AI Now Writes 35% of Web Content, Study Finds

Researchers just mapped how artificial intelligence is reshaping the internet, and the results challenge common fears about AI-generated content.

By mid-2025, roughly 35 percent of newly published websites contained AI-written text, according to a large-scale study from Imperial College London, Stanford University, and the Internet Archive. Before ChatGPT launched in late 2022, that number was essentially zero.

The research team analyzed thousands of websites across 33 months using the Internet Archive's Wayback Machine. They tested six popular theories about how AI might harm online content.

Only two concerns proved true. First, AI text is making the internet more uniform. AI-generated articles were 33 percent more similar to each other than human writing, suggesting the technology gravitates toward average ideas rather than diverse perspectives.

Second, the web is getting weirdly cheerful. AI content scored 107 percent higher on positive sentiment than human writing, reflecting the technology's tendency toward overly agreeable, sanitized language.

But four other worries didn't pan out. Individual writing styles haven't vanished. Websites aren't linking less to outside sources. Information density hasn't dropped. And the study found no increase in verifiably false statements, though researchers acknowledge this particular finding needs deeper investigation.

The team surveyed 853 Americans and discovered a surprising gap between perception and reality. Most respondents believed all the negative predictions about AI, including the four that didn't hold up. People who rarely use AI were more likely to believe it causes harm than regular users.

The Bright Side

Stanford AI researcher Jonas Dolezal sees an opportunity in these findings. Instead of making AI models perfectly compliant and relentlessly positive, he suggests giving them more personality and creative friction. That could help AI serve as a genuine creative partner rather than replacing human voices entirely.

The research team isn't stopping here. They're working with the Internet Archive to create an ongoing monitoring tool that tracks AI content over time, turning their one-time study into a continuous window on how technology is changing online communication.

The real risk isn't necessarily false information flooding the web. It's that overly cheerful, uniform content might push diverse human perspectives to the margins or make people stop trusting online information altogether.

The findings suggest we need smarter solutions than just detecting AI text after it's published, including better standards for marking content origins and search algorithms that reward different ideas over sameness.