What it is
Over the past year, volunteer editors on English Wikipedia kept running into the same problem: new articles that read fluently and looked properly formatted, but fell apart on fact-check — fabricated citations, invented dates, sources that don’t exist. Every one of them, it turned out, was AI-written. Last week, Wikipedia wrote that finding into its editing guidelines: editors may not use AI to write or rewrite articles, because AI-generated text systematically violates multiple core content policies. For an encyclopedia built on verifiability, fluent-but-unreliable text is more dangerous than clumsy human editing. The rule currently covers only the English edition, which has more than 6.9 million articles; other language editions haven’t followed yet.
Why it matters to you
Wikipedia matters for GEO because it carries outsized weight in mainstream LLM training data — getting an entry there means having a dedicated article inside the model’s “head” (see why Wikipedia inclusion is one of GEO’s strongest signals). This rule raises the bar for getting in: submit an AI draft as-is and it’s likely to get caught and reverted. You need genuinely verifiable secondary sources that hold up under a volunteer editor’s line-by-line check. And the failure mode Wikipedia caught — fluent but unsourced — isn’t unique to Wikipedia. It’s a risk baked into any AI-generated content, including pages on your own site that you let AI draft to save time.
Should you act now
If you’re preparing or have submitted a Wikipedia draft, confirm every claim in it is backed by a real, verifiable secondary source before you do anything else — don’t hand the whole thing to AI and paste in the output. The same check applies on your own site: for any AI-drafted content, have a person verify the citations and numbers actually exist before it goes live. That matters more than whether it reads smoothly.