More than a third of new web pages since ChatGPT launched appear to be AI-written

What the study looked at

Researchers at Pew Research wanted to answer a simple-sounding question: how much of the internet is now written by AI instead of humans? To find out, they pulled nearly half a million English-language web pages from a massive public archive of the internet called Common Crawl, covering roughly the last five years, and ran them through a detection tool built to recognize AI-generated or heavily AI-edited text.

What they found

When Pew looked at a broad sample of recent pages, only about 1 in 10 showed strong signs of AI authorship — but that figure also includes many older pages written before AI writing tools even existed. Once they narrowed the sample to only pages published after ChatGPT's launch in November 2022, the figure jumped dramatically: about 35%, or roughly one in three pages, showed genuine signs of AI-written or edited text.

Where it shows up most

Not all corners of the internet are equal here. Ordinary commercial websites (sites ending in ".com") had AI-authorship rates roughly ten times higher than educational (".edu") or government (".gov") sites, which stayed at barely 1%. Nonprofit sites (".org") fell in between, under 5%. That pattern makes sense — commercial content is often produced quickly and at scale to drive search traffic, while university and government sites are typically written more formally and reviewed by more people before publication.

How they actually recognized "AI writing"

Beyond the detection software itself, the researchers noted a number of telltale stylistic habits that show up more often in AI-generated text — such as heavier use of em dashes, more frequent use of a comma before "and" in lists, and repeated sentence patterns like "it's not X, it's Y."

Why this matters

This aligns with other recent findings — including data from Cloudflare suggesting that automated bot traffic has now overtaken human traffic on the web as a whole. Together, these studies paint a picture of an internet where an ever-larger share of both the content and the visitors is no longer human at all.