Pew Research finds AI authorship on a third of new webpages
A new analysis of nearly half a million English-language webpages suggests that 35% of content published since 2022 shows significant signs of being written or edited by artificial intelligence.
A new study from Pew Research indicates that over one-third of English-language webpages published after the launch of ChatGPT in November 2022 exhibit significant signs of being written or substantially edited by artificial intelligence. The report, released on Thursday, provides a detailed look at the changing nature of online content, focusing on what is being browsed rather than who or what is doing the browsing.
To compile the findings, Pew Research analysed nearly half a million webpages from the Common Crawl archive, covering the five years leading up to the present. The study utilised Open Pangram’s detection technology to identify pages likely authored or heavily edited by AI. In a random sample of 10,000 webpages collected in July 2026, approximately 10% initially showed signs of AI authorship. However, after filtering out older pages published before the advent of AI writing tools, the rate of AI authorship among post-ChatGPT content rose to 35%.
The data reveals a stark disparity in AI usage across different domain types. URLs with a .com domain showed signs of AI authorship at approximately 10 times the rate of .edu or .gov domains, both of which recorded rates of around 1%. Meanwhile, .org domains had a lower rate of 4.6%. This suggests that commercial entities are adopting AI for content creation at a significantly higher pace than educational or government institutions.
The study also noted an increase in specific linguistic markers associated with AI writing, including the frequent use of em dashes, Oxford commas, and phrasing structures such as “it’s not X, it’s Y.” While these patterns are not definitive proof, they contribute to the broader picture of AI’s growing influence on digital text.
Pew Research acknowledges that the analysis is not without limitations. Like other AI detection tools, Open Pangram can misclassify pages as AI-written when they were not. Nevertheless, the firm suggests that at scale, the data is likely directionally correct. The findings corroborate other research detailing the increasing role of AI in web content creation.
The report arrives shortly after internet infrastructure provider Cloudflare reported that bot web traffic had overtaken human web traffic. This milestone was reached sooner than Cloudflare had estimated, reinforcing the notion that much of the web is now bots reading webpages written by other bots.
