the Android App

Hi everyone,

Following the Phase 0 exploration we shared earlier, we’re moving into Phase 1 of the Hybrid Search project. Now for Phase 1, we are launching a small, time-limited experiment to learn whether hybrid search approaches can help readers find information on Wikipedia more effectively, learning from real Wikipedia traffic.

Phase 0 findings showed that some readers struggle to find specific information using internal search, especially when queries are phrased as questions or natural language. Based on that we decided that a hybrid approach, e.g. a search with both keyword and semantic capabilities, would be worth exploring in an experiment.

The video shows an early stage design for a hybrid search experiment.

This video shows an early stage design for the experiment, with an explainer note to users, letting them know that with this beta feature they can search using natural language. The example here is "cats and dogs comparison," which, when searched, retrieves the standard keyword match results, as well as beta-marked semantic results that surface relevant snippets from articles.

Who is this experiment for?

The Phase 1 experiment will be shown to a subset of mostly logged-out readers on the Android app on English, French and Portuguese Wikipedias. We will update the project page with the precise percentage of users included in the experiment. Editors will not be included in the experiment group at this stage, but can provide feedback through user testing interviews if interested.

What are we testing?

We’re testing a hybrid search approach that combines:

• Keyword-based (lexical) search, which works well for exact titles and names, and
• Meaning-based (semantic) retrieval, which can better support question-style or exploratory searches.

All results shown come from existing, human-authored Wikipedia articles and sections. This work does not involve generating new answers, rewriting content, or summarizing articles.

How will the experiment work?

Phase 1 will run as a controlled A/B test:

• A control group will see the current search experience.
• An experiment group will see semantic results surfaced alongside existing keyword-based results.

Readers will not be asked to choose between search modes. The experience will be clearly labeled as experimental, and readers will have the ability to opt out and continue using the current search experience unchanged.

Search results may include short excerpts, such as sentences or short paragraphs anchored to article sections, with clear indicators showing where information comes from.

How will we evaluate outcomes?

We’ll evaluate a combination of:

• Reader behavior (such as engagement and time-to-click)
• Qualitative feedback
• Performance guardrails, including latency and overall retention

If results show that this negatively affects the reading experience, the experiment will be adjusted or stopped. If results are promising, the findings will inform whether and how this work should continue.

Timing and next steps

Phase 1 is focused on design exploration, technical feasibility, reader insights and early community feedback. The experiment is set to begin in the second half of February and will be live for two weeks. Initial learnings from this work will be shared for discussion in March.

Learn more and stay involved

More details about Phase 1 scope, goals, and evaluation criteria are available on the project page. We’re especially interested in hearing from you:

• What are your overall thoughts on this early design?
• Are there risks or concerns you think we should be paying closer attention to?

We welcome continued feedback and discussion as this work progresses.

Thank you. EBlackorby-WMF (talk) 15:44, 2 February 2026 (UTC)

Looks great, this could be very useful. I wonder if this could also show results about help & meta pages and forward text snippets from these pages. Sometimes I'm wondering about things already answered perfectly well in a 10 year old VillagePump thread or want to find out sth like 'how to easily add a table column' or 'how to make video start at a specific time' and it's not easy to find the info and even more so for newcomers and as by default help&meta pages are not included in the search results. m:Community Wishlist/W442 includes or could inspire further ideas. I like the design but don't like the window that displays before one can search (maybe that's just temporary early on). Prototyperspective (talk) 17:35, 2 February 2026 (UTC)

Perhaps the first time someone searches on an account (TA or named), we pop up a box that says "Would you like to include help pages within search?" with the options "Yes", "Only while editing", "No". Or maybe we train a model on whether queries are meta or informational.   MetalBreaksAndBends   (talk) (contribs) 16:09, 6 February 2026 (UTC)

@EBlackorby-WMF Why is the video using the example query "cats and dogs comparison" that cannot be answered by pulling snippets from random Wikipedia articles, since there is no Wikipedia article comparing cats and dogs? It may be wise to use a better query when presenting an idea for a feature.

Why is the WMF trying to compete with large companies like Google and Meta with a tiny fraction of their resources? They train their LLMs on gigantic amounts of data, including but not limited to Wikipedia, so they will always have a superior product. Which means that I can make a better tool than what you are proposing, simply by using a bit of JavaScript that makes an API call to an external provider.

How will you add value to or improve upon what those multibillion dollar companies do?

Will the improved privacy, control, no dependency on commercial providers, cost at scale or whatever be worth it when you spend a lot of money making something inferior to tools everyone already uses? Polygnotus (talk) 18:24, 2 February 2026 (UTC)

I wonder why you assume this is about competition with "large companies like Google and Meta" and secondly why you seem to support these companies instead of improvements to Wikimedia. Lots of people also search things specifically within Wikimedia or use the Wikipedia search engine so that other content can be found on Google is no reason not to improve the Wikimedia search. This isn't a Web search engine. And there are article sections comparing cats and dogs and it was just an example anyway so if that example is bad that doesn't imply the concept is bad. Asking bonobos vs chimpanzees for example could surface section Bonobo#Cognitive comparisons to chimpanzees for example. Prototyperspective (talk) 18:31, 2 February 2026 (UTC)

@Prototyperspective why you seem to support these companies I dislike em, but they obviously have more resources than the WMF. why you assume this is about competition with "large companies like Google and Meta" because there is a limited amount of people asking questions? This is a zero sum game, there is a limited amount of people asking questions at any given time and if you ask a question to provider X you do not ask that same question to provider Y. And because those companies don't have this self-imposed limitation on not generating new text their answers would obviously be far superior.

why you seem to support these companies instead of improvements to Wikimedia I obviously support improvements to Wikimedia, I made quite a few, but this is just not a great idea. Or at least, in its current form as it is presented in the comment above.

Lots of people also search things specifically within Wikimedia or use the Wikipedia search engine so that other content can be found on Google is no reason not to improve the Wikimedia search. The amount of people who use the Wikipedia search for answers to questions is actually very low, especially when compared to Googles search box. You can just type your question in the address bar of your browser, actually navigating to en.wikipedia.org and then typing a question in a tiny search field is unnecessary and will lead to inferior results.

That doesn't mean I am anti-Wikipedia-Search improvements; I actually want the WMF to improve the fuzzy search stuff.

This isn't a Web search engine. See Knowledge Engine (search engine). Enjoy!

And there are article sections comparing cats and dogs Nope.

it was just an example anyway so if that example is bad that doesn't imply the concept is bad Agreed. But, as I explained, both the example and the concept are bad.

Asking bonobos vs chimpanzees for example could surface section Bonobo#Cognitive comparisons to chimpanzees for example Yeah there are many better queries they could've picked so its weird they used this one to present the idea to the community. If the work does not involve generating new answers, rewriting content, or summarizing articles then what is the point? If you bother extracting data you might as well present it in a useful format, otherwise the user experience is bad. Also if you want people to actually use the wikipedia search boxes, and type entire sentences into them then it may be wise to make them far more prominent and bigger. And sorry for stating the obvious but these 'success' indicators are so vague that it is impossible that this would not be considered a "success". Polygnotus (talk) 18:41, 2 February 2026 (UTC)

but they obviously have more resources than the WMF and… self-imposed limitation on not generating new text their answers would obviously be far superior

no, they wouldn't; at least often not or depending per what the user wants in the particular case if you ask a question to provider X you do not ask that same question to provider Y. flawed reasoning – this isn't so much about where Internet users are searching but primarily about users that are searching Wikipedia. However, unlike you apparently I wouldn't have an issue with more users coming to Wikipedia to search for a larger fraction of their information searches. I made quite a few I know and big thanks for them but as a personal subjective view you seem to be quite upset whenever WMF develops sth innovative and then complain how this should be left to companies or volunteers; maybe this makes sense in some cases but it doesn't feel like you truly reflect (and/or think more deeply) on whether this applies here or there anymore and what the topic beforehand is. I don't see why it wouldn't be a great idea, you certainly didn't explain why. The amount of people who use the Wikipedia search for answers to questions is actually very low …and? Is that a reason to not improve it? And still quite a few are using it. lead to inferior results only for most cases currently so that's not a good reason to not improve it, quite the opposite. See Knowledge Engine (search engine). Enjoy! Wikimedia search isn't the Knowledge Engine and still isn't a Web search engine. Nope What you're criticizing is not the actual subject, but the example selection. Hopefully better example selection next time, fine. But you're still wrong, there are many comparisons between the two on Wikipedia such as in Cat–dog relationship or if somebody improved that promising article, Comparison of sensory perception in species as well as List of animals by number of neurons and as well as the ability to produce amylase, an enzyme that functions to break down carbohydrates into simple sugars – something that obligate carnivores like cats lack in Dog food. Disproven I'd say. Another approach would be to show articles about comparable things of the two, e.g. Domestication of the cat and Domestication of the dog alongside or right after another and then Dog intelligence & Cat intelligence. the concept are bad It's useful to find info that way for lots of people for lots of cases plus lots of people are increasingly used to this type of searching so the search is good to adapt to also show them good results. so its weird they used this one to present the idea it's not always so easy to find or readily think of a good example but this is one where the general idea is easy to understand, whether or not the results in this case are good was I guess secondary to that. Btw, such searches may also make it clearer to editors that certain info is missing in Wikipedia and editors can also use this to check whether some info already is somewhere in Wikipedia or whether there's value in them looking into maybe adding it. and type entire sentences into them that's what users already do and secondly that's what's a type of searching that's easier to many for lots of cases. you might as well present it in a useful format, generated text is prone to hallucinations and inaccuracies, the text snippets approach is great and text responses is something that could be considered in addition (see the wish I linked) but that doesn't make this any less useful. then it may be wise to make them far more prominent and bigger could be considered later but it's not needed to make it bigger and lots of search input boxes also aren't large; it's large enough. Prototyperspective (talk) 19:39, 2 February 2026 (UTC)

no, they wouldn't; at least often not or depending per what the user wants in the particular case Not sure what you mean. Of course a generated text that directly answers a question is superior to two or more snippets of Wikipedia articles.

flawed reasoning – this isn't so much about where Internet users are searching but primarily about users that are searching Wikipedia. LLMs already have a large amount of users and already can be used to extract information from Wikipedia, which is in the datasets they are trained on. You can't convince people to start using Wikipedia search when it offers a far worse experience. Building a proper search engine or LLM tool is hard, and the WMF already failed at making a search engine. There are next to no people currently using wikipedia search in the way described in the OP. Most people just type in the article name and bash Enter.

as a personal subjective view you seem to be quite upset whenever WMF develops sth innovative and then complain how this should be left to companies or volunteers No, I desperately want them to develop something innovative. But I know that this is one of the pitfalls for the WMF. Making a semantic search thing or an LLM tool is fun. But it is not what we need. And the WMF has a tendency to work on things that are fun, and then bolt a shiny new interface on top of a dinosaur codebase, instead of working on the boring incremental improvements we need.

I don't see why it wouldn't be a great idea, you certainly didn't explain why. I wasn't trying to convince you specifically.

…and? Is that a reason to not improve it? Like I said, I want them to improve it. But stuff like this is very difficult. And they can easily waste a lot of very scarce dev time on a fun project like this while ignoring the important stuff. And there is a pattern of halfbaked things that don't actually help us reach our goal.

only for most cases currently so that's not a good reason to not improve it, quite the opposite No, the WMF will never be able to compete meaningfully with Alphabet or Microsoft, and making something inferior that no one will use because there are many better alternatives out there is not a great use of resources. If we had unlimited devs with unlimited time then it would be a cool project to have some fun tho.

Wikimedia search isn't the Knowledge Engine But making something somewhat related is obviously a bit iffy.

Hopefully better example selection next time Agreed!

there are many comparisons between the two on Wikipedia such as in Cat–dog relationship... You probably already noticed those are not in the video, sadly. Maybe you can help EBlackorby-WMF make a better video? With a better search query and actually relevant results from Wikipedia articles?

Another approach would be to show articles about comparable things of the two, e.g. Domestication of the cat and Domestication of the dog alongside or right after another and then Dog intelligence & Cat intelligence. Yeah and that proves my point. If I ask to compare dogs and cats, and Wikipedia Search shows me those 2 articles next to eachother, then that is a truly terrible answer. ChatGPT and Gemini and Claude and whatever else will provide a far superior answer than "here are 2 somewhat related wikipedia articles, enjoy!".

It's useful to find info that way for lots of people for lots of cases plus lots of people are increasingly used to this type of searching No, that is useful to no one on planet Earth. And people are getting increasingly used to chatbots writing an custom answer for them, which is far superior than just putting 2 Wikipedia articles next to eachother. And you can ask follow-up questions to chatbots....

it's not always so easy to find or readily think of a good example So offer to help them. Make something better.

such searches may also make it clearer to editors that certain info is missing in Wikipedia and editors can also use this to check whether some info already is somewhere in Wikipedia or whether there's value in them looking into maybe adding it. No, that is not how that works.

that's what users already do Well the younger generations mostly just shout at their phones in my limited experience.

secondly that's what's a type of searching that's easier to many for lots of cases Not sure what you mean.

generated text is prone to hallucinations and inaccuracies, the text snippets approach is great Sure, but so are humans. The text snippets approach is obviously terrible because people will stop using it after discovering it doesn't answer their questions.

text responses is something that could be considered in addition (see the wish I linked) but that doesn't make this any less useful. Yes it does, the existence of superior LLM chatbots does make this Wikipedia Search thing less useful. Things don't exist in a vacuum. Having a truck made the horse and cart less useful.

type entire sentences into them that's what users already do No they don't. They type the article name and press Enter. Most natural language questions in English are longer than the ~40 chars you got on 1080p, and the situation is worse on a phone. In the video you can fit ~28 chars in that inputbox.

it's not needed to make it bigger and lots of search input boxes also aren't large; it's large enough Its certainly not large enough for natural language questions in English. And if you want people to use something you need to put it front and center, hence Googles famous design that mostly consists of a textbox and 2 buttons. Polygnotus (talk) 20:11, 2 February 2026 (UTC)

Of course a generated text that directly answers a question is superior to two or more snippets of Wikipedia articles. no, it's not – not for everyone and not for every case. Imo the ideal result would be a preference and a toggle button for which to display where the default is both and I'd mostly use the snippets instead of potentially misleading answers, e.g. because I want to use this to check what exactly is in Wikipedia so that I can add or not add some missing info but there's also sorts of search-uses and users. You can't convince people to start using Wikipedia search strawman since that wasn't even set as the goal. a far worse experience. it's complementary and sometimes you may prefer that and sometimes the other. There are next to no people currently using wikipedia search in the way described in the OP. 1. false 2. not unexpected when that doesn't work.

- - break: please understand one thing and this one thing is important – not everyone uses Wikipedia like you. And also not necessarily like you imagine. There's use-cases of the search and types of searches and purposes you can't even conceive of.

very scarce dev time on a fun project like this I've heard you say this many times. But really I suggest to not see everything as a problem of that type. This isn't a fun project. It's an important over due project on something of impact and I wouldn't be surprised if it can be developed with relatively little volunteer time. Why do I not see you calling for more tech development – this being neglected is the real problem, not them developing sth you don't right now understand the value of (but it's clearly there). No, the WMF will never be able to compete meaningfully with Alphabet or Microsoft people talking like that are usually employed by such companies so could you please just not continue with this talking point… It's not about competing with them. And a little more users coming to Wikimedia instead to search Wikimedia-related things would be great but that's not the goal (/the entire goal) here as I've already explained. making something inferior that no one will use you're comparing apples and oranges as I've already explained. Largely different purposes. Are you next saying we don't need hosting because Amazon could do it for us? What is next, come on. Please understand that saying something is inferior or not requires the use-cases to be the same but they aren't. And even if they were that doesn't imply we shouldn't improve the search. Other search engines being better is just an additional reason to upgrade the outdated tech. and Wikipedia Search shows me those 2 articles next to eachother, then that is a truly terrible answer for. you. Not for me and many others. I'd like to have exactly that in specific with text segments from within the article. ChatGPT and Gemini and Claude and whatever else will provide a far superior answer than "here are 2 somewhat related wikipedia articles, enjoy!". I listed more than 2 articles. I tried AI models, asking them to show me Wikipedia article sections about it and compare and they failed or weren't better. Seems like you seem to want people to use just LLM answers but not Wikipedia text; really I don't get why somebody would want that. No, that is useful to no one on planet Earth. false; it would be useful to me. Don't make assessments so easily. which is far superior than just putting 2 Wikipedia articles next to eachother. It's not 2. It's not just the articles but also text segments (and there's other use-cases than the example). And it's not always better. And fourth, the user who has searched on Wikipedia has searched on Wikipedia, not Google so your premise even if you don't recognize it as such is flawed. So offer to help them. Make something better. why are you so hostile whenever WMF develops sth when I give some feedback and am positive about the development? And OP's post doesn't come down to just one example. I also don't find it so nice to discuss here at such length. No, that is not how that works. good point /s Well the younger generations mostly just shout at their phones in my limited experience. good point /s Not sure what you mean. that to many people searching in natural language questions is easier – in specific not always but at least sometimes. because people will stop using it it's just additional results. No value is lost if people try this search and find they don't find the info they look for. does make this Wikipedia Search thing less useful. well obviously less useful but not unimportant or not having large impact; and I explained why your so-great LLM replies do have severe flaws. They type the article name and press Enter 1. not all 2. probably more and more try natural questions 3. obviously relatively do longer queries or even ask things because that currently doesn't work well/at all In the video you can fit ~28 chars in that inputbox and? It's same for the Web search engine and I still write long queries on the phone. Apparently you assume that there can never be any typo in it and/or the user must see the whole query they write at all time to be able to write it. Its certainly not large enough for natural language questions in English maybe watch the video again; it's more than large enough in the app and if it's too small the size can be increased. I don't want to hear about Google this or that anymore; we're not Google; next maybe you'll be asking we should stop spending scarce volunteer time writing Wikipedia because people can just ask superior AI models their questions. No. Prototyperspective (talk) 23:18, 2 February 2026 (UTC)

If you ask a question and instead of an answer you get 2 vaguely relevant snippets of Wikipedia pages you'd figure something was wrong with that person and back away slowly. Same thing with software.

I'd mostly use the snippets instead of potentially misleading answers No, you wouldn't, and both options are potentially misleading. Have you ever factchecked a Wikipedia article? Pick any Wikipedia article with a lot of edits and factcheck some of the most interesting claims.

I want to use this to check what exactly is in Wikipedia so that I can add or not add some missing info I don't think the tool described in the OP is intended for that usecase (and it wouldn't be great for it).

strawman since that wasn't even set as the goal. That is not what a strawman is, and that was set as the goal. I linked to it and so did the OP.

it's complementary and sometimes you may prefer that and sometimes the other. Nah, people will just continue using whatever tool they are already using. You don't see Google users using Bing once a week.

1. false 2. not unexpected when that doesn't work. 1) lol no its not. 2) sure, but it shows demand is low.

not everyone uses Wikipedia like you Thank god. also not necessarily like you imagine. I am very familiar with how people use computers.

There's use-cases of the search and types of searches and purposes you can't even conceive of. Well, you suggested to check what exactly is in Wikipedia so that I can add or not add some missing info but this tool wouldn't really work for that, sadly.

This isn't a fun project. It is for nerds.

I wouldn't be surprised if it can be developed with relatively little volunteer time. Probably zero, hence the "scarce dev time" thing.

Why do I not see you calling for more tech development I repeatedly have.... and do.

tech development – this being neglected is the real problem And yet these people go to work every day. And they do stuff. So do we want devs to work on this or on something important? We can't clone 'em.

you don't right now understand the value of (but it's clearly there) & people talking like that are usually employed by such companies so could you please just not continue with this talking point… People can just have a different opinion, without them "not understanding" or working for evil companies.

It's not about competing with them Yes it is. They are proposing making something that answers questions by searching Wikipedia and providing relevant snippets. Any AI chatbot can do the same except a billion times better (they are not limited to wikipedia, and can generate an answer to better match the question).

And a little more users coming to Wikimedia instead to search Wikimedia-related things would be great but that's not the goal I think you need to click the links in the OP above. I already linked to https://www.mediawiki.org/wiki/Readers/Information_Retrieval/Phase_1#Success_indicators So you keep making incorrect assertions.

you're comparing apples and oranges as I've already explained. Largely different purposes. Both share the purpose of answering questions.

Are you next saying we don't need hosting because Amazon could do it for us? Doing your own hosting saves a hell of a lot of money compared to just paying Amazon.

Please understand that saying something is inferior or not requires the use-cases to be the same but they aren't. And even if they were that doesn't imply we shouldn't improve the search. Other search engines being better is just an additional reason to upgrade the outdated tech. What is the point if no one is gonna use it because the alternatives are so much better?

and Wikipedia Search shows me those 2 articles next to eachother, then that is a truly terrible answer for. you. Not for me and many others. I'd like to have exactly that in specific with text segments from within the article. If you want that I am sure you can write a script to make that. But it would be terrible at actually answering questions. You can look at User:Polygnotus/CitationVerification for inspiration.

Seems like you seem to want people to use just LLM answers but not Wikipedia text; really I don't get why somebody would want that. Those LLMs are trained on Wikipedia. And those companies generate text. So their service is gonna be infinitely better at providing answers to natural language questions than the proposal above. Sometimes it is not about what people want, there are just cold hard immutable facts we have to deal with.

It's not 2. It's not just the articles but also text segments (and there's other use-cases than the example). And it's not always better. And fourth, the user who has searched on Wikipedia has searched on Wikipedia, not Google so your premise even if you don't recognize it as such is flawed. Wut?

OP's post doesn't come down to just one example. Well using a terrible example does undermine any proposal. For example the simple summaries thing had a terrible screenshot.

No value is lost if people try this search and find they don't find the info they look for. I value the wasted dev time.

I probably missed some bits because much of what you wrote was difficult to understand. Polygnotus (talk) 00:15, 3 February 2026 (UTC)

Well you insist you know how everyone is and will be searching Wikipedia. If you ask a question and instead of an answer you get 2 vaguely relevant snippets of Wikipedia pages This is an issue with the example; at other times it can be exactly what one was looking for. You don't see Google users using Bing once a week. because they're both for the same purpose. to work on this or on something important This is sth important. We can't clone 'em. I'll make an extraordinary claim: more devs could be hired. Additionally, lots of other things. Wut? people do (already) search (within) Wikipedia, the main question is what / how good the results are/can be. What is the point if no one is gonna use it because the alternatives are so much better? you don't understand it. Neither what I just said nor that not everyone always wants to ask an AI but maybe wants to just have human-written Wikipedia snippets and/or search Wikipedia. So their service is gonna be infinitely better at providing answers to natural language questions well I stop here. Prototyperspective (talk) 00:44, 3 February 2026 (UTC)

Well you insist you know how everyone is and will be searching Wikipedia. I did? Can you provide a link please?

This is an issue with the example; at other times it can be exactly what one was looking for. Theoretically. Or it could be the same or worse.

because they're both for the same purpose. Yes, that purpose is answering questions.

I'll make an extraordinary claim: more devs could be hired. I see your claim and raise you: more devs should be hired.

The WMF got a new CEO. Maybe you can ask her to hire more devs?

maybe wants to just have human-written Wikipedia snippets and/or search Wikipedia I understand that people might wanna search Wikipedia, and building a semantic model could be interesting. But very few search queries on Wikipedia are actually in something approaching natural language.[4]

Because of the limited dev time I predict the idea is just to play around with something like paraphrase-multilingual-MiniLM-L12-v2 for a bit. And while that sounds cool, I can foresee a bunch of problems:

When presenting snippets, how can we ensure the relevant bits of context are preserved while everything else is filtered out?

A fact I asked about may be in a long paragraph of text I can barely understand.

Or my understanding may be based upon context that no longer exists when only that paragraph is presented.

And what if its not a paragraph but a cell in a long table, how to present that without context?

If you don't generate your own text these are almost unfixable problems.

Once users discover LLMs give them actual custom answers while Wikipedia search gives them decontextualized snippets, they'll just use their preferred LLM. And you can ask any LLM a follow-up question.

"Don't build anything ever because competitors exist" is not the same as "don't build an inferior version of something that already exists made by a competitor with a thousand times more resources." Polygnotus (talk) 01:10, 3 February 2026 (UTC)

The WMF doesn't have (enough) people who point out flaws in their plans. Someone explaining why a plan sucks is, for a company, not an enemy but the most valuable person you have.

Before working on a plan you need to consider the pitfalls and downsides. So if this is a great plan they already have an answer to counter all criticism, built into the proposal. Polygnotus (talk) 20:33, 2 February 2026 (UTC)

Well I did a bit of investigation and this is a terrible plan as proposed and presented. But when you ignore the proposal and just think "let's play around with something like paraphrase-multilingual-MiniLM-L12-v2 and see if can be used to make Wikipedia Search less terrible" that is a good idea. Polygnotus (talk) 23:53, 4 February 2026 (UTC)

@EBlackorby-WMF, the most important thing about Special:Search for me is that it doesn't (always) try to guess what I'm thinking. Especially for searches outside the mainspace, I'm fairly often looking for an exact string (e.g., a misspelling, a username, a date, a code snippet) that I know will take me to the page I remember. Getting a hundred semantically related results is the opposite of helpful in such cases. WhatamIdoing (talk) 23:11, 5 February 2026 (UTC)

See semantic results surfaced alongside existing keyword-based results. I guess there will also be a setting to disable seeing any such results. Prototyperspective (talk) 23:16, 5 February 2026 (UTC)

Maybe an argument to search? Like how you can prefix searches with a namespace to get results from that namespace. Perhaps "nosem:wp: lta6234".   MetalBreaksAndBends   (talk) (contribs) 16:12, 6 February 2026 (UTC)

@EBlackorby-WMF Will this fix the issue I see, where searching fir "Sense and Sensibility", or "Pride and Prejudice" (without the quotes) does not bring up what's expected?

When I search for Sense and Sensibility, the suggested matches all startcwith "Sensibility"! David10244 (talk) 05:23, 7 February 2026 (UTC)

I don't get that result. Do you have a link? WhatamIdoing (talk) 05:26, 7 February 2026 (UTC)

@WhatamIdoing When I click the magnifying glass next to my name, then type sense and sensibility (actually, all lower case), the first suggested result is the article Sensibility. The next suggestion starts with Sensibility. I'm not at my desktop, so I can't give much more info. I can't make and attach a screenshot easily from my tablet (I don't know how). 🙂 David10244 (talk) 05:35, 7 February 2026 (UTC)

Screenshot of search results

Here's a WP:WPSHOT showing what happens when I type sense and sensibility (all lower case, as you can see) into the search bar at the top of the page.

Is that what you're talking about? Do you get different results when you do the same thing? WhatamIdoing (talk) 07:18, 7 February 2026 (UTC)

@WhatamIdoing Yes, I get different results. My list starts with "Sensibility" (pointing to the article by that name) and the second item is for the book named "Sensibility Objectified". It's very strange that we get different results. I don't have any local scripts in common.css or anything like that. David10244 (talk) 04:00, 11 February 2026 (UTC)

Page not being automatically archived, part 2

User:LaundryPizza03/CSD log is still not archiving. This is a followup to Wikipedia:Village_pump_(technical)/Archive_226#Page_not_being_automatically_archived. –LaundryPizza03 (dc̄) 22:50, 2 February 2026 (UTC)

Pinging participants from last discussion due to lack of participation: @Redrose64, PrimeHunter, Aidan9382, Graham87, and Jonesey95. –LaundryPizza03 (dc̄) 20:40, 8 February 2026 (UTC)

For what it's worth I checked and couldn't find anything wrong with the archiving setup this time. Graham87 (talk) 02:51, 9 February 2026 (UTC)

Deprecating and blacklisting archive.today

As some may be aware by now, the maintainers of archive.today (and archive.is, etc.) recently injected malicious code into all archived pages in order to perform a denial of service attack against a person they disliked (this can be confirmed by the instructions described here.) While the malware has been removed now, it is clear that archive.today can't be trusted to not do this in the future, and for the safety of our readers, these archiving services should be swiftly removed and the websites blacklisted to prevent further use.

The rub is that https://archive.today alone is used in nearly 400,000 pages, and its sister site archive.is is used in around 50,000. We're clearly very dependent on this service, but we must find a way to break away from it. And fast. Children Will Listen (🐄 talk, 🫘 contribs) 00:26, 5 February 2026 (UTC)

I've been wondering the same thing. I don't have an answer but have notified Wikipedia talk:Link rot to get more attention to this issue. ClaudineChionh (she/her · talk · email · global) 01:05, 5 February 2026 (UTC)

Absolutely not. We are dependent on external sites for archiving and verification. I have in the past lobbied WMF to acquire archive.org so it can meet our needs (or set up our own version) but unless and until that happens we have to link to external sites for verification. Hawkeye7 (discuss) 01:24, 5 February 2026 (UTC)

We're not deprecating archive.org (run by the Internet Archive), only the archive services provided by whoever runs archive.today. The maintainers of that service cannot be trusted for the reasons I described. Children Will Listen (🐄 talk, 🫘 contribs) 01:27, 5 February 2026 (UTC)

I don't trust the maintainers of archive.org. Hawkeye7 (discuss) 02:28, 5 February 2026 (UTC)

I trust them a far sight more than someone who has now verifiably used their ownership of a domain we link some 400k times to DDOS another website on the Internet. Izno (talk) 03:39, 5 February 2026 (UTC)

Maybe, but we need verifiability and archival sites are required for that. We shouldn't sacrifice our primary mission. Hawkeye7 (discuss) 03:57, 5 February 2026 (UTC)

Archive sites that are willing to inject malicious javascript aren't particularly good for verifiability. AntiCompositeNumber (they/them) (talk) 04:31, 5 February 2026 (UTC)

They are better than nothing. The only solution is to run our own archive. Hawkeye7 (discuss) 04:46, 5 February 2026 (UTC)

Why insist on framing this as an all or nothing choice? Other archive sites exist and they don't have a demonstrable history of weaponising their service. Treating it as an exceptional case isn't unreasonable.

Also, even if this were an all or nothing choice (which it isn't), Wikipedia's need for citations isn't more important than the security of users. Archive.today has demonstrably abused its users' trust (including Wikipedia's editors and readers) and cannot be considered safe. – Scyrme (talk) 20:46, 5 February 2026 (UTC)

• Trash it completely. Archive.today has proven that it's not trustworthy as an archive source (unlike the Internet Archive) and links to it should be considered potentially malicious in nature. Silverseren^C 04:58, 5 February 2026 (UTC)

  "Archive.today has proven that it's not trustworthy" - there are no (known) examples of its owner tampering with archived pages. "unlike the Internet Archive" - Internet Archive removes archived copies regularly. sapphaline (talk) 14:56, 5 February 2026 (UTC)

  there are no (known) examples of its owner tampering with archived pages: Yes there is, see above. Injecting malicious JavaScript is tampering, visible or otherwise. If they are willing to do that, who knows when they'll decide to exploit zero-days or engage in blatant manipulation. Children Will Listen (🐄 talk, 🫘 contribs) 15:03, 5 February 2026 (UTC)

  They only injected it on the CAPTCHA page. sapphaline (talk) 15:04, 5 February 2026 (UTC)

  Which is only shown when you try to archive something, by the way. sapphaline (talk) 15:05, 5 February 2026 (UTC)

  I always see the CAPTCHA screen when I view an archived page for the first time. Children Will Listen (🐄 talk, 🫘 contribs) 15:06, 5 February 2026 (UTC)

  My point about them being trustworthy when it comes to archived copies stands. Internet Archive is way less reliable in this regard, because archived copies can always be deleted there. sapphaline (talk) 15:20, 5 February 2026 (UTC)

  I'd rather have information lost than readers having to encounter mailicious code whenever an archived copy is visited. Also, we know nothing about the maintainer(s) of Archive.today, how they make money, or even if they're ready to pack up their bags tomorrow and leave. They're in a jurisdiction that's politically unstable and prone to censorship. None of these problems exist with the Internet Archive. Children Will Listen (🐄 talk, 🫘 contribs) 15:30, 5 February 2026 (UTC)

  A jurisdiction that's politically unstable and prone to censorship - you mean, like the United States? (I wish I was joking about my country in 2026) Setting that aside, we shouldn't want any information lost just like that. We need a remedying/replacement process coming before a removal process. See my main comment below. Stefen 𝕋ower ^{Huddle • Handiwerk} 15:34, 5 February 2026 (UTC)

  If this RFC is going to pass (which will be a very unfortunate result!), megalodon.jp archives archive.today snapshots almost perfectly (the only issue is that they're zoomed out and for some reason have a 4000px width, but this is trivially fixed by unchecking some checkboxes in devtools). Maybe WMF could arrange some deal with their operators to archive all archive.today links we have? sapphaline (talk) 15:43, 5 February 2026 (UTC)

  "how they make money" - why should we care about this? "if they're ready to pack up their bags tomorrow and leave" - archive.today has existed for nearly 14 years. It's a snowball chance in hell they're going to shut the site down tomorrow or in any foreseeable future. "They're in a jurisdiction that's politically unstable and prone to censorship" - how do you know? "None of these problems exist with the Internet Archive" - US is extremely prone to censorship and political unstability + Internet Archive removes archived copies on any requests, not just governmental ones. sapphaline (talk) 15:35, 5 February 2026 (UTC)

  It is economically infeasible to hold trillions of archived pages and provide them indefinitely for free. We don't know how they're funding their project, which means we wouldn't know when this funding would dry out.

  Their willingness to inject malware over a petty dispute puts their stability in disrepute. If we get in the bad graces of these maintainers, who knows what they'll be willing to do to us?

  It's fairly well-known that the maintainer(s) of Archive.today live in Russia, and that the main archive storage is also hosted in Russia. Sometimes, they redirect certain IP addresses to yandex.ru, and of course, their official Wikimedia account Rotlink was created in ruwiki.

  However, you are right about the United States, sadly. Children Will Listen (🐄 talk, 🫘 contribs) 15:44, 5 February 2026 (UTC)

Actually the theory is Ukraine, not Russia, and the evidence is they provision on global edge cloud providers (such as CloudFlare - but not CloudFlare). -- GreenC 15:46, 7 February 2026 (UTC)

@ChildrenWillListen makes an important point. People are bringing up that WP already has 500K links to them. What if they introduce malicious code on just some of the archived pages (for instance, because targets of their malice to be more likely to access those links)? Aurodea108 (talk) 01:49, 9 February 2026 (UTC)

I can't think of an explanation for this that isn't malicious. You'd think the maintainer(s) of archive services wouldn't be stupid enough to try to get a blog removed from the internet as a petty retaliation over some alleged doxxing. — DVRTed (Talk) 05:33, 5 February 2026 (UTC)

I agree, they should be blacklisted. Should have happened a long time ago, really, because of massive copyright violation: they distribute lots of content that the copyright owners only made available behind paywalls. See WP:COPYLINK: "if you know or reasonably suspect that an external Web site is carrying a work in violation of copyright, do not link to that copy of the work". — Chrisahn (talk) 11:11, 5 February 2026 (UTC)

I fully appreciate why this needs dealing with, but I am concerned about "the rub". We could end up harming verifiability on a *lot* of our content. Of course, we can leave citations in place without the archive.today links, but without the ready verification of having an article to load, I fear some useful article text could end up being removed by editors who decide they can't trust the listed source due to inaccessibility (typically those with little wiki experience). In cases where the paywalled content still exists, removal would be less likely, but in cases where the original link is permanently dead, it's not available on Archive.org, and we only have archive.today... yikes.

Deprecation makes sense as long as that doesn't include immediate removal before any replacement remedy is pursued. Any process that intervenes with using archive.today should encourage editors to directly replace these sources with archive.org links or newspaper.com clip links, or locate alternate sources. I realize this is generally what deprecation means but if the intervention can be clear and help the editor find an alternative, I would be more relieved of the ramifications of ditching this source. Stefen 𝕋ower ^{Huddle • Handiwerk} 14:51, 5 February 2026 (UTC)

well-said. i support blacklisting as long as it is accompanied by an effort to find alternative solutions instead of just plain removal. ... sawyer * any/all * talk 15:33, 5 February 2026 (UTC)

Oppose - this will greatly harm verifiability. sapphaline (talk) 14:53, 5 February 2026 (UTC)

Agree with those above arguing that archive.today is simply not trustworthy enough to be sending our readers to. Adding malicious code to cause a DDoS on another website is an absurd thing for a website maintainer to do and we shouldn't be facilitating their behaviour by sending more users to their site, nor simply hoping that they won't do something worse which targets our readers. Sam Walton (talk) 15:03, 5 February 2026 (UTC)

You may be interested to see Wikipedia:Requests for comment/Archive.is RFC (2013, with consensus to remove the then 10k links and blacklist for future, overturned by Wikipedia:Requests for comment/Archive.is RFC 4 in 2016.) Andrew Gray (talk) 18:56, 5 February 2026 (UTC)

I saw that, yes. Perhaps it's time for Wikipedia:Requests for comment/Archive.is RFC 5? Children Will Listen (🐄 talk, 🫘 contribs) 19:01, 5 February 2026 (UTC)

Yes, I think so. As you've said above, they can't be trusted to not do that again in the future, so I would support blacklisting their links. Some1 (talk) 01:17, 6 February 2026 (UTC)

I think there needs to be an official RfC on this to get more opinions. Personally I think this shows that archive.today can't be trusted (if they do this over something rather petty, what's stopping them from putting more malicious code into archived pages, not just the captcha?), and it should be at least deprecated - but only if the links can be replaced with a different archive without loss of information. Suntooooth, it/he (talk | contribs) 20:15, 5 February 2026 (UTC)

They blatantly violated Wikipedia:External links#EL3, but you think we need to have a long discussion about whether malware-serving websites are sometimes okay?

If we're going to have an RFC, let's blacklist now and focus the RFC discussion on how to cope rather than whether we should provide links to malware-serving websites. WhatamIdoing (talk) 22:16, 5 February 2026 (UTC)

And if we can't quite bring ourselves to list the sites on MediaWiki:Spam-blacklist, then let's at least put up a warning via Special:AbuseFilter. WhatamIdoing (talk) 22:17, 5 February 2026 (UTC)

I think that formally gaining consensus is important when it affects as many links as this does, especially since even in this thread it hasn't been unanimous. If this affected a much lower number of links (think a couple of orders of magnitude lower) and links that would be easily replaced or removed, then I wouldn't be suggesting a full RfC. Suntooooth, it/he (talk | contribs) 00:40, 6 February 2026 (UTC)

I wonder if there's a way to add a warning in the articles. Something like (and a category showing affected articles) might encourage people to start the process of finding other sources. It might be possible to do this automagically through the CS1|2 templates. I'm assuming that would catch most of them. WhatamIdoing (talk) 02:44, 6 February 2026 (UTC)

It is in the realm of feasible just to turn off the display of archive links that are via archive.today/is in CS1/2. The real question is if we can get everything or if we just have to start off with vanishing the big quantity of links. Izno (talk) 03:59, 6 February 2026 (UTC)

Removing archive links (even by just turning them off, rather than fully removing them) from this number of articles would be a huge hit to verifiability. If consensus is gained to remove archive.today links, there needs to be a mechanism for replacing them with other archives. Suntooooth, it/he (talk | contribs) 16:05, 6 February 2026 (UTC)

would be a huge hit to verifiability I think turning their display off is a fair compromise on the road to removal and replacement. I do agree that half a million pages or links is a big number. A maintenance category would naturally be set up so we can actually find these quicker.

We probably also should notify WP:URLREQ and/or specifically @GreenC about this discussion. Izno (talk) 20:21, 6 February 2026 (UTC)

Good idea. Also considering the RfC above, isn't it possible that many of the archive.today links on the encyclopedia _aren't actually necessary?_ As in, they were added superfluously by the website operators themselves? Perhaps the true scale of the problem is much smaller, and we could vibe code a quick tool to check some of the links. audiodude (talk) 04:50, 8 February 2026 (UTC)

It's not feasible to blacklist it right now. sapphaline (talk) 07:20, 6 February 2026 (UTC)

Why do you think it's not feasible? What do you mean by "feasible"? — Chrisahn (talk) 15:18, 6 February 2026 (UTC)

Because we have ~500k+ (including all of the different domain names) links to it. sapphaline (talk) 18:25, 6 February 2026 (UTC)

That's not a problem. We won't remove them (at least not for a while). Blacklisting means that no new links to these domains can be added. It doesn't mean existing links have to be removed. — Chrisahn (talk) 22:29, 6 February 2026 (UTC)

Every day, links are added because there is no other option. They literally are the only source for a large set of web pages on the Internet. This is why there are so many links. It's the only option. It is pragmatic. You and some others appear concerned about what is best for Wikipedia, but you don't seem concerned about the consequences, which are very real, immediate and large scale - it would cause significant damage to Wikipedia. Unlike the good feelings about punishing Archive.today for some transgression. What is more important? -- GreenC 16:04, 7 February 2026 (UTC)

I've added archive.org URLs to lots of articles. In case a page hasn't been archived by them yet, I click "Save Page Now". I don't recall any significant problems, and I don't recall a URL that couldn't be archived. I'd say such URLs are pretty rare. — Chrisahn (talk) 22:50, 7 February 2026 (UTC)

Actually it's common. I speak from the data not an opinion. -- GreenC 23:27, 7 February 2026 (UTC)

The problems likely depend on the type of source, so some editors may encounter problems fairly often, and others will not ever encounter it. WhatamIdoing (talk) 23:35, 7 February 2026 (UTC)

What data? — Chrisahn (talk) 23:36, 7 February 2026 (UTC)

@Chrisahn, you might want to read the bottom of his User: page... WhatamIdoing (talk) 01:00, 8 February 2026 (UTC)

Agree that there should be an RfC. The implications of the discussion and potential actions taken by consensus will have far reaching effects across the encyclopedia. Additional comment as a technical editor, not one who edits a lot of articles: if archive.today provides a copy of a paywalled or linkrotted news article, but the article was actually published by the news organziation in question at some point, what does it matter if the archived copy isn't available? The citations are still technically valid right? Does wikipedia remove citations to books that are out of print? Does information exist if it's not on the internet lol? audiodude (talk) 04:47, 8 February 2026 (UTC)

Yes, you're right that a Wikipedia:Convenience link (to the original and/or an archive) is not required, if the news article is archived in some place that is accessible to the general public. For example, it's traditional for ordinary print newspapers to keep a copy of all their old newspapers, and many will either let the general public take a look or send the older ones to a local library or historical society. However, not all publications have a print edition, and some news outlets put more information/additional articles on their web. I have, for example, been disappointed that paper copy of The Atlantic has fewer articles than their website. A web-only source needs a working URL, because sources must be WP:Published#Accessible. WhatamIdoing (talk) 06:04, 8 February 2026 (UTC)

Has anyone linked to the circus that occurred when archive.today first appeared? As I recall, they used extremely advanced (for the time) techniques to attack Wikipedia by edit warring their links into pages. The views that we have to keep using them miss the big picture: these guys are obviously up to something bad. The infrastructure and operational maintentance to support their system would cost a vast amount and someone is planning to get a return on that investment eventually. It's much more effort than some libertarian philanthropist would support. Johnuniq (talk) 02:47, 6 February 2026 (UTC)

Yes linked above: Wikipedia:Requests for comment/Archive.is RFC (2013, with consensus to remove the then 10k links and blacklist for future, overturned by Wikipedia:Requests for comment/Archive.is RFC 4 audiodude (talk) 04:54, 8 February 2026 (UTC)

Technical question: How would blacklisting work? If I understand correctly, the idea is that blacklisting prohibits adding new archive.today (and archive.is etc.) links, but we'll keep the existing ones for now. Specifically: If I edit an article and try to add a new archive.today link, I get an error message and can't save my changes. But if I edit an article (or section) that already contains one or more archive.today links and I make unrelated changes, there's no such error message. Is that correct? Can we make that work? A "dumb" edit filter (that simply checks whether such links occur anywhere in the text I'm trying to save) won't work – it won't let me save unrelated changes. I can think of a few ways to implement a smarter filter, but I don't know if edit filters have access to the required information, or how efficient smarter checks would be. — Chrisahn (talk) 09:18, 6 February 2026 (UTC)

@Chrisahn Yes, this is how the built-in tools like MediaWiki:Spam-blacklist already work, and edit filters can also be made to work that way. They forbid adding new links to blacklisted domains, but if a link is already present in the article, it can be edited without tripping the blacklist. There are still some scenarios that cause problems (e.g. if a vandalism deletes a citation that links to archive.today, you won't be able to revert it without removing those links first), but that hasn't stopped additions to the blacklist before. Matma Rex talk 14:41, 6 February 2026 (UTC)

Thanks for the explanation! Sounds good. — Chrisahn (talk) 15:17, 6 February 2026 (UTC)

There have been concerns about archive.is/archive.today going back more than twelve years, see for example Wikipedia:Requests for comment/Archive.is RFC. --Redrose64 🌹 (talk) 11:13, 6 February 2026 (UTC)

Hi!

archive.today is just a very useful website that can be used if archive.org is not helping.

The hosters of archive.today are not as reliable as the guys that host archive.org. However, we don't know any case, where a snapshot was false, do we?

In this case, they "just" abused their visitors for a DDOS attack. Of course we should not support this. But this does not mean, we have to definitely block the website.

By the way, blacklisting (via WP:SBL) without a previous removal of all links is not a good option, because this leads to several problems:

• Moving of parts of pages to other pages (e.g. archiving) is not possible any longer, if the moved text contains a link.
• Modifying an existing blacklisted URL (e.g. link fixing) might trigger the SBL.
• It's not possible to add blacklisted links to a discussion which is challenging for some not so technical users.

In my opinion a technical solution could be:

• replace all links with a (unsubstituted) template (yes, this is a lot of work, but could be automized partially);
• if any problem with the domain occurs again, modify the template such that there is no link any longer to archive.today (and .is and all the other domains);
• when the problem is solved, revert the template change;
• if anybody adds a link to archive.today without the template, a bot could try fix that afterwards and the bot could write a message on the linker's talk page that they should check whether they could find something better.

By a solution like this we would would still have the benefits of the archived versions. But we could remove all links fast and at once, if needed.

-- seth (talk) 17:04, 6 February 2026 (UTC)

A couple hundred thousand bot edits is not a good solution, either.

Trappist the monk might have some ideas about whether the citation templates could special-case these domain names for a while, while the work is done. A maintenance category, as Izno mentioned above, would also be a good idea. And even if we don't want to use the MediaWiki:Spam-blacklist quite yet, for fear that it will interfere with rearranging pages, we could implement a Special:AbuseFilter that would prevent people from adding any new ones. WhatamIdoing (talk) 21:33, 6 February 2026 (UTC)

I still advocate going back to WMF with a proposal to create our own archive. This will get us off dependence on external archive sites that we cannot control. Hawkeye7 (discuss) 22:25, 6 February 2026 (UTC)

@Hawkeye7: It may open us up to legal difficulties we can't easily handle, especially if we refuse to delete content as instructed by copyright owners. Children Will Listen (🐄 talk, 🫘 contribs) 22:27, 6 February 2026 (UTC)

Setting up an internet archive requires years of planning and work. I'd like us to start making tangible progress on resolving this problem today, or at least in the next week. Even if we thought that was legal and a good idea, creating our own archive isn't going to address the problem right now. WhatamIdoing (talk) 23:24, 6 February 2026 (UTC)

cs1|2 can special case archive.today (and companion domains) if/when there is a consensus to deprecate/blacklist.

—Trappist the monk (talk) 22:58, 6 February 2026 (UTC)

One major problem with the edit filter (and SBL/BED) is that many unexperienced people who trigger a rule just don't know what that means and what they should do. We often see that people who wrote large paragraphs and failed of first try to safe, just run away, although the warning said that if they are sure about what they are doing, they just should try to save again.

The filter (and SBL/BED) should be used if people intentionally (try to) spam. If they actually just want to help, then there's a risk of annoying/frustrating them. That's why -- over time -- I more and more tend to use notification bots and maintenance lists instead of the blacklist-like tools in cases where links are mostly added by non-spammers.

-- seth (talk) 23:27, 6 February 2026 (UTC)

XLinkBot can deal with link additions, and we can modify the citation modules to not accept/show archive.today links anymore. After the number of links becomes manageable, an edit-filter based solution would be a good idea. Children Will Listen (🐄 talk, 🫘 contribs) 22:36, 6 February 2026 (UTC)

Also, am I the only person concerned by how OP downplays doxxing someone as a "petty dispute"? sapphaline (talk) 18:25, 6 February 2026 (UTC)

Might be concerning, but that's "two people have been bad people" and each should be judged on their own merit accordingly. You don't treat someone DDOSing another person off the Internet as a stable individual meriting half a million links from the most popular source of collated information on the Internet (and that ignoring the prior dramas, as linked above). Izno (talk) 21:37, 6 February 2026 (UTC)

@sapphaline: I wasn't fully aware of what the DDoS was a retaliation for when I started this thread, but either way, as @Izno said, they can't be trusted anymore, regardless of intention. Children Will Listen (🐄 talk, 🫘 contribs) 22:28, 6 February 2026 (UTC)

Doxing? Hardly. Quote: "While we may not have a face and a name, at this point we have a pretty good idea of how the site is run: it’s a one-person labor of love, operated by a Russian of considerable talent and access to Europe." [5] — Chrisahn (talk) 22:40, 6 February 2026 (UTC)

Probably not important, but interesting: Less than a day before this discussion was started, User:Masharabinovich, an account probably connected to archive.today, was renamed by request. — Chrisahn (talk) 22:47, 6 February 2026 (UTC)

Well, the article they tried to remove did say that "Masha Rabinovich" is a pseudonym they use when creating online accounts. Children Will Listen (🐄 talk, 🫘 contribs) 23:13, 6 February 2026 (UTC)

Another aspect: Depending on how the FBI case against archive.today goes, there's a chance that these ca. 500,000 archive links in our articles will become useless in the not too distant future. — Chrisahn (talk) 01:05, 7 February 2026 (UTC)

This I agree with. Stay the course and wait and see. -- GreenC 17:11, 7 February 2026 (UTC)

• Prior to about 2015, the Wayback Machine did not systematically archive all links on Wikipedia. There are huge gaps prior to that date. Between 2012(?) and 2015, Archive.today systematically archived Wikipedia. Thus many dead links are only archived on Archive.today. The one time Archive.today got blacklisted a long time ago, it didn't last long. People reversed it. Why? Because Archive.today is incredibly useful. It's that simple. It's pragmatic. They have the goods nobody else does. This incident with the CAPTCHA will soon be forgotten as inconsequential to Wikipedia. But blocking Archive.today will cause daily conflict with editors who need to use it because there is no other option. -- GreenC 17:11, 7 February 2026 (UTC)

  The decision was reversed because the maliciousness of the maintainers used to be pure speculation. This is no longer the case. Children Will Listen (🐄 talk, 🫘 contribs) 17:14, 7 February 2026 (UTC)

Your wish to punish Archive.today over this silly incident (which they undid) would cause widespread and deep collateral damage to Wikipedia. -- GreenC 18:32, 7 February 2026 (UTC)

I think that would depend on how it's implemented. First, just to remind everyone, WP:Glossary#verifiable means someone can find a reliable source. It does not mean that the Wikipedia article already has a little blue clicky number (that's WP:Glossary#cited) or that the ref contains a functional URL. This means that if the Wikipedia article says "The Sun is really big", and there's no cited source, or the cited source is a dead URL, then that sentence is still verifiable, because an editor (or reader) could look up Alice Expert's book, The Sun is Really Big, and learn that the material in the Wikipedia article matches the material published in at least one reliable source. Removing archive links therefore doesn't (usually) destroy verifiability (unless that was the only source in the world that ever said that, and the original is a dead URL – in which case, are we really sure we should be saying that now?); it just makes verifying the information take more work.

Having looked at a too-small sample size (=4 articles) with these links, I think that some of these links are unnecessary and others deserve a {{better source needed}} tag no matter what the archive status is. I therefore think that checking and replacing sources might be a good thing, overall. WhatamIdoing (talk) 19:20, 7 February 2026 (UTC)

A citation to a book is always verifiable. So is the NYT and other news outlets. Referring to everything else online-only, which is most of it. Without an archive, a dead website is unverifiable. Maybe wait 10 years for an archive to surface, but eventually it's gone. You might find other sources, but who is going to do that for half a million links? Certainly not the few people engaged in these conversations. Most people don't even verify sources, much less try to replace them with other sources. People are busy creating new citations with future dead links that nobody fixes. The debt continues to grow, and one of our best tools for dealing with it is now being threatened with removal. -- GreenC 19:44, 7 February 2026 (UTC)

Please look at the definitions I linked. We don't care whether "a dead website is unverifiable". (It's really none of our business whether people can double-check that some other website's content was taken from a reliable source vs is an original work.)

We care whether the content in the Wikipedia article is verifiable – and we care whether it's verifiable in any reliable source, not just the cited one.

Yes, you're right: half a million sources is a problem, and the debt continues to grow. To stop the bleeding, I think we should deprecate/discourage future additions of this source. To get the existing ones checked, I think we should have a tracking category, and maybe even a way to make this a more mobile-friendly and/or newcomer-friendly task. Based on my experience the other day, we're looking at about five minutes per source. Also based on my experience the other day, half the sources are unreliable ones anyway (at least for medical content). WhatamIdoing (talk) 19:55, 7 February 2026 (UTC)

If Archive.today actually goes offline, then we have another problem. But treating it like it's already offline by adding {{dead link}} templates is backwards since we don't know the future. The assumption there are alternatives to Archive.today is a mistake. Most Archive.today links are added because Wayback can't do it. There's really only two games in town, and we are eliminating one. And you can't go back and fix it, either you save the web page before it dies or it's gone forever. Archive.today has a monopoly on many archive pages, and many citations are the only game in town there are no better sources. Most people don't read these forums, but if you start blocking or hiding links, there will be many editors complaining. It's a major resource for our community that has a large following. Nobody has really been notified about the RfC. -- GreenC 21:00, 7 February 2026 (UTC)

Wikipedia:Requests for comment/Archive.is RFC 5 is live. This should probably be added to WP:CENT to get a more global consensus. Children Will Listen (🐄 talk, 🫘 contribs) 17:12, 7 February 2026 (UTC)

The bulleted editing icon

I know that in the Source editor, we use colons to indent entire paragraphs the equivalent of one tab stop from the left margin, but I prefer to work in the Visual editor and hadn't figured out how to do it there.

— So I went to the Help Desk to ask how. Another editor suggested using a bulleted icon that I would see at the top left of the screen, providing indentation options that could indent from either the left or the right. However, both he and I found those two options were greyed out. Would you please make them operative?

— One other related comment: I found that the placement of the bulleted icon was at the top left of the screen only in the Source editor, whereas in Visual it appears in the Tool bar. That's a bit confusing. Couldn't it be placed on the Tool bar in both the editors? Augnablik (talk) 10:42, 5 February 2026 (UTC)

The visual editor was designed for use in articles/mainspace, where you shouldn't normally be :indenting paragraphs. WhatamIdoing (talk) 22:17, 5 February 2026 (UTC)

Note that semantically, the colon syntax is part of the description list syntax (see Help:Wikitext § Description lists). In mainspace, this syntax should only be used for lists that define terms, such as glossaries. (Yes, the colon syntax is misused everywhere else, but we shouldn't do the same in actual articles.)

There are templates that will indent text, but generally it is preferable to use a template that is semantically appropriate for the content in question. This provides the opportunity to make the text more accessible for various assistive technologies. isaacl (talk) 02:03, 6 February 2026 (UTC)

@WhatamIdoing and @Isaacl, I think I see your points. I'd been thinking that my need for a way to indent in the VE occasionally came up in some other situations, but now I think that was mistaken.

So, then, let me slightly change what I also asked about:

1. Because all the legitimate ways to indent are covered by the first two options on the bulleted icon, bulleted and numerical, then why are there the two other indenting-from-the-let-and-right-margin options also on the menu? Although they're greyed out, it would appear to editors that they are—and should be— available.

2. Could you place the bulleted icon in the same location in both the SE and the VE for consistency and thus ease of use?

Augnablik (talk) 08:45, 6 February 2026 (UTC)

As per the documentation at mw:Help:VisualEditor/User guide, the bottom two menu items are for increasing/decreasing the nesting level of the list items. Requests regarding the user interface will have to go to the Wikimedia Foundation developers, but note since a change like that would affect all users editing pages, my suspicion is that they'd be wary about changes. isaacl (talk) 18:00, 6 February 2026 (UTC)

Tracked in Phabricator
Task T39938

It's been on the list since 2012, and AFAIK is was last considered in 2015. I think it unlikely to happen in the next few years. WhatamIdoing (talk) 05:30, 7 February 2026 (UTC)

How to see Wikipedia articles in a category or a WikiProject list missing images?

I think it would be useful to see lists of articles that do not include any image, maybe with a column for the linked Commons category if it exists and a column for the image(s) set on the Wikidata item if there are any. The articles could be those in a category or especially some WikiProject list like Wikipedia:WikiProject Climate change/Popular articles or Category:High-importance science articles (corresponding articles, not the talk pages though). I think it's not unlikely that there is some way for this.

Asking this in the context of c:Commons:List of science-related free media gaps – this could be useful not just for adding images if a useful relevant high-quality one exists for the article, but also for identifying media gaps.

By the way, I was wondering whether to post this here or at c:Commons:Village pump/Technical.
Prototyperspective (talk) 14:49, 5 February 2026 (UTC)

You could find articles that have been tagged with Template:Image requested, but I'm not aware of any way to look for untagged articles. https://pagepile.toolforge.org/ will let you define a list of target pages, and that list can be used be used by other tools for various purposes, but, again, I'm not aware of any tool that would import such a list and identify missing images.

Images are one of the key things that readers want to find in a Wikipedia article. It would be nice to have more emphasis on finding and adding appropriate images. WhatamIdoing (talk) 23:59, 5 February 2026 (UTC)

Good idea – that method shows 191 pages in this query which is something one can start with.

A way to list articles without images would probably show way more results, would be more dynamic, and could be useful in more ways. It would not rely on users adding that template which is relatively rarely done. Additionally, having that template doesn't mean the article is short of even an image illustrating the main subject and entirely lacking images (also implies there is no image for the article in the page preview hovercard and in the Wikipedia app).

Agree on what you said there. Also of note that only very few users know of, see, and click the Commons category linked to an article – there's often high-quality files there but pageviews stats show that few go to these pages. After creating many Commons categories, I think most of them over a year later weren't even linked via the small often overseen {{Commons category}} somewhere in the article. One can often find images in categories that have been there for years but nobody ever added them to the article including articles including not even one image. Prototyperspective (talk) 00:26, 6 February 2026 (UTC)

Well, as I like to tell the people fighting over infoboxes: It'd be better to start with Category:Wikipedia articles with an infobox request than any random article. Searching by templates might be better if you get into larger groups of pages; deepcat searches sometimes time out for me. WhatamIdoing (talk) 03:38, 6 February 2026 (UTC)

Regarding the search link that checks for templates instead of the categories: don't know why it only shows 70 results instead of all 191.

Regarding the search link that checks via the two categories: I've looked into it further and excluded all articles that are biographies or films. Now it contains just 58 items instead of 191 and most of these are niche low-importance articles where I can't see how an image would be very useful or they already have an image for the article's topic (as in the case of Gypsum concrete). I nevertheless added the search query to the media gaps page.

deepcat searches sometimes time out for me this happens for deeply-nested categories which is why it won't really work for Category:Science currently. This may also be an issue here because not yet all relevant articles in that category branch have been tagged with the WikiProject template. Additionally, it doesn't look like one can scan if an article is in a category and its associated talk page in another. This would be useful because the WikiProject category is only set on the talk page. There's also ways to scan for articles in a category branch that don't yet have the WikiProject template but it's complicated and I guess barely anybody uses that (a tool for that would be great btw). Prototyperspective (talk) 15:55, 6 February 2026 (UTC)

Petscan gets about 95% of the way there - you can ask for pages in a category that don't have "a lead image", which I think is the single image returned in the API. Pages with no images will presumably also have no lead image. Andrew Gray (talk) 09:56, 6 February 2026 (UTC)

Following up on this - it seems "lead image" is defined from mw:Extension:PageImages and is a) one of the first four images in the lead section which b) has a certain range of ratios and c) is not explicitly excluded. So it is possible to have an article with images that nonetheless show up as no-image here. But having said that...

• Pages with a non-free lead image linked from Wikipedia:WikiProject Climate change/Popular articles (53/1000)
• Pages with no lead image linked from Wikipedia:WikiProject Climate change/Popular articles (137/1000)

It doesn't seem to be possible to do this in one step starting with a talkpage category (like importance tags), but it is possible in two steps via PagePile.

There's also an interesting query which I've found on Quarry and tweaked - thanks to @Cryptic for coming up with it originally - which identifies six "top/high" importance Science articles with no image links on the page. Andrew Gray (talk) 18:20, 6 February 2026 (UTC)

Interesting, thanks, I didn't know about the petscan feature to only show articles without lead image.

I tried to run it on Category:Science but the problem is that it's not possible because that cat has too many subcategories and also when limiting it to e.g. just 3 layers, it (query) shows too many results (>60.000).

I first thought maybe the approach of that petscan filter isn't really adequate as it also shows articles with images even lots of images – but looking more closely, I'm not so sure anymore: e.g. Agricultural science is listed buts its image in infobox does not illustrate agricultural science; Artificial intelligence is listed despite having many images but it does not have an image at the top that's some diagram explaining AI types and/or how AI works. Articles like Anthropology also miss some image that illustrates the subject well. So maybe the issue is not with the methods but simply that there's soooo many articles missing images (I think the community hasn't really begun to systematically address this).

What would be the best ways to address this that takes into account these issues: prioritizing articles that miss images, using only other methods that check whether there is any image at all in the article², somehow further filtering the petscan, somehow extracting fields or large-order topics lacking images?

² here's one additional way to check if there's any image whatsover (or animation or video or audio) in an article: deepcategory:Science -insource:"[[File:" (82,511 articles with incomplete results) Note: in this query it also shows articles with image in infobox so these would also need to be excluded somehow (maybe via filtering out things like .png?). One could combine this with incategory:"Commons category link is on Wikidata" to see just articles with no image but a Commons category (2,086 so this one seems quite actionable).

Pages with no lead image linked from Wikipedia:WikiProject Climate change/Popular articles (137/1000) Nice query, this one seems quite actionable as well. I'll probably link that on the science-related media gaps page as well and will look for other similar WikiProject pages for which to also create such a query for and maybe extract some topics in need of illustration/image (note that an article with lots of images illustrating the various subtopics may not be missing an image much even when there no lead image and ideally we'd like to have one).

interesting query which I've found on Quarry and tweaked … identifies six "top/high" importance Science articles with no image links weird that it only shows 6 items. So it seems like currently this query is not useful but maybe it can be tweaked further until it is. The description says that have no images of any sort (not even those from templates like {{unreferenced}}) so that seems to be the cause here; maybe one could exclude images in such templates but I also wonder why Research statement shows despite that there's several PDF document icon images on the page(?)

Thanks for your investigations and very helpful contributions on this issue. Prototyperspective (talk) 14:12, 8 February 2026 (UTC)

The pdf icons aren't added by image links - in a template or otherwise - but by a css class. They're not detectable with queries against the database even if we wanted to (other than by searching for external links ending in ".pdf", which isn't practical).

Excluding images included by templates isn't possible either. We've been asking the developers for an equivalent for simple links in WhatLinksHere for over two decades. And it wouldn't help anyway, since it would also exclude images in infoboxes.

What would help is a list of specific files to ignore, like {{unreferenced}}'s File:Question book-new.svg. Or I can write queries for non-free/non-existent lead images by talkpage categories/wikiproject ratings/etc. Asking at WP:RAQ is the best way for such requests not to get lost; my free time and attention is very limited this time of year. —Cryptic 18:59, 8 February 2026 (UTC)

Suggestion Mode – new Beta Feature on Tuesday

Suggestion Mode is a new Beta Feature for the VisualEditor that proactively suggests actions that people can consider taking to improve Wikipedia articles, such as "add citation", "improve tone", or "fix an ambiguous link". The feature is locally configurable, and can be locally expanded. It will be available here as a new Beta Feature on Tuesday (and thus, in practical terms, only visible to experienced editors to begin with).

The goal of this limited early release is for us to work together to:

1. Identify what issues and improvements need to be addressed before evaluating the impact of the feature on newcomers through a controlled experiment.
2. Generate ideas for new suggestions you think would be worthwhile to implement. More on this below.

The feature is closely related to the existing Edit Check feature which shows actionable feedback to newcomers as they edit, and shares many configuration details with it.

Why Suggestion Mode?

Suggestion Mode is meant to benefit two audiences:

• [Primary] Newcomers who are eager to edit and struggle with how to start doing so constructively, plus giving them encouragement to explore the policies and guidelines.
• [Secondary] Experienced editors seeking easier ways to find out what might need fixing, and assembling the context needed to decide how and if to act.
  • Note: volunteers have helpfully created many tools/gadgets/user scripts to help with the above.[6][7] Suggestion Mode seeks to make the functionality these tools offer easier for more people and in more languages to access.

How it works

When an editor who has the Beta Feature enabled opens an article with VisualEditor, if there are any of the available types of suggestion within the article content, then one or more suggestion cards will be shown alongside. Each card contains a description of the potential problem, a link to the policy or guideline the suggestion is based on, a button to start resolving the problem, and a way to provide feedback about the suggestion itself. You can see some examples and the feedback flow below. See mw:VisualEditor/Suggestion Mode#Design for more examples.

• 
  "Add a citation" example on the article w:en:Mango
• 
  "Convert citation" example on the article w:en:Gooseberry
• 
  "Revise tone" example on the article w:en:Melon
• 
  Desktop feedback workflow
• 
  Mobile feedback workflow

The team has started with an initial set of suggestions to demonstrate the concept. They are derived from existing tools, policies, and content guidelines. We're very interested in your recommendations for additional types of suggestions, to add to the growing list in T360489. The complete list of initial suggestions can be seen at Special:EditChecks. You can test the feature immediately, via this user script.

Local configuration

The aspects of a suggestion that are community configurable will vary on a case-by-case basis. They can be configured by admins at MediaWiki:Editcheck-config.json, to enable/disable individual suggestion types and control parameters for each type (e.g. the categories and sections it should/should not be shown within, the cumulative edits someone must have made to see a suggestion, etc.). The listing of available parameters is at mw:Edit check/Configuration. In particular, the textMatch suggestion type is a relatively simple system that finds words or phrases within the text, and suggests either replacing, deleting, or thinking about the text (along with a contextual guidance link). That sub-feature is easily expanded/adapted in any way you wish. In the future, we hope to support regex for these suggestions.

Known issues

The team is currently working on: YAdding the ability to include links within the text-match types of Suggestions (e.g. The "English variant specified" type will link to MOS:RETAIN next week) (T416511); YAdding the editsuggestion-visible tag to monitor edits that are made when any Suggestions have been seen (T413419); Adding the ability to see the specific suggestions someone acted on within a given edit session (T416535); Improving the feedback flow to be more streamlined (T401739); Adding the ability to toggle the visibility of the Suggestions cards entirely (T415589).

Get involved

For now, Suggestion Mode will be available as a Beta Feature, in order to collect your recommendations for: changes to the default "descriptions" (both the wording and the links), feedback on the individual suggestions and their results, and requests/ideas for further types of suggestions. The team and some volunteers have been experimenting with the checks for the last few weeks, plus discussing the tool in Discord and Phabricator, and the team has fixed a number of issues, but we need your help finding more ways to improve this feature. We also hope you will have additional ideas for new types of suggestions, that can either be implemented entirely locally as text-match suggestions, or requested for developer-assistance in making more complex suggestions; there is a listing of existing suggestions in T360489. Please share your thoughts on the feature either here or at mw:Talk:VisualEditor/Suggestion Mode, and use the built-in feedback system to share any details about problems with specific suggestions. Much thanks, Quiddity (WMF) (talk) 00:29, 6 February 2026 (UTC)

Looks pretty cool. can't wait to try it out. —TheDJ (talk • contribs) 09:08, 6 February 2026 (UTC)

Agree, I think so too. This could be quite useful for the community. What I don't like here is that it's just for the VisualEditor and not the wikitext editor (albeit this feature is probably most useful for newish editors and not so much for active editors already overburdened with tasks anyway who don't benefit much from any such further ones – those are probably mostly using the wikitext editor but I could be wrong about that). Prototyperspective (talk) 15:28, 6 February 2026 (UTC)

Personally I think the difficulty involved in making it work in both modes would not be worth the extra dev effort. -- asilvering (talk) 18:36, 6 February 2026 (UTC)

Yeah, the problem is that you need two completely different systems when looking for suggestions to make and when applying them to the document. VisualEditor is easier to do, because it offloads the whole "you must parse and modify wikitext without breaking it" part to Parsoid, and lets us work with something that's already got some level of semantic meaning applied to it.

We actually could reuse this, kind of, by (essentially) running VisualEditor in the background, and having your wikitext sent into the API and parsed, working out what suggestions there are, then asking the API again to tell us what ranges in the wikitext source they correspond to. Then doing similar things when you want to take an action in response to a suggestion, etc. It'd be painful and slower, as you might imagine. DLynch (WMF) (talk) 17:55, 9 February 2026 (UTC)

This feature is now available. You can enable it at Special:Preferences#mw-prefsection-betafeatures. (If you have previously selected the preference for "Automatically opt-in to new Beta Features", then you still need to open your Preferences page once, in order to enable any new type of Beta Feature.)

Please do share your thoughts and feedback (and especially your ideas for other types of Suggestion that could be implemented (either by the team, or by yourselves locally via textMatch), that might be helpful for newcomers to act-on and learn-from) so that we can continue to improve it for you. Thanks. Quiddity (WMF) (talk) 18:52, 11 February 2026 (UTC)

Geohack template broken

Can someone with the relevant permissions and technical knolwedge please revert Template:GeoTemplate back to a state where English language geohack works? The error seems to have been introduced yesterday. Tæppa (talk) 13:28, 6 February 2026 (UTC)

It's not protected so I suggest reverting to an earlier working version. Pinging @Tæppa — Martin (MSGJ · talk) 14:03, 6 February 2026 (UTC)

It is extended confirmed protected. Tæppa (talk) 14:10, 6 February 2026 (UTC)

Wrong user! @Mapeh — Martin (MSGJ · talk) 14:04, 6 February 2026 (UTC)

This is not a problem with Template:GeoTemplate (see Template talk:GeoTemplate#What happened?). It's at toolforge, but I cannot work the best way of filing a phab: ticket for a toolforge problem. --Redrose64 🌹 (talk) 22:49, 6 February 2026 (UTC)

(Discussion also at w:Template talk:GeoTemplate#What happened?)

Pinging @Trappist the monk: who edited the w:Module:Lang the day before (some) of the geohack language pages broke. My original thought was that the use of <br />{{lang|ar|خَرائط فلسطين المَفتوحة|rtl=yes}} broke the page, but <br />{{lang|he|עמוד ענן|rtl=yes}} has been there a while.

Just for interest:

Arabic (rtl) works [8], Hebrew (rtl) works [9], Pashto (rtl) doesn't work [10], Yiddish (rtl) doesn't work[11], Chinese (ltr) doesn't work [12], Russian (ltr) works [13]. Tæppa (talk) 00:08, 7 February 2026 (UTC)

The change I made at Module:Lang was for {{transliteration}}. At this writing, there are ten {{lang}} templates in {{GeoTemplate}}. All ten are used for presentation and have nothing to do with the &language= query portion of the geohack url.

—Trappist the monk (talk) 01:12, 7 February 2026 (UTC)

Thanks for the clarification.

For the record, all of the above links (Arabic through Russian) are now "broken" so it's probably nothing to do with the templates that the wikipedias have for each language. Tæppa (talk) 22:02, 8 February 2026 (UTC)

As someone who reads a lot of geographic articles, this GeoHack glitch has been annoying for the last couple of days. I think the correct place to file a bug report is here. In case it helps to debug: I notice that, upon opening GeoHack, the table of links to Google Maps, OSM, and other services very briefly displays but disappears after a split second. If I use Firefox's "reader view", all the links are visible. HTH. ~2026-87494-1 (talk) 01:49, 9 February 2026 (UTC)

The reason Geohack is broken seems to be related to this change: 1234316. Geohack was depending on html comments to find the main content. These html comments are removed in the latest mediawiki update. --wimmel (talk) 08:57, 10 February 2026 (UTC)

User:BrandonXLF/AJAXUndo doesn't work

There's no "ajax undo" button. How to fix this? sapphaline (talk) 20:27, 6 February 2026 (UTC)

@Sapphaline: Have you tried, perhaps, leaving a message at User talk:BrandonXLF? --Redrose64 🌹 (talk) 22:53, 6 February 2026 (UTC)

@Sapphaline: It appears you have only tried it at the end of a huge faulty User:Sapphaline/common.js. Does it work if loading it is the only command? Then the problem is not in the loaded script but your own JavaScript. PrimeHunter (talk) 00:15, 7 February 2026 (UTC)

I did try to load it from another source; it doesn't work. sapphaline (talk) 07:30, 7 February 2026 (UTC)

@Sapphaline: It works for me. Load it as the ONLY command as in line 1 of 1, not line 1435. PrimeHunter (talk) 09:59, 7 February 2026 (UTC)

Omg, thats the most common.js i have EVER seen. Id be surprised you can get anything done, the page must be dancing before your eyes and take seconds to load. —TheDJ (talk • contribs) 11:40, 7 February 2026 (UTC)

Its size doesn't affect page loading speed on my end in fact, at least in a way that's noticeable. sapphaline (talk) 14:52, 8 February 2026 (UTC)

Fixed by moving the script to the top of the page. sapphaline (talk) 15:14, 8 February 2026 (UTC)

Clip history link by CSS

On no.wiki, hist links in recent changes and user contributions have been expanded to historikk (the Norwegian equivalent of "history"). Five letters are much space wasted on a narrow screen. Please, is there some CSS that will clip this link to a given number of pixels so that it doesn't occupy so much space? I would like to know what it is. Utfor (talk) 00:41, 7 February 2026 (UTC)

@Utfor: See Wikipedia:Village pump (technical)/Archive 226 § "hist" changed to "history" in interface. The best solution is to wait for T244411. NguoiDungKhongDinhDanh 01:06, 7 February 2026 (UTC)

Waiting won't work in this case. Jon Harald Søby, a translator over on translatewiki marked the "historikk" translation as not outdated even after the English was reverted to "hist". This implies that there are no plans to bring the short form back in the Norwegian interface. No matter how long OP waits, it's not going back to "hist" unless someone else goes to translatewiki and changes it. Warudo (talk) 01:12, 7 February 2026 (UTC)

Note that the translation was changed 40 minutes after I left this comment. Warudo (talk) 21:31, 7 February 2026 (UTC)

You can ask a Norwegian administrator to create no:MediaWiki:Hist with a wanted text for everybody at that wiki. PrimeHunter (talk) 01:37, 7 February 2026 (UTC)

Monospace fonts look smaller than sans-serif

Exactly what the title says. For example:

Test 1 test 2

The latter text ("test 2" in monospace) will look smaller than the sans-serif text. This leads to, for example, my signature appearing smaller than others. Without using font-size:150% as it is against signature rules, how can I make my signature font size the same as the sans-serif text? TheTechie^[she/they] | talk? 06:43, 7 February 2026 (UTC)

If it helps, I am on (Arch) Linux, on LibreWolf (a Firefox fork; WebGL is enabled). I can't recall if this occurs on other OSes. TheTechie^[she/they] | talk? 06:44, 7 February 2026 (UTC)

Hmmm, on my screen, it only looks a little bit smaller, and only vertically. Horizontally, the monospace looks wider, if anything.

Cute signature, by the way! I don't think you need to fret so much over the size. MEN KISSING (she/they) T - C - Email me! 06:55, 7 February 2026 (UTC)

I know, but it annoys me enough that I had to find a solution :) TheTechie^[she/they] | talk? 18:43, 7 February 2026 (UTC)

This seems like an instance of Wikipedia:Typography#The monospace "bug". I think the best solution here is to use font-family: monospace, monospace instead of font-family: monospace:

Test 1 test 2

LightNightLights (talk • contribs) 08:27, 7 February 2026 (UTC)

Yeah, that appears correctly on my system. Thank you; I will edit my signature to look like that. TheTechie^[she/they] | talk? 18:40, 7 February 2026 (UTC)

£, s, d?

Crosspost (sort of) from Template talk:Pounds, shillings, and pence#Alternative coding, I've been looking for something that lets me output non-decimalised numbers. So far I've been using {{GBP|10 8s 9d}} to produce £10 8s 9d for example, but leaving what is effectively text inside the curly brackets doesn't feel right and might break downstream applications e.g. inflation calculators.

I suppose the ideal outcome might be {{GBP|x|y|z|nd}}, with numbers in place of x, y, z, and "nd" for "non-decimalised". Empty y or z values would return a negative sign, e.g. £x/-/z or £x/y/-, and an empty x value would skip the "£" symbol as well e.g. y/z, y/- or -/z. Of course, values y and z would also need to be capped, with any excess being moved to the prior column e.g. 30 shillings would be displayed as £1/10/-, unless some extra term like |abbr=on was included?

Critically, the template system cannot require typing the £ symbol, because not everyone has that on their keyboard. Also, whatever system ends up working for this should be copy-able to other areas e.g. Australia where the three-piece currency system used to apply. ({{AUD|}} doesn't currently support £sd but it might one day.)

Anothersignalman (talk) 16:03, 7 February 2026 (UTC)

@Anothersignalman: While this could be implemented, I'm unsure why you want this. What's the problem with just {{GBP|10}} 8s 9d to get £10 8s 9d without including any letters? – Scyrme (talk) 23:08, 9 February 2026 (UTC)

Also, after looking at the code for {{£sd}} and {{GBP}}, it would probably be much more straightforward and cleaner to implement this as a new template rather than modify the existing ones. (And if you want the inflation calculator to work, that would probably require converting the input to decimal then converting it back. It doesn't look like the inflation calculator is set up to take nondecimal inputs.) – Scyrme (talk) 23:21, 9 February 2026 (UTC)

I've since swapped over from {{GBP}} to {{Australian pound}} for my articles, which I didn't know about before and seems to have solved my problem. My original concern was that I wanted to keep all numbers related to specific figures within a set of curly figures on a matter of principle, but the inflation element was a secondary concern. If I still needed this, I'd have accepted a modification to {{GBP}} so that it could take a decimal input and output the £sd arrangement, because I could then request or find a separate template that took £sd inputs to generate the decimal output and put that inside the GBP set. This would also have solved issues where a source reported a cost as say 30s instead of £1 10s. Anothersignalman (talk) 05:46, 10 February 2026 (UTC)

@Anothersignalman: Are you only using {{Australian pound}} with Australian predecimalised currency? The template links to Australian pound when it uses £. If you're also intending to use this with UK currency, maybe it would be helpful to modify the template to allow that link to vary (or to display no link by default), and move the template to a broader title? ({{Australian pound}} would exist as a redirect after the move, so the existing uses would still work.) – Scyrme (talk) 17:02, 10 February 2026 (UTC)

No, I'm writing an Australian article, I just used GBP because it had the right symbols and the two currencies were tied to each other in the relevant time frame. Anothersignalman (talk) 08:10, 11 February 2026 (UTC)

Page move failed

I attempted to move the page maximum weight matching to maximum-weight matching. Formerly, the latter had been a redirect to the former. I wanted it to be the other way around. It didn't work and I have not been able to restore the material. What is going on? Michael Hardy (talk) 18:22, 7 February 2026 (UTC)

You somehow managed to move the page twice, which deleted the actual page in the process. I've sorted it out now. * Pppery * _{it has begun...} 18:32, 7 February 2026 (UTC)

I've done that when I was an admin once or twice, I think,, by pressing enter twice quickly while moving the page. Graham87 (talk) 01:32, 8 February 2026 (UTC)

Differences between languages of Wikipedia

On Chinese and English, also French Wikipedia, you can directly link to 345; but on German, you can only have 51.
And German has many things special. For example, the font of the title is very different (on computer). Why?

If there is a specific page helpful to answer my questions, please link it. Noordpunt (talk) 19:14, 7 February 2026 (UTC)

@Noordpunt: I don't know what you mean by "you can directly link to 345". de:345 is a working link to a German article. Each Wikipedia language is edited independently. The German Wikipedia sets font-family for page titles at de:MediaWiki:Vector-2022.css#L-46 which can be edited by local administrators. There is a link to a German survey. PrimeHunter (talk) 20:48, 7 February 2026 (UTC)

Sorry, I did not describe clearly.

I meant that, "from the FR, ZH and EN Wikipedia home page, I can visit (through the label 文A) from there to 345 versions of languages; however, on the main page of DE ([14]), I can only visit 51 languages." I feel confused on this. And I know that diff. versions of Wikipedia are linked together in Wikidata. But does it mean that German does not link to enough number?

And another question, for ;, how to input this in Visual Editor? Noordpunt (talk) 15:31, 8 February 2026 (UTC)

Here on the English Wikipedia, the main page which is linked to the Wikidata item Q5296. All the other language Wikipedia main pages that also share that item are available from the language menu. It looks like the German Wikipedia main page is also linked with the same Wikidata item so, in theory, by default it should have all the same language options as the English version. However, clearly the German Wikipedia is overriding this.

Looking into it, it seems the German Wikipedia has a consensus to not link every available language on its mainpage. For information, see de:Wikipedia:Interwiki-Links auf der Hauptseite. – Scyrme (talk) 17:19, 8 February 2026 (UTC)

Regarding ;, I don't know if the Visual Editor has an equivalent for that Wikitext markup. I wasn't able to find an option to input it. It might just not be available. – Scyrme (talk) 17:23, 8 February 2026 (UTC)

Our own Main Page has a selection {{Wikipedia languages}} of large languages as part of the page itself and then the normal full list of 345 languages made by MediaWiki from the Wikidata item Wikimedia main page (Q5296). The German main page de:Wikipedia:Hauptseite made another system. The page itself has no links to other languages but they made a selection of 51 languages to replace the Wikidata list (I don't know how they suppressed the full list). The box at top of their main page has a link on "Sprachversionen" (Language versions) to a German wiki page de:Wikipedia:Sprachen with all languages. It's a manually made list and currently says 344 languages instead of 345 so maybe it isn't completely up to date. PrimeHunter (talk) 18:04, 8 February 2026 (UTC)

This sounds like it might be to do with mw:Universal Language Selector/Compact Language Links. --Redrose64 🌹 (talk) 23:22, 8 February 2026 (UTC)

It's not, it's done by good ol' interwiki links. Nardog (talk) 01:53, 9 February 2026 (UTC)

Cite error: The <ref> tag has too many names

Hi, I can't see how to fix the giant red "Cite error: The <ref> tag has too many names" error message on Reference 2 in John Chard. Can anyone? Thanks, DuncanHill (talk) 20:56, 7 February 2026 (UTC)

HerbertGP36 recently converted several references to use {{sfn}} but forgot to remove a couple of <ref (literal) wikitext statements, which made the system think that they were opening a big reference. Izno (talk) 21:19, 7 February 2026 (UTC)

@Izno: Thanks, now I can see the diff I'll have a better idea what to look out for next time. Now I'll fix the no-target error introduced by your edit! DuncanHill (talk) 21:21, 7 February 2026 (UTC)

Correction: You didn't introduce it, it was hidden by the ref tags! DuncanHill (talk) 21:24, 7 February 2026 (UTC)

Global lock for QBA-bot

If you can't ask a question related to the bot error of the administrator of the Russian wikipedia, then where do I have the right to ask so that my question is not deleted? Where and to whom should I contact? Is there a legal section on Wikipedia? Question HERE. ~2026-86834-9 (talk) 06:16, 8 February 2026 (UTC)

Пишите в рувики. Тут бесполезно. Saiduzbek (talk) 12:26, 8 February 2026 (UTC)

• Global lock requests are handled at meta:SRGL. — xaosflux ^Talk 13:13, 8 February 2026 (UTC)

  The article is protected from editing. Saiduzbek (talk) 13:44, 8 February 2026 (UTC)

campaign-external-machine-translation

I noticed this tag seems to have appeared on quite a few edits recently, but I have no idea what it's supposed to mean. Does anyone else have an idea? Thanks. Sugar Tax (talk) 19:01, 8 February 2026 (UTC)

That came up a few years ago (see Wikipedia:Village_pump_(technical)/Archive_204#New_tag?), not sure why it would start appearing again. Schazjmd (talk) 19:44, 8 February 2026 (UTC)

A test [15] shows the tag is applied if you use VisualEditor with this in the url: veaction=edit&campaign=external-machine-translation. I constructed the url manually but others are unlikely to do that so I guess there is still something out there which can generate such url's. Whether or not external machine translation was actually used, the tag is so rare (often one daily edit) that it seems harmless. PrimeHunter (talk) 20:43, 8 February 2026 (UTC)

Such links are generated when you visit a Google Translate view of a Wikipedia article (e.g. [16]), then click the "Contribute" link next to "Automatic translation", and then click the "Edit the original version" button on that page. It doesn't indicate that anything in the edit was machine-translated, but that the editor followed the prompts to edit from a machine-translated version of a page.

I filed a bug about these tags not being labelled properly: T416876. Matma Rex talk 13:51, 9 February 2026 (UTC)

columns

Hello, please advise me how to set the WORLDS & EURO columns width to 50%. I mean, they should be the same width:

Rank	Team	Qualified for the following tournaments
		WORLDS	EURO
1	A	yes	yes
2	B	no	yes
3	C	no	no

I'm trying everything, but it's not working. Thanks, Maiō T. (talk) 22:35, 8 February 2026 (UTC)

@Maiō T.: I ignored "50%" which is far too wide on desktop screens. Here I use style="width:10em;" in both columns to give the same width if there isn't wide content forcing a larger width:

Rank	Team	Qualified for the following tournaments
		WORLDS	EURO
1	A	yes	yes
2	B	no	yes
3	C	no	no

Don't use it in wider tables which will require horizontal scrolling on many smartphones. Then it's better to let the browser pick the smallest possible width. 10em would normally be too large for a brief word but I chose it to avoid line wrapping in the header. If you want "half of however wide the colspanned header becomes to fit its content" then I don't know whether it's possible. PrimeHunter (talk) 23:48, 8 February 2026 (UTC)

Exactly, PrimeHunter, I meant what you wrote at the end. But sometimes it works; see the following code and result:

{| class="wikitable" style="text-align:center"
! style="width: auto" colspan="2" | Qualified for the following tournaments
|-
! style="width:50%" | WORLDS
! style="width:50%" | EURO
|}

Qualified for the following tournaments
WORLDS	EURO

But when I add a column to the left side, it all goes wrong. Bad luck... Maybe it's an HTML bug, or something. It would be really cool if someone fixed it... Maiō T. (talk) 11:50, 9 February 2026 (UTC)

No, it's because 50% + 50% + something is more than a hundred percent. HTML defines an order of priority. So it will go: something + 50% of total space + 50% of total space... oh, that won't fit, you will only get the remaining space instead (so 50% of total space - something). —TheDJ (talk • contribs) 16:58, 9 February 2026 (UTC)

Broken duplicate args warning

Tracked in Phabricator
Task T416270

Not sure where to raise this issue, but MediaWiki:Duplicate-args-warning seems to be broken... Preview this permalink and see that the Duplicate args warning has 2 red links where there should be valid blue links. Anyone know what is going on here??? Zackmann (^{Talk to me}/_{What I been doing}) 01:29, 9 February 2026 (UTC)

Looks like it was broken by gerrit:c/1227448. The code switched from fetching the warning messages with ->text() and then parsing the wikitext later to ->parse(). That interacts poorly with how gerrit:c/731169 changed the parameters from this message from normal to "plaintext" even though they contain page titles, resulting in the placeholders (which include various kinds of quotes for a historical security thing) getting percent-encoded in the URL. phab:T310015 appears to be a similar past issue. I see someone else has also filed phab:T416467 about the same issue with this message. Anomie⚔ 02:25, 9 February 2026 (UTC)

Also, looks like MediaWiki:template-loop-warning may have the same issue, as it was changed to "plaintext" in the same way in gerrit:c/731169. Anomie⚔ 02:27, 9 February 2026 (UTC)

Good to know someone is on it! Thanks. Zackmann (^{Talk to me}/_{What I been doing}) 03:11, 9 February 2026 (UTC)

@SomeRandomDeveloper FYI Matma Rex talk 13:57, 9 February 2026 (UTC)

Thanks for notifying me. I've uploaded a fix at gerrit:c/1237914. SomeRandomDeveloper (talk) 14:37, 9 February 2026 (UTC)

Template:Infobox theatre

Template:Infobox theatre redirects to Template:Infobox venue

venue is real estate, what about the organizations that perform live performances ?

e.g.: Naked Angels (theater company) which changes venues

Piñanana (talk) 02:23, 9 February 2026 (UTC)

@Piñanana: For infoboxes you just pick the one where the parameters are the best fit for the article and what you want to say. The infobox name is not seen by readers. A theater company could also use {{Infobox performing arts company}}. PrimeHunter (talk) 04:14, 9 February 2026 (UTC)

Help with conditional expressions

I am drafting a template at Template:Career FNCS results middle/sandbox. It works fine when I use #switch as follows:

{{#switch:{{{made_grands|}}}|true=Yes|false=''Eliminated in {{{elimination_stage}}}''}}

However, it seems to me that the code above could be re-written using #if as follows:

{{#if:{{{made_grands|}}}|Yes|''Eliminated in {{{elimination_stage}}}''}}

If I do this, and enter made_grands=false, the parameter is still treated as true, and thus produces "Yes". I am very new to conditional expressions; am I doing something obviously wrong? Rockfighterz M (talk) 21:37, 9 February 2026 (UTC)

@Rockfighterz M: #if tests for non-empty. See mw:Help:Extension:ParserFunctions##if. PrimeHunter (talk) 21:52, 9 February 2026 (UTC)

@Rockfighterz M I think you're looking for #ifeq:, not #if:, but even then you'd need two nested if statements to cover the case where the value is neither "true" nor "false". --Ahecht (TALK
PAGE) 18:47, 10 February 2026 (UTC)

Tech News: 2026-07

Latest tech news from the Wikimedia technical community. Please tell other users about these changes. Not all changes will affect you. Translations are available.

Updates for editors

• Logged-in contributors who manage large or complex watchlists can now organise and filter watched pages in ways that improve their workflows with the new Watchlist labels feature. By adding custom labels (for example: pages you created, pages being monitored for vandalism, or discussion pages) users can more quickly identify what needs attention, reduce cognitive load, and respond more efficiently. This improves watchlist usability, especially for highly active editors.
• A new feature available on Special:Contributions shows temporary accounts that are likely operated by the same person, and so makes patrolling less time-consuming. Upon checking contributions of a temporary account, users with access to temporary account IP addresses can now see a view of contributions from the related temporary accounts. The feature looks up all the IPs associated with a given temporary account within the data retention period and shows all the contributions of all temporary accounts that have used these IPs. Learn more. [17]
• When editors preview a wikitext edit, the reminder box that they are only seeing a preview (which is shown at the top), now has a grey/neutral background instead of a yellow/warning background. This makes it easier to distinguish preview notes from actual warnings (for example, edit conflicts or problematic redirect targets), which will now be shown in separate warning or error boxes. [18]
• The Global Watchlist lets you view your watchlists from multiple wikis on one page. The extension continues to improve — it now properly supports more than one Wikibase site, for example both Wikidata and testwikidata. In addition, issues regarding text direction have been fixed for users who prefer Wikidata or other Wikibase sites in right-to-left (RTL) languages. [19][20]
• The automatic "magic links" for ISBN, RFC, and PMID numbers have been deprecated in wikitext since 2021 due to inflexibility and difficulties with localization. Several wikis have successfully replaced RFC and PMID magic links with equivalent external links, but a template was often required to replace the functionality of the ISBN magic link. There is now a new built-in parser function {{#isbn}} available to replace the basic functionality of the ISBN magic link. This makes it easier for wikis who wish to migrate off of the deprecated magic link functionality to do so. [21]
• Two new wikis have been created:
  • a Wikipedia in Jju (w:kaj:) [22]
  • a Wikipedia in Nawat (w:ppl:) [23]
• View all 23 community-submitted tasks that were resolved last week.

Updates for technical contributors

• A new global user group has been created: Local bots. It will be used internally by the software to allow community bots to bypass rate limits that are applied to abusive web scrapers. Accounts that are approved as bots on at least one Wikimedia wiki will be automatically added to this group. It will not change what user permissions the bot has. [24]
• Detailed code updates later this week: MediaWiki

Meetings and events

• The MediaWiki Users and Developers Conference, Spring 2026 will be held March 25–27 in Salt Lake City, USA. This event is organized by and for the third-party MediaWiki community. You can propose sessions and register to attend. [25]

Tech news prepared by Tech News writers and posted by bot • Contribute • Translate • Get help • Give feedback • Subscribe or unsubscribe.

MediaWiki message delivery 23:28, 9 February 2026 (UTC)

Help with transliteration style

I need help with my custom CSS.

I am specifying special fonts for the Hebrew text and different ones for transliterated Hebrew text (he-Latn), however the former always overrides the latter, lately.

I have the following line to specify Hebrew text style: span[lang|=he]:not(.he-Latn-fonipa, .he-Latn) { font-family:...}

And the following line to specify transliterated Hebrew text: span[lang|=he-Latn]:not(.IPA) { font-family:...}

I was told about the phrase not(.IPA) here before to prevent it from overriding my IPA style specification: .IPA { font-style:...} and span[lang|=he-Latn-fonipa]

What to do to still specify the transliterated Hebrew text without having conflicts with the Hebrew text? --Esperfulmo (talk) 04:26, 10 February 2026 (UTC)

span:lang(he-Latn):not(:lang("*-fonipa")). Nardog (talk) 04:57, 10 February 2026 (UTC)

Thanks, but after trying that, the problem persists. The transliterated Hebrew still uses the Hebrew style. :( --Esperfulmo (talk) 06:14, 10 February 2026 (UTC)

Oh, I thought you were trying to specify a font for transliterated Hebrew as opposed to Hebrew IPA. If you want to specify a font for Hebrew but not romanization or IPA, :lang(he):not(:lang("*-Latn")) should work. Nardog (talk) 06:18, 10 February 2026 (UTC)

So, it worked after I set the font specification to !important. Thanks. --Esperfulmo (talk) 06:20, 10 February 2026 (UTC)

Mobile table of contents experiment: Phase 1

Hi everyone,

I’m posting on behalf of WMF's Reader Growth team. The week of February 16 we will launch an A/B/C test that adds a Table of Contents on mobile web and auto-expands all article sections as a way to get readers information faster by addressing navigation difficulties. Our hypothesis is that by giving readers a table of contents to see sections at a glance, they will be able to more easily find what they’re looking for.

Why are we doing this?

We see this work as a part of addressing the decline in pageviews on Wikipedia. We want it to be easier to access content on the site, especially on mobile where newer readers tend to come in. WMF’s Reader Foundational Research found that difficulty with in-article navigation, in particular mobile, is a top complaint among readers. We’re trying out a table of contents on mobile web to see if it supports ease of browsing based on data that it can be helpful for navigating. The Wikipedia Android app, for example, has a table of contents, which on average gets opened almost 4 times per user, much more often than users start a search, which is only an average of 1.5 times a session. The app also sees 71.1% clickthrough rate, indicating strong usage on small screens.

These screenshots show the two different table of contents buttons that will be shown to experiment participants in the two treatment options.

What idea are we testing?

Article sections are currently collapsed by default on mobile, which was intended to save users time in navigating as they scroll through long paragraphs of text. However, we suspect that this default may contribute to navigation difficulties since users must first open individual sections before reading. In December 2025, we conducted an experiment on Arabic, Vietnamese, French, Chinese, and Indonesian wikis to 1) auto-expand all sections in an article by default and 2) pin the header of the section in the viewport to the top of the page.

We found that this change actually lowered the retention rate for readers by about 1.5% and shortened the amount of time they spent onwiki. We suspect that auto-opening all the sections on mobile ended up causing navigation difficulties by creating a wall of text, resulting in readers feeling overwhelmed or frustrated and leaving. So we decided to try out something different.

Now we want to see if offering a Table of Contents will improve those navigation needs. The new test will add a Table of Contents button on mobile. When users tap it, a panel slides up from the bottom showing the article’s section headings, which they can then click to jump to different parts of the page.

These screenshots show the two different table of contents interfaces that the two treatment groups will see.

What stage is this project in?

This project is in phase 1: launching a small test with an early version of these ideas.  It’s not yet clear whether this feature will be an improvement for readers, so we want to test it to determine whether to proceed into phase 2: building a feature.  

What is the timeline?

The experiment will go live the week of February 16 and will run for four weeks. It will affect up to 10% of mobile users on Arabic, Vietnamese, French, Chinese, and Indonesian Wikipedias and up to 1% of mobile users on English Wikipedia. Once we have the results, we will come back here to discuss results and decide whether we want to proceed with this idea.

Thank you! EBlackorby-WMF (talk) 15:59, 10 February 2026 (UTC)

Something to entertain since you're poking around would be to display some lesser set of the table of contents, perhaps all items pointing to an (h1), h2, or h3, and excluding any pointing to h4s. Izno (talk) 19:30, 10 February 2026 (UTC)

Indexed article omitted from Google

Hi folks, not sure if this is the best place to post this but it seemed like a high-visibility spot where someone might have an answer. Feel free to move or copy my message elsewhere if you'd like.

I created the article Nova Scotia Guard on 1 July 2025. The article appears to be indexed, and is the third result on DuckDuckGo. However, it will not show up on Google at all. Even when searching "Nova Scotia Guard Wikipedia", you'll get articles it's linked to and even a category it's in, but not the article itself. I was particularly perturbed by the fact that the Grokipedia clone of the article shows up in Google, but not the one I created. I mentioned this in the Wikipedia Discord server some time ago and my results were replicated by several other users. Since then I created a redirect, edited the Wikidata item, and added more links to the article, but it hasn't changed anything.

My biggest concern here is that there may be other articles which Google is not showing in search results for one reason or another. If someone might be able to look into this I'd appreciate it. Thanks, MediaKyle (talk) 16:04, 10 February 2026 (UTC)

It was marked as reviewed in September which should allow it to be indexed, but I checked Google Search Console and for some reason Google hasn't crawled it since July when it was noindexed. I requested a re-crawl, so hopefully it will start showing up soon. the wub "?!" 16:44, 10 February 2026 (UTC)

@MediaKyle: It hasn't been edited since 20 August 2025 where it was still noindexed. I get the impression Google is watching our edit logs and often revisits a page shortly after it has been edited so any edit (except an unlogged null edit) may influence them. PrimeHunter (talk) 17:31, 10 February 2026 (UTC)

Thanks for the replies, I appreciate you both looking into this. I thought I edited the article the other day but I guess I did everything except that... Just made an edit. Hopefully it will show up soon and this is just an isolated incident. Cheers, MediaKyle (talk) 17:49, 10 February 2026 (UTC)

@MediaKyle, fyi I just searched and we were the 3rd result. Dw31415 (talk) 18:55, 10 February 2026 (UTC)

Thanks for letting me know, just checked and it shows up for me now as well. Seems this is resolved now... still have to wonder what other articles might be caught by this oddity but I can't imagine it's too widespread. MediaKyle (talk) 19:01, 10 February 2026 (UTC)

Hm. I think maybe quite a lot, actually, if the reason something wouldn't be indexed is "no edits since an NPR hit the reviewed button". -- asilvering (talk) 05:22, 11 February 2026 (UTC)

This is pretty common (and pretty damaging for our Google scores probably) —TheDJ (talk • contribs) 13:27, 11 February 2026 (UTC)

Temporary Wikipedian userpages

In the past couple of weeks Special:WantedCategories has seen more than one recurrence of a redlinked Category:Temporary Wikipedian userpages that was deleted in 2016. Both times, it was populated entirely by the user talk pages of editors who were blocked for vandalism in 2008, the first time exclusively editors whose usernames began with W, and today exclusively editors whose usernames began with V — and the culprit appears to be that said talk pages have recently been undeleted on WP:DELTALK grounds, after having been previously deleted, and were thus put back into a category that existed at the time of the original deletion but has not existed for a decade.

So is there another way that this can be resolved without forcing me to gnome it out in an WP:AWB run every time it comes back again? Bearcat (talk) 17:28, 10 February 2026 (UTC)

The category is currently empty. Please always include an example. I found one in your contributions: [26]. There is no way to prevent the categorization before you removed it. The page was undeleted by Hex. Her logs show she undeleted many such pages beginning with V on 7 February and with W on 29 January. If she is planning to undelete more pages then you could ask if she will remove the category afterwards but now I have also pinged her. PrimeHunter (talk) 18:03, 10 February 2026 (UTC)

Hiya - yes, I'm repairing a mass deletion of user talk pages by a former admin back in 2008 before we decided not to do that. It's slow and tiresome because I check each of them to ensure there's nothing requiring RevDel before hitting the button, so I was planning to ask someone to get a bot to clear off the category afterwards. I guess you'd like me to arrange that now? Given that I've done about 600 out of 11,000, this is going to take a while. By the way Bearcat, since you evidently checked the logs, you could have written to me first before posting here. It's also kind of odd that you didn't mention me and so PrimeHunter had to send a ping.  — Hex • talk 19:58, 10 February 2026 (UTC)

@Hex: Now that the topic is raised, I do have to wonder what purpose is served by undeleting these talk pages. Was there a discussion somewhere that concluded these 11000 pages would be useful to mass-restore 18 years later? Anomie⚔ 23:59, 10 February 2026 (UTC)

I have the same question. I looked at just one example, which had five Linter errors and one nonexistent category. I expect to see deleted templates as well. Restoring these pages will make work for a lot of gnomes; what is the benefit, and where was the discussion about this restoration? – Jonesey95 (talk) 00:32, 11 February 2026 (UTC)

No discussion was required. We established consensus a long, long time ago that user talk pages shouldn't be deleted except in rare circumstances because they form an important part of the historical record. When that happened, someone should have done this job, but nobody did. I'm rectifying that error. The effort of dealing with a small number of linter issues will be outweighed thousands of times over by the benefits of not having a massive chunk of user interactions and block log context missing for no good reason.  — Hex • talk 01:43, 11 February 2026 (UTC)

And what advantages are those, exactly? Since no one has cared in 18 years, it seems unlikely they're that significant. Anomie⚔ 12:57, 11 February 2026 (UTC)

thousands of times over? Actual human and bot editors are going to have to make thousands of edits to remove errors from these restored pages. That is a guaranteed downside if this crusade project goes forward. Hex: Please enumerate concrete instances that balance the downside of those thousands of edits with benefits. I won't ask you to justify the obviously unjustifiable orders of magnitude that you claim. Just a simple positive or break-even counterbalance would be fine. – Jonesey95 (talk) 15:09, 11 February 2026 (UTC)

A quick and dirty sql says that among the undeleted pages with linter issues, most have issues with obsolete tags, no background inline (which a number of wikipedians regard having a lot of false positives) and no end tags. Most of this can be automatically fixed. Everything else is less than 20 pages with issues.

Do users need to have a set number of edits to have userpages, and if so, why? Snævar (talk) 17:01, 11 February 2026 (UTC)

When you ran your SQL query, did you run it on the originally restored pages to account for the Linter errors that have already been fixed and the categories that needed to be removed? – Jonesey95 (talk) 17:37, 11 February 2026 (UTC)

@Anomie - who can say, really? People are interested in anything and everything. Even the most seemingly mundane detail in an archive may be exactly what some future historian is looking for as part of a research project. You could really say that about almost all of our archives, which we generate at a ferocious rate - page revision histories, system logs, talk archives. 18 years is also not very long at all. The discussion we're having right now will get archived, and then nobody might care about it at all for 25, 50, 100 years. But there may be a single historian in 2126 who it's useful to and is reading it right now. (Hello! Do you live in space? I'm sorry for what we did to the planet.) It's that possibility that we keep archives for.  — Hex • talk 20:30, 11 February 2026 (UTC)

Re redlinked categories, I've told JJMC89 bot III (the bot that normally processes WP:Categories for discussion outcomes) to empty Category:Temporary Wikipedian userpages, so that problem shouldn't happen anymore. And I'll add (to counter Jonesey95's comment that no one has cared in 18 years), that I approve of what Hex is doing, and have expressed dissatisfaction with this trend since at least 2019 (for example Wikipedia:Requests_for_undeletion/Archive_344#c-Pppery-2019-11-07T21:55:00.000Z-User_talk_pages:_A). Admittedly that scattergun REFUND request wasn't my finest hour and I'm inclined to agree now with some of the comments criticizing that request, but the claim that nobody has cared isn't true and if Hex wants to spent their time doing this then I would say more power to them. * Pppery * _{it has begun...} 18:26, 11 February 2026 (UTC)

Oh, thanks for the bot-herding, and your comments!  — Hex • talk 21:02, 11 February 2026 (UTC)

Section of text shows up orange but only in some cases

It is the section starting with "include obviously. It is absurd to say that we should say "he had never been arrested before" in [27]. See the discussion there about this. Thanks. Doug Weller talk 18:40, 10 February 2026 (UTC)

You are using one of the scripts that check links for reliability (Headbomb's I believe). It highlights the entire list item in which the unreliable link appears. I skimmed it so I can't say which specific link. Izno (talk) 19:28, 10 February 2026 (UTC)

Thanks. . Doug Weller talk 19:54, 10 February 2026 (UTC)

Side box + floatright combination squishes text on mobile

Hello there :) Apologies if this has been discussed everywhere or is otherwise known, but I noticed an ugly visual bug resulting from the combination of a {{side box}} and (a table with) the floatright class. You can see the effect at Ejective consonant, opening the mobile version from a narrow enough screen (or emulator - the "iPhone SE" preset in chrome devtools is perfect). I tried fiddling with it for a bit but didn't find a convincing solution, or one in which I'm sufficiently confident (e.g., would it make sense to add a content-based width to {{side box}}?). I'm also not familiar with the available layout templates and classes here on enwiki, so y'all may already have a simple solution that I'm not aware of. Daimona Eaytoy (Talk) 22:04, 10 February 2026 (UTC)

This should be changed in the floatright class definition. Memory says this class (and its friends) used to be wrapped in a media query on mobile such that it only took effect above a certain width. I have been meaning to make that how it works globally and just have been ~lazy~. cc @Jdlrobson Izno (talk) 00:51, 11 February 2026 (UTC)

Those rules exist but they only work on responsive skins (not resized skins). They work fine on mobile devices and people using desktop site on mobile.

Generally people shout at you when you give any kind of impression you are making their favorite pre-2011 skin "responsive" or mobile like which is why we unfortunately intentionally dont have a response version of the Vector 2022 (or Vector classic) skin whicha makes me sad.

https://gerrit.wikimedia.org/g/mediawiki/core/+/257e5d8c0a298d31870727cba144dc6c46ecec5f/resources/src/mediawiki.skinning/content.thumbnails-common.less 🐸 Jdlrobson (talk) 02:21, 11 February 2026 (UTC)

@Jdlrobson Ok, 1) I'm not crazy, and 2) the styles there aren't applying in Minerva right now? https://en.wikipedia.org/wiki/Ejective_consonant?useformat=mobile only has the clear-right-float-right rule, not the clear-both-float-none rule, at the appropriate width. Izno (talk) 02:42, 11 February 2026 (UTC)

Because Minerva isn't using that version, it has its own. [28] Izno (talk) 02:45, 11 February 2026 (UTC)

Phab:T368469 🐸 Jdlrobson (talk) 02:56, 11 February 2026 (UTC)

Thanks for the context :) I agree that the floatright class is ultimately responsible, although I guess I was also wondering if there's a simpler fix to apply to either of the involved templates while we wait for the proper fix. --Daimona Eaytoy (Talk) 12:01, 11 February 2026 (UTC)

"we unfortunately intentionally dont have a response version of the Vector 2022" - are you sure about that? sapphaline (talk) 12:05, 11 February 2026 (UTC)

Nesting templates

1. Is there are way for a template to read the parameters of a template nested inside it? For example, if I had {{template one| {{template two |para1=some |para2=thing}} }} is there some way to code "template one" to read the parameters from "template two"? Or can "template one" only ever read the output of "template two", not the inputs?
2. I've seen source code for some templates use #invoke where they could use an existing template. Is there a benefit to doing this, particularly for sidebars? What's the rationale for invoking a Lua module rather than just using {{sidebar}}?

I don't have a specific goal in mind with these questions, I'm just trying to understand how this works a bit better. – Scyrme (talk) 05:06, 11 February 2026 (UTC)

@Scyrme:

1. A template cannot do this. Via a module it can read the source text of the whole page and search this source text for strings like a specific template name, but we only do that in special cases like Module:Auto date formatter. It doesn't sound suitable for your purpose. It also relies on the parameter being present in the source text.

2.It reduces the post-expand include size to invoke a module directly instead of via a template. See Template:Navbox#Technical details.

PrimeHunter (talk) 05:29, 11 February 2026 (UTC)

It should not be possible to make generic section headers in Village Pump discussions

Occasionally editors will start a discussion on one of the village pump subpages and create a subheader with a title like "Discussion" or "Survey", which inevitably creates navigation problems when there end up being multiple identical subheaders on the page under different discussions. It should just not be technically possible to create a generic subheader on these pages. If someone tries to create one, they should be prevented from saving until they change it a unique subject-specifics subheader (like "Discussion (section headers)"). BD2412 T 19:00, 11 February 2026 (UTC)