Skip to content

ProductNot yet confirmed elsewhere1 publisher2 min readPublished

Wikipedia editors who expected AI to do the housekeeping banned it from writing by March 2026

Princeton University's study found about 5% of nearly 3,000 new English Wikipedia articles from August 2024 contained substantial AI-generated text. Wikipedia had banned writing with large language models by March 2026, a Wired book excerpt reports.

The Product Desk

How we use AISend a correction

Illustration accompanying Wikipedia editors who expected AI to do the housekeeping banned it from writing by March 2026
Generated illustration
AI text appeared in new articles before the ban Share of new English Wikipedia articles in August 2024 with substantial AI content, per a Princeton study, and time to the March 2026 ban.

About 5% of new English Wikipedia articles from August 2024 contained a substantial amount of AI-generated content, per a Princeton study. That is roughly 150 of nearly 3,000 articles. The ban came about 19 months after the sampled month.

AI text appeared in new articles before the ban
FigureValueNoteClaim
New articles with substantial AI content, Aug 20245% about2
Articles in that 5%150 roughlyout of nearly 3,000 sampled10
Sampled month to ban19 months, about11

What happened

  • Editors had predicted that language models might improve Wikipedia by automating formatting, sources, citations and missing links, leaving humans free for creative work.
  • Signs of AI text included ChatGPT's fondness for em dashes and curly quotation marks.
  • Wikipedians first tried to shape AI output, weeding out generic fluff about village scenery and fake sources.
  • The ban confines the models to translation and minor proofreading.
  • Google's AI Overviews became as central to the search experience as Wikipedia was until recently, with Wikipedia now a source one step removed.

Why it matters

  • decision A team that opens a shared knowledge base to AI drafting has to set the line between checking tasks and composing tasks before volume arrives. Wikipedia tried managing the output first and drew the line later.
  • cost Review labour lands on volunteers, the same people the 'Death of Wikipedia' essay expected to contribute less as AI spread through the platform.
  • exposure Errors that enter Wikipedia can reach readers who never open it, because AI Overviews draw on it one step removed, and the excerpt's author asks how anyone would prove those answers wrong.
  • contradiction The Foundation says Wikipedians' care is something AI cannot replace, and the excerpt's author points out that this leaves open whether it gets replaced anyway.

Sometimes the residue of a prompt stayed in the text, and a page could open with a line such as "Here is your Wikipedia article" [3]. That is what users actually did. What the community told itself, by Wired's account, was that AI would come and go [12], and that in the meantime it would take over chores [1].

The Princeton figure needs a denominator. Five percent of nearly 3,000 new articles is roughly 150 pages [10]. That is a small count, drawn from one month on the English edition, and the excerpt does not report a later measurement. The ban came 19 months after that month [11].

The punctuation tells were weak evidence once a writer knew to look for them, since em dashes and curly quotation marks are quick to delete [4]. The harm the editors named sat in the content: write-ups praising a village's scenic countryside, and fake sources [5].

Katherine Maher, then head of the Wikimedia Foundation, had argued in a 2019 New York Times op-ed that artificial intelligence was too often presented as an all-powerful solution to problems and a scalable replacement for people [7]. The "Death of Wikipedia" essay described the version editors feared most: models composing the articles while humans were left with formatting and cleanup [9].

In our view, the line Wikipedia drew is a test any operator can run on their own tools. Translation and minor proofreading start from text a person already wrote [6]. Composition starts from nothing, so each claim in the output needs a source that a reviewer has to go and find.

Sort each AI-assisted task on two attributes: whether the output contains claims nobody supplied, and whether the pool of people able to check it is growing or shrinking. A task that adds no claims and keeps a steady or growing set of reviewers can go ahead. A task that adds claims while reviewers leave is where Wikipedia ended up with a ban. The two mixed cases are candidates for a pilot with a stated sampling rate. The cost of the line is whatever drafting speed a model would have supplied, and Wikipedia chose to pay it.

What to watch

  • Whether a measurement later than the August 2024 sample appears for the share of new English articles with substantial AI text.
  • Whether the translation and minor proofreading exception to the March 2026 ban holds, or editors narrow it further.
  • Whether contributor numbers fall as the 'Death of Wikipedia' essay predicted.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence45
Adoption30
Hype gap+20
Incentives
Insufficient
Confidence45
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    Wikipedians predicted that large language models might even improve Wikipedia by automating more of the housekeeping (formatting, sources, citations, missing links), freeing humans to do creative work.

    ReportedSupportedSource: Wired book excerpt, 'Wikipedians Believed AI Would Come and Go. They Were Wrong'View cited source
  2. [2]

    A Princeton University study in October 2024 found that about 5 percent of nearly 3,000 newly created articles on English Wikipedia in August 2024 contained a substantial amount of AI-generated content.

    ReportedSupportedSource: Princeton University study, as reported by WiredView cited source
  3. [3]

    Residues from the prompts used to compose articles were sometimes left behind in the text, for example 'Here is your Wikipedia article'.

    ReportedSupportedSource: Wired book excerptView cited source

Sources

1 independent publisher whose own reporting we read for this story.

  1. wired.com

    1 article · October 13, 2026

    Wikipedians Believed AI Would Come and Go. They Were Wrong

Share your take

Let Clarity write the post for you.

Signed-in readers get a short post drafted on this story in the register they choose — narrative, analytical, or a direct position — editable to the last word before it goes anywhere. The share buttons at the top of this story work without an account.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Loading related stories