5% of New Wikipedia Articles Had AI Text, Study Found

A Princeton University study found that roughly 5 percent of nearly 3,000 articles created on English Wikipedia in August 2024 contained a substantial amount of AI-generated content, according to WIRED's October 13, 2026, account of the encyclopedia's fight with large language models. That single-month snapshot is the clearest public figure on how fast machine-written text moved into a project built on human sourcing and citation.

The number matters because of what came next. By March 2026, per WIRED, the Wikimedia Foundation banned editors from using large language models to write article text outright, restricting AI tools to translation and minor proofreading. The gap between the August 2024 study and the March 2026 ban — roughly 19 months — is the window in which Wikipedia's editing community decided containment had failed and prohibition was the only workable rule.

How much AI-written content has appeared on Wikipedia?

The Princeton study's 5 percent figure, drawn from a single month of nearly 3,000 new English-language articles in August 2024, is the only hard count in the record WIRED cites. Editors flagged a cluster of tells that pointed to machine authorship: AI models' habitual em dashes, curly quotation marks, and in some cases literal prompt residue left in the text, such as the phrase "Here is your Wikipedia article." Early AI contributions also produced a specific kind of filler — WIRED describes language "praising a village for its fertile farmlands, livestock, and scenic countryside" — alongside citations to sources that did not exist.

Why did Wikipedia ban large language models from writing articles?

Wikipedia's initial approach was not prohibition but correction. Editors tried to make AI output "acceptable and constructive," per WIRED, fixing or removing it when it wasn't. That changed by March 2026, when the Wikimedia Foundation concluded AI content was violating the project's core principles too often to manage case by case. The resulting policy confines large language models to translation and minor proofreading, cutting off the drafting role entirely.

By the Numbers

• 5 percent: share of nearly 3,000 new English Wikipedia articles found to contain substantial AI-generated content, per a Princeton University study of August 2024 data, as reported by WIRED on October 13, 2026.
• March 2026: when the Wikimedia Foundation banned large language models from writing article text, limiting their use to translation and minor proofreading, per WIRED.
• 2018: the year WIRED reports Google began training AI systems to produce "Wikipedia-like articles," predating the current AI Overviews traffic shift by roughly six years.

What did Wikipedia's leadership warn about AI in 2019?

Five years before the Princeton count, then-Wikimedia Foundation head Katherine Maher warned in a 2019 New York Times op-ed, cited by WIRED, that "[t]oo often, artificial intelligence is presented as an all-powerful solution to our problems, a scalable replacement for people." Her headline, "Without Humans, A.I. Can Wreak Havoc," framed the risk as the same one now confronting other knowledge institutions: retaining human control over systems trained to imitate human output.

Could Wikipedia's human editor pool shrink because of AI?

The more lasting worry inside the Wikipedia community, according to WIRED, isn't a one-month content percentage — it's a feedback loop. A widely read essay titled "Death of Wikipedia" argued that as AI content spreads, human contributors scale back, and the resulting gaps invite still more AI to fill them, compounding the problem each cycle. One Wikipedian quoted by WIRED put the stakes in blunt terms: "The potential 'whittling down' each year of our pool of talented writers is the greatest threat to Wikipedia, as we must increase the number of editors, especially expert editors, to be able to fix the sprawling hellscape of weak, inaccurate and incomplete articles that drag down the project."

"I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes."

That sentiment, quoted by WIRED from a widely shared post, captures the inversion Wikipedians say they're living through: automation arriving in the creative layer instead of the housekeeping layer it was expected to relieve.

What is Google's role in the AI-Wikipedia loop?

WIRED traces the loop's origin to 2018, when Google was already training AI to generate "Wikipedia-like articles." Nearly eight years later, that same company's AI Overviews are now cited, per WIRED's summary of the broader piece, as a factor in declining Wikipedia site traffic — meaning the search engine that once trained on Wikipedia's text is now positioned to intercept readers before they reach the source. The full WIRED story, adapted from the book "The Last Best Place on the Internet," lays out how that circularity — AI trained on Wikipedia, then AI content flowing back into Wikipedia, then AI summaries diverting readers away from Wikipedia — is reshaping the encyclopedia's position in the information ecosystem it helped build.