🌍Pew: 35% of Post-ChatGPT Web Pages Show AI Authorship
TL;DR
Pew analyzed nearly 500,000 English-language pages from Common Crawl and found AI authorship signals in 35% of those published after ChatGPT. Commercial domains are roughly 10x more affected than .edu and .gov.
Pew analyzed nearly 500,000 English-language pages from Common Crawl and found AI authorship signals in 35% of those published after ChatGPT. Commercial domains are roughly 10x more affected than .edu and .gov.
Key Points
A random 10,000-page July 2026 sample showed ~10% AI signals before filtering out pre-2022 pages
.edu and .gov sat near 1% AI authorship; .org came in at 4.6%
Detection used Pangram, which Pew concedes can misclassify human text
Em dashes, Oxford commas and 'it's not X, it's Y' phrasing all rose over the period
Lands shortly after Cloudflare reported bot traffic overtaking human traffic
Why It Matters
A third of new pages carrying model fingerprints is the clearest number yet on training-data contamination, and it sets a floor on how much of the next crawl is model output.
Quick Facts
Frequently Asked Questions
Why does this matter?
A third of new pages carrying model fingerprints is the clearest number yet on training-data contamination, and it sets a floor on how much of the next crawl is model output.
What happened?
Pew analyzed nearly 500,000 English-language pages from Common Crawl and found AI authorship signals in 35% of those published after ChatGPT. Commercial domains are roughly 10x more affected than .edu and .gov.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,294 builders reading daily.