Pew Study Finds AI Writing Traces on About a Third of Webpages Since ChatGPT

Pew Research released a study analyzing roughly 500,000 English webpages from the Common Crawl archive, finding that about one third of webpages show signs of AI-written or substantially AI-edited content since ChatGPT's launch — one of the most representative quantifications yet of generative AI's penetration into the web's content ecosystem.

Confirmed

• The study scanned nearly 500,000 English webpages from Common Crawl with AI-detection tools; roughly one third show AI writing or substantial AI editing since ChatGPT's launch (@rohanpaulai, @emmanuelvivier, @TuhinChakr); @luisdans cites an alternate framing of about 10% of samples showing "clear signs."
• Distribution is highly uneven: by 2026, 9.4% of .com commercial pages show significant AI editing or creation versus 1% for .edu and 0.8% for .gov; @TuhinChakr notes growth is driven almost entirely by .com sites.
• Style shifts include em-dash usage roughly doubling versus 2023, Oxford comma use up 63%, more "Not X, but Y" constructions, and surges in AI-favored words like "delve," "interplay," and "testament."

Unconfirmed

• AI detectors carry error rates and are unreliable for individual pages, a limitation @luisdans highlights; the one-third versus 10% figures reflect differing statistical framings — the original report is authoritative.

Why it matters

• @luisdans argues the data lends quantitative support to the "dead internet theory," showing generative AI is now a major force in web text production.
• The .com versus .edu/.gov gap suggests AI content concentration tracks commercial incentives, as @rohanpaulai notes commercial publishers benefit most from rapid drafting.
• Detectable stylistic fingerprints offer practical linguistic clues for AI-content identification and provenance research.

2026-08-21 ~ 2026-08-22 · 12 related posts

• [source] Pew Study: 1 in 3 Web Pages Show AI Writing Signs Since ChatGPT Launch — rohanpaul_ai · 2026-08-21
• Pew: AI Writing Found in 9.4% of .com Sites vs 1% of .edu — rohanpaul_ai · 2026-08-21
• AI's Stylistic Footprint: Dashes and 'Delves' Surge Across the Web — rohanpaul_ai · 2026-08-21
• Pew Research: AI content growth driven almost entirely by commercial websites — TuhinChakr · 2026-08-22
• Rohan AI Newsletter: 1 in 3 Web Pages Written by AI — rohanpaul_ai · 2026-08-22
• AI Weekly: 33% of Web Pages AI-Written, Zhipu Calls Trillion-Param Models a Detour — rohanpaul_ai · 2026-08-22
• [source] Pew study finds 10% of web samples show signs of AI authorship — luisdans · 2026-08-22
• Pew Research: How Much of the Internet Is Written With AI? — Justgototheeffinmoon · 2026-08-22

4 near-duplicate retellings: rohanpaul_ai · rohanpaul_ai · rohanpaul_ai · emmanuelvivier