The Pew Research Center just put a number on something anyone who reads the internet has probably sensed: a lot of what gets published these days wasn’t typed by human hands. Analyzing nearly half a million English-language pages from the Common Crawl archive, Pew found that roughly 10 percent of all pages showed clear signs of AI authorship. But filter for pages published after ChatGPT launched in late 2022, and the picture shifts dramatically — more than a third show machine fingerprints.
The commercial web is driving the trend. About one in ten .com pages carries detectable AI text, compared to 4.6 percent for .org domains and a hair over 1 percent for .edu and .gov sites. That ten-to-one gap between commercial and educational or government pages says something about who’s rushing to automate content production. Pew used the Open Pangram detection tool for the analysis, which also surfaced some stylistic tells. Em dashes now appear twice as often as they did in 2023. Oxford comma usage jumped 63 percent. And AI-favorite vocabulary — “delve,” “pivotal,” “tapestry,” “meticulous” — has more than doubled in frequency. The “it’s not just X, it’s Y” construction nearly tripled.
Those linguistic patterns do more than confirm the obvious. They suggest that even text written partly by humans is absorbing the cadence of machine prose, which tracks with how people actually use these tools. A separate April 2026 study from Imperial College London, the Internet Archive, and Stanford reached a similar conclusion, putting the share of fully or partly AI-generated new websites at roughly 35 percent.
The uncomfortable caveat — and Pew acknowledges this — is that detection tools can’t reliably distinguish between fully automated text and writing where a human drafted something and an AI polished it. I’ve tested these detectors on my own work enough to know they misfire regularly. Yet public debate keeps treating AI involvement as binary, with Anthropic’s planned Claude watermark recently reigniting the fight. The reality is messier: adoption isn’t slowing, and the spectrum of human-machine collaboration is getting harder to define, not easier. That fuzziness is exactly where the next wave of disputes about authorship, authenticity, and credibility will land.
