Pew Study: 35% of Web Pages Since ChatGPT Show AI Authorship

More than a third of web pages published since ChatGPT's debut carry the fingerprints of AI writing, according to a Pew Research Center study released on August 20, 2026. The organization analyzed nearly half a million English-language pages spanning five years, starting before OpenAI's chatbot launched in November 2022.

Pew put the figure at 35% for content produced after that release date. The detection relied on Open Pangram's authorship-classification technology, which flags patterns strongly associated with generated or heavily edited text.

Those patterns include stylistic tics that have become shorthand for machine-written prose: frequent em dashes, Oxford commas, and rhetorical constructions like "it's not X, it's Y." Pew found these markers grew more common with each passing year since ChatGPT's release, suggesting AI involvement in web publishing has climbed rather than plateaued.

A snapshot from a fresher sample tells a somewhat different story. Among 10,000 pages collected in July 2026, only about 10% showed strong signs of AI authorship, a smaller share than the five-year average, though still a meaningful chunk of new content.

Where the AI writing concentrates

The study also broke results down by domain type, and the gap is stark. Commercial .com sites showed AI-authorship signals at roughly ten times the rate of .edu or .gov domains, both of which sat near 1%.

Nonprofit .org sites landed in between, at 4.6%. The disparity points to where AI-assisted content creation has taken hold fastest: high-volume commercial publishing, rather than academic or government sources bound by stricter editorial or institutional review.

The findings land alongside a separate data point from Cloudflare, which has reported that bot traffic on the web has now overtaken traffic generated by humans, a milestone that arrived faster than the company had previously projected. Taken together, the two findings describe a web increasingly shaped by non-human actors on both the production and consumption sides.

Pew's methodology does not claim to detect every instance of AI use, since lighter editing assistance may leave fewer traceable markers than fully generated text. But the scale of the sample, drawn from nearly 500,000 pages, gives the 35% figure more weight than smaller spot-check studies that have circulated since ChatGPT's launch.