A Pew Research Center released a study on Thursday finding that over 35% of English‑language web pages published after the release of ChatGPT show indications they were written or substantially edited by AI. The report aligns with other research suggesting a growing share of newer web content is produced or heavily modified by automated systems.
Data collection and methodology
Pew used the Common Crawl web archive to collect nearly half a million English‑language pages from roughly the past five years, with the sample extending a couple of years before ChatGPT’s November 2022 launch. The organization then applied Open Pangram’s detection technology to estimate which pages were likely AI‑authored or significantly AI‑edited.
In a random sample of 10,000 web pages gathered in July 2026, about 10% showed “significant signs of AI authorship,” Pew reported. The researchers note that a purely random sample necessarily includes older pages published before generative AI tools existed, which cannot be AI‑authored.
Results for content published after ChatGPT
When the analysis was restricted to pages published after ChatGPT’s release, the share of pages exhibiting signs of AI authorship rose to more than one‑third (35%). This suggests AI involvement is substantially higher among newer web publications.
Differences by domain
Pew also examined results by top‑level domain. URLs with a .com domain showed signs of AI authorship at roughly ten times the rate observed for .edu or .gov domains (both near 1%). .org domains had a 4.6% rate of AI authorship.
Detection limits and stylistic indicators
The report acknowledges limitations: Pangram and other AI detection tools can misclassify human‑written pages as AI‑generated. Still, Pew argues that at scale the directional pattern is likely accurate. The study also found that several stylistic cues commonly associated with AI writing have become more frequent over time, including the use of em dashes, the Oxford comma, and constructions such as “it’s not X, it’s Y.”
Context and implications
Pew’s findings complement recent network‑level observations: for example, Cloudflare recently reported that bot web traffic has overtaken human web traffic, a milestone that occurred earlier than the company expected. Whereas Cloudflare’s analysis focuses on who or what is browsing the web, Pew’s work examines what is being browsed — and indicates that a substantial portion of newly published content may originate from or be shaped by AI, with implications for the quality and provenance of online information.
