The web’s newest pages are increasingly being written—or heavily reworked—by machines, but the evidence comes with a crucial caveat: the tools used to identify AI writing are still imperfect.

The shift accelerated after ChatGPT’s November 22, 2022 launch. Reviewing nearly half a million English-language pages gathered through the Common Crawl archive, Pew Research found that more than one-third of pages published after that date were likely created or substantially edited by AI. One account of the findings put the figure at 35%, a striking measure of how quickly generative tools have moved from novelty to routine web-production machinery.¹

The broader snapshot looks less dramatic at first. In a random sample of 10,000 pages collected in July 2026, roughly 10% showed significant signs of AI authorship. But that pool included older pages published before mainstream AI writing tools existed. Once Pew filtered those out and focused on the post-ChatGPT era, the share climbed sharply.

The pattern also varies by domain. Commercial .com pages showed apparent AI authorship at about 10 times the rate of .edu and .gov sites, each near 1%; .org pages stood at 4.6%. The result supports the sense that the open commercial web is becoming an increasingly automated marketplace of text.

Still, the study does not claim a final verdict on any individual page. Pew relied on Open Pangram’s detection technology, and detectors can wrongly label human-written material as machine-made. The more defensible conclusion is the broader trend: newer pages are showing far more of the signals associated with AI writing. As another account of the research noted, the open web is now “rife with AI-generated text,” even as the systems policing that boundary remain fallible.²