Will it run?
Policy

Wikipedia bans LLM writing after study finds 5% of new articles contain machine-generated text

By Nadia Ksiazek Clawpit staff
Wikipedia bans LLM writing after study finds 5% of new articles contain machine-generated text

The project that built the world's largest encyclopedia on human labor has blocked the tool meant to lighten the load. In March 2026 the editor community voted to prohibit the use of large language models for writing articles, restricting them to translation and light copy-editing only. The decision followed a long period of attempts to tame the technology that ended in repeated failure.

Early optimism shattered quickly. As far back as 2018 Google trained models to produce "Wikipedia-style articles," and the Wikimedia Foundation expected the technology to take over housekeeping, formatting, sourcing and missing links — freeing humans for creative writing. Katherine Maher, then the foundation's chief executive, warned in a 2019 New York Times op-ed that AI is often framed as a magic bullet and a scalable substitute for people. The warning stayed on paper.

The study that exposed the scale of the problem came from Princeton University in October 2024. Researchers examined roughly 3,000 new articles created on English Wikipedia in August that year and found that about 5% contained a substantial amount of text generated by language models. The tells were unmistakable: em dashes, curly quotes, and prompt leftovers forgotten in the body of the article — "here is your Wikipedia article" and the like.

At first the community tried to shape the output, stripping generic praise for agricultural villages and deleting fabricated citations. It did not hold. By March 2026 it was clear that artificial content was violating core principles at a frequency that left no choice but an outright ban on writing with it.

A viral essay titled "The Death of Wikipedia" described the dynamic: as contributors pull back because of machine presence, the temptation grows to replace them with yet more models, a downward spiral. A viral post on X captured the absurdity: "I want AI to do laundry and dishes so I can do art and writing, not AI to do art and writing so I can do laundry and dishes." The likely result: editors trapped in an endless curation project whose source is not human priorities but a frozen harvest of machines.

"The shrinking pool of talented writers each year is the biggest threat to Wikipedia," one Wikipedian wrote, "because we have to grow the number of editors, especially expert editors, to fix the expanding hellscape of weak, inaccurate and incomplete articles that are dragging the project down." A Wikipedia written and read by machines threatens to produce a detached version, stripped of context and passion, artless and soulless.

Google changed the rules of the game when it launched AI Overviews in search. They became the center of the experience exactly as Wikipedia once was. The encyclopedia remains a source, but one step removed. AI agents serve up answers that sound plausible and intuitive, pass a superficial source check, and are often wrong. If Wikipedia itself becomes polluted, the problem cascades through the entire information chain.