Pew study: more than one in three new web pages shows signs of AI authorship
Almost 490,000 English-language pages were analysed. Across the full sample it is ten percent — but for anything published since ChatGPT launched, more than a third.
Illustration · AI-generated (AI IN LIFE)
At a glance
- Data: close to 490,000 English-language pages from Common Crawl, January 2021 to July 2026
- 10 percent of all analysed pages show clear AI authorship signals (as of July 2026)
- More than one third among pages published after ChatGPT's November 2022 launch
- By domain: .com around 10 percent, .org 4.6 percent, .edu and .gov around 1 percent each
- Detection via the Open Pangram model; Pew explicitly warns against conclusions about single texts
The Pew Research Center analysed close to 490,000 English-language web pages from the Common Crawl archive, published between January 2021 and July 2026. The finding: as of July 2026, ten percent of all pages examined show clear signs of AI authorship. Restrict the analysis to pages that went online after ChatGPT's release in November 2022, and it rises above one third.
The distribution is anything but even. Commercial .com addresses sit at around ten percent — roughly ten times the rate of .edu and .gov pages, which both stay near one percent. The .org space falls in between at 4.6 percent. Where reach translates directly into money, machine writing arrived fastest.
For detection, the researchers used Open Pangram, a model that looks at linguistic patterns. Some of the signals are strikingly concrete: since 2023, em dashes appear about twice as often, the Oxford comma is up 63 percent, AI-typical vocabulary has more than doubled, and negative parallelism — phrasing along the lines of "not X, but Y" — has nearly tripled.
Pew states the key limitation itself: detection models are not perfect and can misclassify individual documents. The claim holds for large bodies of text, not for any single page. Treating this study as a tool to call "AI" on one specific article stretches it past what it can carry.
The second caveat matters just as much. The web is a mix of old and new material. Ten percent looks low precisely because it counts millions of pages created long before any language model existed. The number that shows the direction of travel is the other one — more than a third of everything added since.
FAQ
Does this mean a third of the internet is written by AI?
No. The figure applies to pages published after ChatGPT launched and measures signs of AI involvement, not necessarily fully machine-generated text. Across the whole sample it is ten percent.
How does the model detect AI involvement?
Through linguistic patterns. Pew cites a doubling of em dashes since 2023, 63 percent more Oxford commas, more than doubled AI-typical vocabulary, and nearly tripled negative parallelism.
Can I use this to check whether a specific text is AI-written?
Not reliably. The authors stress that detection models can misclassify individual documents and that the signal only becomes meaningful across large collections of text.
