A new analysis from Pew Research, released Thursday, reveals that more than a third of webpages published since ChatGPT's debut show strong indications of AI authorship. This finding aligns with earlier research highlighting how much of the internet's newer content is now either written by or heavily shaped by artificial intelligence. The study comes on the heels of a recent report from internet infrastructure firm Cloudflare, which found that bot traffic had surpassed human traffic online—a threshold reached sooner than the company had predicted. Pew's investigation, however, isn't concerned with who or what is visiting websites, but rather with the content itself. And it turns out, a significant portion of that content appears to be bots reading pages composed by other bots.

To conduct the study, Pew relied on the Common Crawl web archive, gathering nearly 500,000 English-language webpages spanning the past five years or so, with the sample starting a couple of years before ChatGPT launched in November 2022. The researchers then applied Open Pangram's detection technology to assess how many of these pages were likely written or substantially edited by AI. In a random selection of 10,000 pages collected in July 2026, roughly 10% displayed "significant signs of AI authorship," according to Pew. Still, the organization noted that such a random sample would naturally include older pages—those published before AI writing tools existed and therefore couldn't have been AI-generated.

To get a clearer picture of how much of the web is currently being produced by AI, Pew filtered out the older pages and concentrated solely on those published after ChatGPT's release. In this refined snapshot, signs of AI authorship appeared in over a third—35%—of the pages.

Looking deeper into specific domains, Pew discovered that URLs ending in .com showed signs of AI authorship at a rate roughly ten times higher than .edu or .gov domains, both of which hovered around 1% AI-authored. Meanwhile, .org domains showed a 4.6% rate of AI involvement. Of course, this analysis isn't flawless. Pangram, like other AI detection tools, can sometimes misidentify human-written content as AI-generated. But on a large scale, the findings are likely at least directionally accurate. Pew also observed that other supposed markers of AI writing have grown more common over time, including the use of em dashes, Oxford commas, and phrases like "it's not X, it's Y," among other stylistic patterns.