Over on the official weblog of the Wikipedia group, Marshall Miller untangled a current thriller. “Round Could 2025, we started observing unusually excessive quantities of apparently human site visitors,” he wrote. Greater site visitors would usually be excellent news for a volunteer-sourced platform that aspires to achieve as many individuals as doable, however it could even be shocking: The rise of chatbots and the AI-ification of Google Search have left many massive web sites with fewer guests. Possibly Wikipedia, like Reddit, is an exception?
Nope! It was simply bots:
This [rise] led us to analyze and replace our bot detection methods. We then used the brand new logic to reclassify our site visitors knowledge for March–August 2025, and located that a lot of the unusually excessive site visitors for the interval of Could and June was coming from bots that had been constructed to evade detection … after making this revision, we’re seeing declines in human pageviews on Wikipedia over the previous few months, amounting to a lower of roughly 8% as in comparison with the identical months in 2024.
To be clearer about what this implies, these bots aren’t simply vaguely inauthentic customers or some incidental aspect impact of the overall spamminess of the web. In lots of circumstances, they’re bots engaged on behalf of AI corporations, going undercover as people to scrape Wikipedia for coaching or summarization. Miller bought proper to the purpose. “We welcome new methods for individuals to realize information,” he wrote. “Nevertheless, LLMs, AI chatbots, search engines like google, and social platforms that use Wikipedia content material should encourage extra guests to Wikipedia.” Fewer actual visits means fewer contributors and donors, and it’s simple to see how such a state of affairs might ship one of many nice experiments of the net right into a demise spiral.
Arguments like this are intuitive and simple to make, and also you’ll hear them past the ecosystem of the net: AI fashions ingest loads of materials, typically with out clear permission, after which supply it again to customers in a type that’s typically straight aggressive with the individuals or corporations that supplied it within the first place. Wikipedia’s authority right here is bolstered by the way it isn’t attempting to generate profits — it’s run by a basis, not a longtime business entity that feels threatened by a brand new one — but additionally by its distinctive place. It was based as a stand-alone reference useful resource earlier than settling ambivalently into a brand new position: A website that folks largely simply discovered via Google however in higher numbers than ever. With the rise of LLMs, Wikipedia grew to become essential in a brand new manner as a uniquely giant, numerous, well-curated knowledge set concerning the world; in return, AI platforms are actually successfully holding customers away from Wikipedia whilst they explicitly use and reference its supplies.
Right here’s an instance: Let’s say you’re studying this text and change into interested by Wikipedia itself — its early historical past, the wildly divergent opinions of its unique founders, its funding, and so on. Except you’ve been being attentive to these items for many years, it might really feel as if it’s all the time been there. Absolutely, there’s extra to it than that, proper? So that you ask Google, maybe as a shortcut for attending to a Wikipedia web page, and Google makes use of AI to generate a blurb that appears like this:
That is an AI Overview that summarizes, amongst different issues, Wikipedia. Formally, it’s fairly near an encyclopedia article. With a number of formatting variations — discover the bullet-point AI-ese — it hits loads of the identical factors as Wikipedia’s article about itself. It’s a bit shorter than the highest part of the official article and comprises far fewer particulars. It’s high-quality! However it’s a abstract of a abstract.
The subsequent choice you encounter nonetheless isn’t Wikipedia’s article — that reveals up additional down. It’s a immediate to “Dive deeper in AI Mode.” Should you do this, you see this:
It’s one other abstract, this time with a little bit of commentary. (Additionally: If Wikipedia is “usually not thought-about a dependable supply itself as a result of it’s a tertiary supply that synthesizes data from different locations,” then what does that make a chatbot?) There are hyperlinks within the type of footnotes, however as Miller’s put up suggests, individuals aren’t actually clicking them.
Google’s remedy of Wikipedia’s autobiography is about as pure an instance as you’ll see of AI corporations’ efficient relationship to the net (and possibly a lot of the world) round them as they construct unusual, sophisticated, however typically compelling merchandise and deploy them to tons of of hundreds of thousands of individuals. To those corporations, it’s a useful resource to be consumed, processed, after which changed into a product that makes an attempt to render all the pieces earlier than it’s out of date — or at the very least to bury it beneath a heaping pile of its personal output.
