Wikipedia's Editors Declare War on AI-Generated Slop, Banning LLM Hallucinations

The editors of Wikipedia have officially drawn a line in the digital sand, issuing a hard ban on using large language models to write or rewrite articles. This new policy is a direct response to growing fears that AI-generated text is basically a factory for content that violates Wikipedia's sacred principles of verifiability and reliable sourcing—principles that, unlike some tokens, actually have real backing.

The freshly minted guideline puts it bluntly: "Text generated by large language models often violates several of Wikipedia's core content policies. For this reason, the use of LLMs to generate or rewrite article content is prohibited." There's a tiny carve-out, allowing AI to suggest basic grammar fixes to an editor's own work, but only if the bot doesn't start inventing new "facts." Editors are told to review these suggestions with the same skepticism they'd apply to a random crypto influencer's trading advice.

While the policy doesn't lay out specific penalties, getting caught repeatedly misusing AI qualifies as a "pattern of disruptive editing" under existing rules. This could lead to a block or ban faster than a rug pull, though accounts can theoretically appeal—a process likely more rigorous than most project revivals.

Linguistics professor Emily M. Bender acknowledged that some uses, like spell-checking, are reasonable. The real trouble starts when systems graduate from fixing commas to fabricating content. She points out that LLMs lack the accountability human contributors bring, operating with the ethical transparency of an anonymous dev wallet.

"Using large language models to produce synthetic text, it is a fundamental property of these systems that there is no accountability, no connection to what someone believes or stands behind," Bender said. She warned that widespread AI edits could degrade the site's value and reputation, turning the encyclopedia into a confidence game of citations.

Joseph Reagle, a professor who studies Wikipedia's culture, said the community's move echoes its long-standing obsession with accuracy and sourcing. "Wikipedia is wary of AI generated prose. They take the accurate characterizations of what reliable sources state about a topic seriously. AI has had serious limitations on that front, such as 'hallucinated' claims and fabricated sources," he noted, describing the kind of creative writing that would make even a meme coin whitepaper blush.

Reagle also highlighted the irony that many LLMs were trained on Wikipedia's own content. He pointed to a simmering resentment among Wikipedians about services that scrape community content only to generate a "glut of AI slop" that the community then has to clean up—a classic case of doing the hard work for others to profit.

Despite the broad prohibition, Wikipedia will allow AI tools to translate articles from other language editions into English, as long as editors verify the original text isn't pure fiction. The policy also advises editors not to rely on writing style alone to spot AI content, suggesting they instead check for policy compliance and editing history, because some humans write with the bland charm of a chatbot.

"Some editors may have similar writing styles to LLMs," the update dryly admits. "More evidence than just stylistic or linguistic signs is needed to justify sanctions." In other words, you can't ban someone just for being boring.

In a telling sign of the times, the Wikimedia Foundation reported in October that human visits to Wikipedia dropped about 8% year-over-year, as search engines and chatbots increasingly serve answers directly—the digital equivalent of front-running. Then, in January, the Foundation announced deals with AI giants like Microsoft, Google, Amazon, and Meta, letting them use Wikipedia material through its commercial Enterprise product. So, the data can be sold for training, but the resulting slop isn't welcome back home. The circle of web3 life, but for knowledge.