Wikipedia’s traffic is falling, AI part of the reason, data does not yet show dying
The excerpt argues that AI search is draining Wikipedia of readers, that the encyclopedia is in a worse position than its defenders expected, and that the deeper loss is cultural, the end of an Enlightenment ideal of one shared body of knowledge. The reporting underneath is mostly accurate. The conclusions go beyond what the numbers show.
The usage figures check out, with one important caveat on where they come from. The Pew Research Centre reports that Wikipedia articles were viewed 1.9 trillion times over the past decade, about 508 million views a day, with 49% in English. The excerpt’s claim of a decline of almost a quarter in readership, over 1.1 billion visits a month since 2022, comes from Christophe Henner’s anniversary essay, which cites Similarweb data showing a 23% fall between 2022 and 2025. That is a third-party estimate. The Wikimedia Foundation’s own number is smaller: it reported an 8% year-on-year fall in human page views last October, after correcting for bot traffic. The two figures cover different periods and methods, so “a quarter” and “8%” are not in conflict, but readers should not treat the larger one as an official count. Pew also notes that bots, crawlers and AI agents produced more than 88 billion views in 2025 alone, which means raw traffic is a messy measure.
The excerpt’s description of the AI ban is correct on substance and slightly off on timing. English Wikipedia’s editors voted on March 20, 2026 to prohibit using large language models to generate or rewrite article content, replacing an older rule that only barred writing articles from scratch. The vote is reported as 44 to 2 by most outlets and 40 to 2 by a few, so I would not rely on the exact tally. The exceptions are basic copyediting of text an editor wrote and translation under existing guidance, as the excerpt says. The rule applies only to the English edition, and Spanish Wikipedia has a stricter total ban. The vote followed a suspected AI agent called TomWikiAssist that wrote and edited articles in early March with no human oversight. Detecting AI text remains imprecise, so enforcement depends on human moderators, who can miss cases on lightly watched pages.
Henner’s essay deserves a closer look, because the excerpt quotes him in a way that complicates its own thesis. The essay was published in January as Wikipedia turned 25, and it argued that the readership decline was linked to a shrinking pool of editors, with new sign-ups down about 36% since 2016. He called the AI-summary explanation “convenient” and said he was skeptical of it, preferring a broader shift in how younger people consume information, with AI as an accelerant. That is a more cautious claim than the headline “AI is killing Wikipedia.” The click-through figures he cites, 0.33% from AI chatbots against about 9% from Google search, are his, and I could not verify them independently. One commentary also reports that Henner co-wrote the essay with the AI chatbot Claude, and the Signpost chart credits Claude, which is a notable irony given the subject.
The excerpt also leaves out two pieces of context that cut against the “parasitic” framing. First, Wikipedia’s anniversary campaign was built around the slogan “Knowledge is Human.” Second, the Wikimedia Foundation’s commercial arm, Wikimedia Enterprise, announced licensing deals in January with Amazon, Meta, Microsoft, Mistral AI and Perplexity, letting those companies’ AI models reuse Wikipedia content. So the Foundation is at once banning AI inside the encyclopedia and selling the encyclopedia to AI firms. Whether that is pragmatic funding or a quiet acceptance of the dynamic Henner and Jemielniak warn about is the real strategic question, and the excerpt does not address it.
Several other claims I did not verify. The Princeton finding that about 5% of new English articles in August 2024 contained substantial AI content, Katherine Maher’s 2019 New York Times op-ed, and the quoted passages from WIRED and Nature are plausible but outside what my searches covered. A summary of later data says mobile page views fell 19% in a year by April 2026, but that comes from an AI-generated summary on a wiki page, so I would treat it as unconfirmed.
The analytical point is that the excerpt blends three different claims. The first, that traffic is falling, is measured and partly disputed in size. The second, that AI is the cause, is plausible but is contested even by Henner. The third, that the encyclopedic age is ending, is a cultural argument that no dataset can confirm. A reader can accept the first, doubt the second and take the third as one writer’s thesis. The most useful evidence to watch is not page views but editor numbers and article quality, since the essay itself concedes that content quality and article counts are up.
Sources: the WIRED excerpt from Richard Cooke’s The Last Best Place on the Internet, Pew Research Center, The Signpost (Henner essay), Meta-Wiki (Henner essay), Marketing Brew, Storyboard18, Computing, Beutler Ink, Implicator, MediaNama, SiliconANGLE, Dataconomy, PCWorld and GIGAZINE.
