Pew finds over a third of webpages published since ChatGPT show signs of AI authorship

A third of the internet’s new pages carry the fingerprints of a machine writer. Pew Research Center published the measurement on August 20, 2026.

What was measured

Pew analyzed nearly 500,000 English-language webpages published between January 2021 and July 2026, running each through Pangram’s AI detection model.

Over one third of pages published after ChatGPT’s launch in November 2022 show signs of AI authorship.

The distribution by domain is uneven. In the July 2026 snapshot, about one in ten .com pages shows significant AI signs. That falls to 4.6 percent of .org pages and roughly 1 percent of .edu and .gov.

The style shift is visible

The prose itself changed in ways that can be counted. Since 2023, em dashes appear about twice as often. Oxford commas are up 63 percent. AI-favored words including “delve” and “interplay” have more than doubled. The “it is not X, it is Y” construction has nearly tripled.

Pew attaches its own caveat: detectors are imperfect, and the signal indicates that AI took part in the writing rather than proving a page is fully machine-made.

The loop underneath

Three systems draw from this same pool. Search results are ranked from it. Research and reference work is done on top of it. And the training data for the next generation of models is scraped out of it.

That last one is what makes the trend self-reinforcing rather than merely notable. Models trained largely on model output inherit the tics the detectors already measure, which is how a stylistic quirk turns into a durable feature of written English rather than a passing artifact of one product cycle.