Study estimates over a third of post-ChatGPT web pages show signs of AI authorship

A Pew Research analysis released Thursday estimates that over a third of web pages published after ChatGPT’s November 2022 release show signs of AI authorship or substantial editing. The estimate is based on a study of nearly half a million English-language web pages drawn from the Common Crawl web archive, spanning roughly five years of web history.

To detect AI-generated or heavily edited content, Pew used Open Pangram’s technology on a July 2026 snapshot of 10,000 pages. In that sample, about 10% showed significant signs of AI authorship. However, because the sample included older pages published before AI writing tools existed, Pew filtered the dataset to focus only on pages published after ChatGPT’s launch. In that filtered set, 35% of pages showed signs of AI authorship.

Domain-level analysis found that .com pages were far more likely to show AI authorship signals than .edu or .gov pages. The study reported AI authorship rates around 10 times higher in .com domains compared to .edu or .gov domains, which were each near 1%. .org domains had a 4.6% rate of AI authorship.

Pew acknowledged limitations in the detection methodology, noting that tools like Open Pangram can misclassify content as AI-generated when it is not. Despite this, the researchers argue that the scale of the analysis suggests the findings are directionally accurate. They also observed increases over time in stylistic markers often associated with AI writing, such as em dashes, Oxford commas, and repetitive phrasing patterns like 'it’s not X, it’s Y.'