Pew study finds AI authorship signals across 35% of newer webpages

Pew Research has found significant signs of AI authorship or substantial AI editing in more than one-third of English-language webpages published after ChatGPT launched in November 2022. In the study’s filtered snapshot, 35% of pages created after that release showed such signals.

The research examines the content of the web rather than the traffic moving across it. It follows Cloudflare’s report that bot traffic had surpassed human web traffic, but Pew’s analysis asks a different question: how much of the material being read online may have been produced or heavily revised by generative AI.

How Pew measured AI authorship

Pew assembled nearly half a million English-language webpages from the Common Crawl archive, covering roughly five years and beginning before ChatGPT’s release. It used technology from Open Pangram to identify pages likely to have been written or substantially edited by AI.

In a random set of 10,000 webpages collected in July 2026, about 10% displayed “significant signs of AI authorship.” Pew noted that this broad sample necessarily included older pages published before modern AI writing tools were available. Removing those older pages produced the higher 35% result for pages published after ChatGPT’s launch.

Differences across domain types

The rate varied markedly by top-level domain. URLs ending in .com showed signs of AI authorship at about 10 times the rate found on .edu and .gov domains, where both rates were around 1%. The figure for .org domains was 4.6%.

Pew also observed growth over time in patterns often associated with AI-generated prose, including em dashes, Oxford commas and constructions such as “it’s not X, it’s Y.” These patterns are supporting indicators rather than conclusive evidence about any individual webpage.

Limits of detection and the business implication

Pew cautioned that Pangram, like other AI-detection systems, can misclassify human-written material as AI-generated. The findings are therefore more useful as a large-scale directional view than as a definitive judgement on any single article or site. The broader context includes ChatGPT’s changing position in the AI market, where changes in the competitive AI landscape underline how quickly generative tools are becoming part of everyday digital production.

For businesses that use web research, publishing workflows or automated content tools, the practical implication is to retain meaningful human review, verify high-stakes claims against reliable evidence and treat AI-authorship signals as prompts for scrutiny rather than proof.