Overview
- Pew Research published its analysis Thursday, August 20, 2026, and reported that about 35% of pages published since ChatGPT’s November 2022 release show meaningful signs of AI authorship or substantial AI editing.
- The study tested 490,000 English-language pages from 49 Common Crawl snapshots and counted any page with a Pangram score of 0.2 or higher as containing meaningful AI authorship.
- .com pages display the highest measured rates of AI-derived text at roughly 10%, compared with about 4.6% for .org and about 1% for .edu and .gov, showing uneven adoption across site types.
- Pew and independent peer-reviewed work led by Jonas Dolezal give consistent ~35% estimates but both teams warn that detection is probabilistic, Open Pangram can overcount older pages, and public crawl data underrepresents paywalled or login-protected content.
- Researchers say the rise in AI-typical punctuation and phrasing and infrastructure signals such as Cloudflare’s report on rising bot traffic underline risks to authenticity, spurring calls for verification tools like cryptographic provenance and algorithmic changes to surface human-created content.