Pew Research assembled a corpus of nearly half a million English‑language webpages from the Common Crawl archive, covering roughly 2020 – mid‑2026. Each page was run through Open Pangram’s detection model to estimate whether the text had been generated or substantially edited by an AI system.
When the analysis was limited to pages published after ChatGPT’s November 2022 release, over one‑third (≈35 %) exhibited “significant signs of AI authorship.” A broader random sample of 10 000 pages from July 2026, which included older content, yielded a lower baseline of about 10 % AI‑likelihood.
- Sample size: ~500 k pages total; 10 k for the July 2026 snapshot
- Detection tool: Open Pangram AI‑authorship classifier
- Timeframe: pre‑ChatGPT (≈2020‑Oct 2022) vs. post‑ChatGPT (Nov 2022‑mid‑2026)
- AI‑authored share: ≈35 % of post‑ChatGPT pages; ≈10 % when older pages are included
Why this matters
The finding that a substantial fraction of newer web content bears AI‑generated markers suggests a shift in the web’s information ecosystem, where machine‑produced text increasingly populates the same surfaces that both human readers and automated crawlers consume. This trend may affect content provenance, downstream model training data quality, and the effectiveness of bot‑traffic mitigation strategies, as more pages become both sources and targets of automated agents.
Share this article
Found this insightful? Share it with your community on Reddit, X, or copy the link.