A 500,000-page Pew Research study from Common Crawl measures the footprint of language models on the recent web — and the huge gap between the .com and .edu domains.
On August 20, 2026, Pew Research published a measure of the extent of automated writing on the web. The organization drew from the Common Crawl archive about 500,000 pages in English covering five years, starting two years before the release of ChatGPT in November 2022, and then submitted these texts to Open Pangram's detector to identify content that was likely written or edited by a model.
Out of a random sample of 10,000 pages drawn in July 2026, 10% show marked signs of AI writing. But if we only look at pages put online after the arrival of ChatGPT, the proportion rises to 35%. The breakdown by domain extension is just as telling: sites in .com have a rate about ten times higher than that of .edu and .gov, which remain around 1%, while .org is at 4.6%.
The authors point out the inherent limitation of the exercise: a detector can make a mistake on a given page. On the scale of hundreds of thousands of documents, on the other hand, the order of magnitude is probably reliable — and it draws a web in which a growing proportion no longer has a human author in the full sense.
Read the full article on the original source
Go to the source →