Nassim Nicholas Taleb recently put it bluntly: Wikipedia is a victim of AI. He's right, and the damage is coming from multiple directions at once.
The most visible threat is traffic loss. When ChatGPT, Perplexity, or Google's AI Overviews answer a factual question, they're drawing heavily on Wikipedia content - but the user never visits Wikipedia. Wikimedia Foundation data showed a noticeable dip in pageviews starting in late 2023, coinciding with the mass adoption of AI chat interfaces. Fewer visits means fewer donation prompts, fewer people discovering the "edit" button, and fewer new volunteers joining a community that was already shrinking before AI entered the picture.
Then there's the content quality problem. Wikipedia editors have been fighting a rising tide of AI-generated submissions - articles and edits that read fluently but contain subtle factual errors, hallucinated citations, or synthesized claims that never appeared in any source. Detecting this stuff is labor-intensive, and the people doing that labor are unpaid volunteers who now have to work harder to maintain the same quality bar.
The Feedback Loop Nobody Talks About
Here's the part that should concern anyone who uses AI tools daily: these models were trained on Wikipedia. If Wikipedia's quality degrades because AI drained its contributor base, the next generation of models trains on worse data. It's a slow-motion collapse where AI tools gradually poison their own well.
For the average person using ChatGPT to look something up, this feels abstract. But if you rely on AI for research, content creation, or fact-checking, the accuracy of your outputs is downstream of Wikipedia's health. The encyclopedia isn't just a website - it's infrastructure. And right now, nobody is paying for the maintenance.