Stanford, Imperial College London, and the Internet Archive studied 33 months of web data, from August 2022 to May 2025. That's the period right after ChatGPT launched.
Before late 2022? Zero. By mid-2025? Around 35% of newly published websites flagged as AI-generated or AI-assisted.
Three years. That's all it took.
Here's what's interesting though. The researchers tested six common critiques about AI content, stuff like more lies, fewer sources, shrinking viewpoints. But only two of them actually turned out to be true.
➡️ AI is making the web less semantically diverse. Fewer distinct ideas, more of the same thing said in slightly different ways.
➡️ AI is making the web more positive and cheerful. Which sounds fine until you realize it's also more sanitized.
The disinformation angle? Didn't confirm it. Not because AI content is accurate, but because the internet wasn't great at truth to begin with.
One researcher put it well. He said the real risk isn't fake facts. It's unverifiable claims. Stuff fact-checking tools can't even test.
For anyone creating content online, the gap between human voice and AI voice is still real. And it's probably the only thing that holds attention long term.
Maybe I'm wrong.
But the web getting blander and more agreeable doesn't feel like a win for anyone trying to build something that actually gets read.
Source: 404 Media