The Pew Research Center released a new study on Thursday: since ChatGPT’s launch, more than a third of new webpages show signs of AI-generated writing.

The finding comes from an analysis of the Common Crawl web archive. Pew collected nearly 500,000 English-language webpages over the past five years or so, starting a few years before ChatGPT’s release in November 2022. They then used Open Pangram’s detection technology to determine which pages were likely written or heavily edited by AI.

Among a random sample of 10,000 webpages from July 2026, about 10% showed “clear signs of AI writing.”

But that random sample was mixed with plenty of older pages — pages published before AI writing tools existed, which could not have been written by AI.


If you filter out older pages and only count pages published after ChatGPT’s launch, the share jumps to 35%.


Breaking it down by domain is even more interesting. Webpages on .com domains show signs of AI writing at a rate about 10 times higher than .edu and .gov — both of which sit at only about 1%. .org domains are slightly higher, at 4.6%.

In other words, commercial websites are the most eager to use AI to produce content. Academic and educational institutions haven’t done so yet.

This study comes shortly after Cloudflare reported that bot traffic had already surpassed human traffic — earlier than they had predicted.

Pew’s data focuses not on who is browsing the web, but on what webpages themselves are. The result: a large portion consists of bots reading pages written by other bots.

Of course, this kind of detection is imperfect. Like other AI detection tools, Pangram may misclassify human-written pages as AI-written. But at scale, the direction is probably correct.

Pew also noted that certain so-called AI writing traits have increased year over year: em dashes, Oxford commas, and constructions like “not X, but Y.”

Reading this, you might be wondering whether this article was written by AI. Don’t worry — this one was written by a human. At least for now.