More than a third of web pages published since OpenAI ChatGPT show signs of being written by artificial intelligence, study finds pew research center.
Image source: pewresearch.org
Cloudflare previously reported that web traffic from bots exceeded traffic generated by humans – something that happened sooner than the company expected. The Pew Research Center study is not about who or what is browsing the web, but about the material presented on those pages. Apparently, bots now regularly read content written by other bots. In conducting its analysis, Pew Research Center used the Common Crawl web archive, which obtained nearly 500,000 web pages published over the past five years, starting in the years before ChatGPT was released in November 2022. Next, Open Pangram’s technology helps us understand how many of these may have been written by artificial intelligence or heavily edited.
In a random sample of 10,000 web pages saved in July 2026, about 10% showed “significant evidence of artificial intelligence authorship.” But such random samples inevitably include old web pages created before the advent of artificial intelligence writing tools. So the Pew Research Center filtered out older content, leaving only content published after ChatGPT was released — and more than a third of the pages in the updated sample (35%) found evidence of AI authorship. .com sites showed signs of AI authorship about ten times more often than .edu or .gov sites, about 1%; on domains in the .org zone, AI was detected in 4.6% of cases.
Study results may not be perfect. Like other similar tools, Open Pangram can show false positives and show signs of AI authorship, even if the text was written by people with good intentions. But in large samples, the statistics appear to be generally true.
If you find an error, select it with your mouse and press CTRL+ENTER.