Headline: One-Third of Post‑ChatGPT Web Pages Show AI Signatures — and .com Sites Are Driving the Surge, Pew Finds The web has quietly been refashioned by AI. A new Pew Research Center analysis of roughly 490,000 pages from the Common Crawl archive (January 2021–July 2026) finds that about 1 in 10 English‑language pages now show “significant signs of AI authorship.” Narrow the window to content published after ChatGPT’s November 2022 debut, and that share jumps to roughly 35%. How they measured it - Pew ran the text through Open Pangram, an AI‑detection model from Pangram Labs. The detector doesn’t flag single words but looks at statistical patterns across large blocks of text. - The researchers caution that the model can misclassify individual pages, and “significant signs of AI authorship” doesn’t mean a page was entirely machine‑generated—much could be AI‑assisted. What’s changing stylistically Pew tracked changes in stylistic patterns that have become more common since 2023, including: - More em dashes (about twice as common), - A 63% rise in Oxford comma usage, - A more than twofold increase in AI‑favored words like “delve,” “interplay,” and “testament,” - A near tripling of “negative parallelism” constructions (e.g., “it’s not just X, it’s Y”), though still rare overall. Where the AI content is concentrated The AI footprint is uneven across the web: - .com domains show AI signals at roughly 10 times the rate of .edu and .gov sites (each near ~1%), - .org sites sit around 4.6%, - In 2021 the four domain types looked much the same, but .com climbed steeply—Pew reports the .com AI‑authorship rate rose from ~1% in Jan 2021 to about 9.35% by Jan 2026. Why the domain gap exists Pew attributes the divergence to who’s publishing and how fast. Academic and government pages undergo editorial review, institutional sign‑offs, and slower publication cycles. .com includes everything from legitimate newsrooms to high‑velocity affiliate marketing farms producing large volumes of content—workflows that lend themselves to heavy AI use. Broader context and next steps Pangram’s detector has also appeared in other research, including a study finding AI‑generated text in roughly 9% of U.S. newspaper articles this year (including opinion pages). The limits of current detection matter, but detection could get more reliable in other ways: large AI companies (Anthropic and others) are exploring model‑level text fingerprints that would make machine‑generated text easier to identify with fewer errors. What this means for crypto and Web3 For crypto audiences, the trend has several practical implications: - News, analysis, and token marketing can be inflated by fast AI content production—raising risks of misinformation, washed‑out quality, and manipulation of investor sentiment. - On‑chain oracles, reputation systems, and decentralized curation protocols that rely on off‑chain text or media could be exposed to AI‑generated noise unless provenance and verification are strengthened. - Conversely, improved detection or model‑level fingerprints could enable better labeling or metadata standards for AI‑assisted content—an opportunity for builders focused on content authenticity in Web3. Bottom line: AI assistance and generation are now a pervasive part of the post‑ChatGPT web, especially on commercial domains. That shift matters not just for media quality, but for any crypto infrastructure or community that depends on trustworthy off‑chain information. Read more AI-generated news on: undefined/news