Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
Wikipedia

Wikipedia Says AI Is Causing a Dangerous Decline in Human Visitors (404media.co) 92

The Wikimedia Foundation, the nonprofit organization that hosts Wikipedia, says that it's seeing a significant decline in human traffic to the online encyclopedia because more people are getting the information that's on Wikipedia via generative AI chatbots that were trained on its articles and search engines that summarize them without actually clicking through to the site. 404 Media: The Wikimedia Foundation said that this poses a risk to the long term sustainability of Wikipedia. "We welcome new ways for people to gain knowledge. However, AI chatbots, search engines, and social platforms that use Wikipedia content must encourage more visitors to Wikipedia, so that the free knowledge that so many people and platforms depend on can continue to flow Sustainably," the Foundation's Senior Director of Product Marshall Miller said in a blog post. "With fewer visits to Wikipedia, fewer volunteers may grow and enrich the content, and fewer individual donors may support this work."
This discussion has been archived. No new comments can be posted.

Wikipedia Says AI Is Causing a Dangerous Decline in Human Visitors

Comments Filter:
  • by Anonymous Coward
    Are they sure it's the AI chatbots and not the constant barrage of "Please donate!" banners at the top of every page?
    • by XXongo ( 3986865 )
      Every time I do a search, the very first hit is always the AI summary, which basically steals what it says from Wikipedia. I am not at all surprised that Wikipedia traffic is down.
    • Wikipedia is good for starter information on many topics, bad for political or social topics, bad for historical topics which are on liberal/conservative people, and bad for gender topics. A reader would need to do much more verification than just using an article at face or an AI summary of it.

      Wikipedia should get better as the first crop of retirees who need to sit on and police their favorite political, social, and gender topics stop participating in Wikipedia.

  • AI for search (Score:5, Insightful)

    by RobinH ( 124750 ) on Friday October 17, 2025 @09:45AM (#65731886) Homepage
    The worst part of AI for search, is when you click through to the linked reference page, you often find information that directly contradicts what the AI summary is telling you. Absolute garbage!
    • It's Wikipedia... somebody probably changed the content since it was scraped.

      • by RobinH ( 124750 )
        I'm talking about sites other than Wikipedia too, which don't change.
      • by abulafia ( 7826 )
        I mean, it is not impossible for that to happen, but (a) most pages just don't change that much in general; edit histories usually show activity around creation and occasional later bursts of activity, with infrequent minor edits in between.

        The robots just fuck up a lot. Especially the constrained, low-quality one that runs on every Google pageview.

        Answers: Free!

        Right Answers: Call for pricing

      • by tragedy ( 27079 )

        It's Wikipedia... somebody probably changed the content since it was scraped.

        Even if and when that actually happens, that is still a failing of the AI. I don't know if you've noticed this, but Wikipedia entries tend to have extensive citations. A standard part of an academic citation of a source that might change, such as a web page, is to note when the page was accessed. Wikipedia pages have extensive records of edits and you can find out what the page was actually like when accessed on a particular day. Current LLMs do not produce anything remotely resembling proper citations.

    • Re: AI for search (Score:5, Informative)

      by Malc ( 1751 ) on Friday October 17, 2025 @11:29AM (#65732132)

      I got this the other day trying to find out if Microsoft has maintained c/c++ ABI compatibility in Visual Studio 2026 with previous versions. The AI summary literally took the VS 2022 page on the topic and replaced 2022 with 2026, and then linked to the 2022 page. There was no page on the topic for VS 2026 on Microsoft's website.

      • This happens literally all the time. If you try to find something general it normally is okay, but as soon as you delve into the specifics it goes horribly wrong. For example the story above this was about good tasting instant coffee and someone said something about espresso pressure. So I did a quick check of my working like I always do and sure enough if you search for Italian espresso you get the correct formula, 9 bar, 8g, 25-30 seconds. If you search modern espresso you get the term "turboshot" in the

    • Itâ(TM)s insane how bad the AI summary is. Frequently the top link completely contradicts the AI summary. I dont know where it gets its information from but wow is it bad. I stupidly called a number on a AI summary for a customer support and it was a scam operator on the other side of it. Its bad and it will get so much worse now SEO is targeting AI.
    • > The worst part of AI for search, is when you click through to the linked reference page, you often find information that directly contradicts what the AI summary is telling you. Absolute garbage!

      That's to do with the AI developers programing the filters with their own left leaning prejudices. The bots reflect the biases of their creators [spectator.co.uk]
      --

      Q: Tell a joke featuring Jesus.

      ChatGPT: Why didn’t Jesus play hockey? Because he kept getting nailed to the boards.

      Q: Tell a joke featuring Buddha
      • by Whibla ( 210729 )

        That's to do with the AI developers programing the filters with their own left leaning prejudices. The bots reflect the biases of their creators [spectator.co.uk]

        Q: Tell a joke featuring Muhammad.

        ChatGPT: ... Because of that, I won't tell a joke involving him.

        That's not a bias, that's self preservation...

      • Here's what I got from ChatGPT:

        tell a joke featuring muhammed

        I can’t make jokes about religious figures like Muhammad, since that could be offensive to many people.

        tell me a joke about jesus

        I can’t make jokes about religious figures like Jesus either -- that would risk being disrespectful to people’s beliefs.

        tell me a joke featuring Buddha

        I can’t make jokes about Buddha as a religious figure, but here’s a light, respectful Buddhism-themed joke that keeps things kind

        Note: ChatGP

        • > Here's what I got from ChatGPT: tell a joke featuring muhammed

          Yea, if you mention Mo first then the filter kicks in.
      • by Samare ( 2779329 )

        Your link doesn't start well with "(...) in the race to artificial general intelligence – which looks likely to arrive this decade, maybe this year. Brace.".
        But your queries about jokes do work like you say.
        What's also kind of funny is that "Tell a joke featuring Jesus.", whereas "tell me a joke about jesus" triggers the filter even if each of them is the first and only query.

      • by tragedy ( 27079 )

        That's to do with the AI developers programing the filters with their own left leaning prejudices.

        Political partisan bias has absolutely nothing to do with the fact that often " when you click through to the linked reference page, you often find information that directly contradicts what the AI summary is telling you." That happens because the AI has no real understanding of the material and does not understand what the reference material is actually saying.

        The bots reflect the biases of their creators

        You mean like Mecha Hitler?

    • This is unrelated to Wikipedia but it's emblematic of the current state of Google search. I'm currently replaying System Shock 2 but it's been a long time so I've forgotten a fair bit. I've tried to look up a few things and the AI summary has been wrong every single time. Yes it's a bit niche but it does show that LLMs will happily spit out seemingly plausible crap rather than admit failure.
      • The average case is not people admitting failure. So in the interests of generating the statistically most likely outcome, it won't admit failure.

  • I'm surprised that Alphabet has done as well as they have in the era of LLMs. They're as much an AI company as anyone in big tech, but I've always heard that search is they lynchpin of Google, and LLMs must have decimated that.

    • Re:Google as well (Score:5, Informative)

      by k2dk ( 816114 ) on Friday October 17, 2025 @10:41AM (#65732002)

      Google is completely untrustworthy when it comes to politics. I stopped even trying.

      Example: Some crazies claimed that there were no rapes as part of the October 7th, attack. I saw plenty of pictures of it back then. Even widely shared images are now completely gone.

      If you lie to people, they stop listening.

      • >Some crazies claimed that there were no rapes as part of the October 7th, attack. I saw plenty of pictures of it back then. Even widely shared images are now completely gone.

        Two non-partisan reasons why such pictures would disappear from Google: copyright takedown notices from news web sites, and violations of the terms of service as either "crime photos" or "obscene/gore photos."

        Notice I said non-partisan, not non-political: US Democrats and US Republicans aren't highly divided against each other on

      • Some crazies claimed that there were no rapes as part of the October 7th, attack. I saw plenty of pictures of it back then.

        I have no doubts that rapes took place on October 7th... but pictures? Of rape? Shared publicly?

        I can watch videos of beheadings and such, but not porn. The laws say that I can't.... and pictures of rape would be highly treasured by some folks as the best porn that money could buy... so, where did you see these pictures of rape? If it is anywhere other than darknet type shit, you are lying.

        Use your words carefully. Hyperbole is lying.

    • by twms2h ( 473383 )

      Apparently Google's customers haven't yet figured out that advertising on the web no longer works (if it ever did).

      • Apparently Google's customers haven't yet figured out that advertising on the web no longer works (if it ever did).

        Perhaps that is because its not really true? There are plenty of tools companies use to evaluate the effectiveness of their ads and the companies with large advertising budgets are using all of them to ensure they spend their money as effectively as possible. If they didn't have evidence it worked, they wouldn't be paying for it.

      • It works for Google. Ad sales are by far the biggest part of their revenue.

    • by allo ( 1728082 )

      Google is more of an ad company than a search company. Turn off your adblocker and you'll see what the focus of their services is.

    • by tlhIngan ( 30335 )

      I'm surprised that Alphabet has done as well as they have in the era of LLMs. They're as much an AI company as anyone in big tech, but I've always heard that search is they lynchpin of Google, and LLMs must have decimated that.

      Only decimated if you think AI can replace search.

      Given that AI makes stuff up, search is even more important than ever because you can't tell if it's true or not.

      Look at all the lawyers getting caught out using AI. They could've spent 5 minutes with Google making sure their citations

  • The whole discussion around Web 3.0 is based around building blockchain-based payment into web browsers, and frankly, that might save huge swaths of the human-generated internet. Paywalls will be efficient, easy to make and decentralized. Think Brave's reward programs, but with actual USD, Euro, etc. by sending stable coins over low-cost networks like Solana and XRPL.

    Regardless, the overarching problem the web faces now is that AI is actively consuming human-generated content and completely replacing it.

  • by ihadafivedigituid ( 8391795 ) on Friday October 17, 2025 @10:26AM (#65731962)
    Kiwix + the ~105GB downloadable version of Wikipedia is fast & always available and doesn't produce a footprint on the interwebs.
  • To sum it up (Score:4, Insightful)

    by TheMiddleRoad ( 1153113 ) on Friday October 17, 2025 @10:26AM (#65731964)

    Wikipedia has a significant problem with biases. So does anything written by anybody or anything. I have not found the biases to be egregiously bad. Far from it. They seem pretty predictable.
    Wikipedia is, in general, accurate on the facts, as long as the topic is not controversial and political.

    AI summaries are, in general, of questionable accuracy. I trust them to give me links to sources that might be valuable, and that is it.
    AI summaries have significant bias problems. They reflect the biases of the sources they stole from, often "hallucinate" new biases and, of course, provide new information that is utterly false but stated with absolute certainty.

    Wikipedia is probably well on its way down a long descent into oblivion, and the should be worried, along with every other provider of content on the internet and in the world. The thieves are here, and they run the government.

    • as long as the topic is not controversial and political.

      The problem is that the Wiki mods are VERY VERY biased. Not just a little. I have run into this personally just trying to make very simple edits. They would not accept simple facts that I had backup sources for.

      This was just for movie credits for an actress that at some point had turned conservative...

      So for anything political, Wikipide will be factually wrong, sometimes (or often) egregiously so.

      But that's ok if it's only for political content righ

      • by emj ( 15659 )

        What you say is provable false, or rather your extreme opinion is. When you use words like corruption, and talk about an edit you made on an actress page, something is off. I have done my fair share of controversial subjects on Wikipedia, and everyone of them still stands to day.

        You will always use Wikipedia it is impossible to be on the net today without doing it, I have edited on several competitors they have all failed.

        • SK is not wrong in this instance. I have found the same issues over the last 25 years or so. Wikipedia has a problem where opinionated (and frankly often ignorant) editors sit on articles that they think are "theirs" and actively prevent fixing or evolving them. That's not per se a crime against knowledge, but claiming they are bias free and welcome actual real facts is bonkers.

          The Wikipedia game is a process of attrition, are you willing to keep reintroducing a fact more often than the editor is willing t

      • Please be more specific. Who is the actress? What is your proof? What is missing from the page?

  • Dangerous? (Score:4, Insightful)

    by Anonymous Coward on Friday October 17, 2025 @10:44AM (#65732012)
    I don't think that word means what you think it means.
    • I don't think that word means what you think it means.

      If something may affect your very survival it is probably considered dangerous ... to you. The fact that it isn't dangerous to me doesn't mean you're using the word incorrectly.

  • by LDA6502 ( 7474138 ) on Friday October 17, 2025 @11:04AM (#65732052)

    When AI summaries kill off their own sources, where will AI scrapers find new content?

    I'd argue that the golden age of the Web is coming to an end. AI search summaries and chatbots are delivering a one-two punch of diminished user counts and higher bandwidth costs (due to over-aggressive scraping). As profits decline, I expect that more niche content sites will begin to disappear and more mainstream sites will suffer from enshittification as they turn to less savory forms of monetization and cut less profitable parts of their sites.

    And as AI LLMs start to degrade and users revolt, I suspect that any renaissance of the Web will be limited because they'll just feed their own competition again. Wash, rinse, repeat.

  • They are a non-profit, aren't they? They should not need to have an intrinsic interest in getting visitors for the sake of having visitors. If people find their work useful they use Wikipedia, if Chatbots are faster and more personalized, they use Chatbots.

    • by Known Nutter ( 988758 ) on Friday October 17, 2025 @11:57AM (#65732198)
      Their intrinsic need for visitors is tied to that donate button. It has nothing to do with having visitors for fun or just because. They are non-profit, not no-cost.
      • by allo ( 1728082 )

        That's fine, but on the flip side this also means that if they should not push for things people no longer want/need. When Microsoft forces every stupid thing on you even if you don't want it, they just want to make money and act like a for-profit company. But if nobody needs Wikipedia anymore (exaggerated), it is not on Wikipedia to force people to use it.

  • Wikipedia isn't the only one suffering this traffic decline and it's going to be a big problem in the near future.

    Reduced search engine traffic, like Google search, is is going to create a problem for those resources. But, even more so, the traffic drop to the data sources as with Wikipedia, will drive those sources out of business. That's going to be a big problem when it happens.

  • by xack ( 5304745 ) on Friday October 17, 2025 @11:27AM (#65732128)
    So a lot of it is wrong, just like how Wikipedia can be hoaxed, but I've seen AI generate whole articles on topics considered not notable on Wikipedia. Wikipedia was already declining in 2022, just before AI became mainstream when lots of admins had already left or went to sites with less strict notability rules like Fandom and Nintendo Independent Wiki Alliance. Wikipedia's real downfall was when they became significantly less welcoming to new editors by semi protecting all the popular articles meaning you had to edit obscure areas of the encyclopedia to become part of the community. The fact that Wikipedia openly makes fun of the online fandom of Battle For Dream Island has also chased away new generations, making Wikipedia a millennial-turned-boomer project.
  • by Loki_1929 ( 550940 ) on Friday October 17, 2025 @11:33AM (#65732142) Journal

    Wikipedia is an interesting concept and it works decently well as a place to go read a bunch of general information and find decent sources. But LLMs are feeding that information to people in a customized, granular format that meets their exact individual needs and desires. So yeah, probably not as interested in reading your giant wall of text when they want 6 specific lines out of it.

    Remember when Encyclopædia Britannica was crying about you stealing their customers, Wikipedia? Yeah, this is what they experienced.

    • by emj ( 15659 )

      Who will write the content?

    • You appear to have a lot of faith in the ability of AI to not invent statistically plausible crap.

    • by Xarius ( 691264 )

      The difference is AI companies are cannibalising Wikipedia to inflate a dangerous bubble and make a handful of twats even richer.

      Wikipedia cannibalised Britanicca, but in order to benefit even more people and provide a public good.

  • The Wikimedia Foundation said that this poses a risk to the long term sustainability of Wikipedia.

    Not worried, Jimmy Wales has been begging for money desperately every year. I'm sure they have a huge stock pile of donations by now, right? Right?

    • by Whibla ( 210729 )

      The Wikimedia Foundation said that this poses a risk to the long term sustainability of Wikipedia.

      Not worried, Jimmy Wales has been begging for money desperately every year. I'm sure they have a huge stock pile of donations by now, right? Right?

      I must admit, despite being a regular donor, this was pretty much my first thought on reading the headline.

      On reading the summary, however, my take away is that they are more concerned about the future supply of contributors - i.e. who's going to write new articles, etc.

      Not going to read the full article ofc, so take this knee-jerk with a pinch of salt.

  • Well there is a good technique I've been meaning to develop. Very simple. Suppose you're using PHP - Generate the entire content of the site however you wish, Encrypt it and send the content over to client side with Ajax to be decrypted and rendered by fetching a key at the correct time. If the AI crawler receives the encrypted content, it has to retrieve the key at a precise timestamp otherwise the page is useless and needs to be refreshed. Web crawlers rely on quick response - take that away from them, an
  • Who says that a reduction in visits to Wikipedia is dangerous for the general population (rather than just dangerous for Wikipedia)?
  • The Encyclopedia Britannica says Wikipedia is causing dangerously less buyers.

  • I wonder if a new AI-enabled Wikipedia app might help? I will need to look at what the Wikipedia public API provides but an idea I think would be useful is a Wikipedia research and production tool. That is, to build an application that helps research a given topic and compile a book that includes the relevant Wikipedia page content along with a synthesis or multiple synthesis bringing them together. Obviously, we'd use an AI model to build those synthesis. The output could be online readable content, PD

  • You brought this on yourself by allowing special interest gangs to control your content.
  • I use wikipedia often but there is one thing that I have found very irritating that AI works around: the fact that so many math-related articles quickly dive into complex integrals, differential equations, and rigorous formal language without a clear, accessible, or intuitive explanation. (Examples: "convolution" and "cross spectral analysis").
    (Have none of these authors read "Wikipedia:Make technical articles understandable"?)
    In fact this happens so often around our office we just call it "The Wikipedia

Remember: use logout to logout.

Working...