Pew Research Center (2026) — How much of the internet is written with AI?

Summary:

This Pew Research Center analysis looks at how much of the English-language web now appears to be written or heavily edited by AI. It asks a simple question: after ChatGPT and similar tools took off, how common did AI-written web content become?

The work was done by researchers at Pew Research Center, a nonpartisan research organization. The goal was to measure how AI-style writing has spread across the web over time and where it shows up most.

To do that, the team pulled almost 490,000 English-language webpages from the Common Crawl web archive, covering the past five years. They then ran the page text through Open Pangram, an AI detection model, to flag pages that were likely written or substantially edited by AI. They also looked at a random sample of 10,000 pages from July 2026 as a snapshot, and compared patterns across domain types like .com, .org, .edu and .gov.

The headline result is that 10% of pages in the July 2026 sample showed strong signs of AI authorship. The share has been rising since late 2022, around when ChatGPT was released.

When the researchers looked only at pages published after ChatGPT launched, the share was much higher: more than one-third of pages in the July 2026 snapshot showed signs of AI authorship.

AI-style writing was much more common on .com sites than elsewhere. In 2026 samples, about 9.4% of .com pages showed signs of AI authorship, compared with 4.6% on .org pages and about 1% on .edu and .gov pages. The report also says AI-linked writing habits have become more common online overall. Em dashes appeared about twice as often as in 2023. Oxford commas rose 63%. Certain words the model links to AI, like “delve,” “interplay” and “testament,” more than doubled. A comparison style like “it’s not just X, it’s Y” nearly tripled, though it was still uncommon.

The main catch is that this depends on an AI detection tool, and Pew openly says such tools are not perfect. The model can misclassify individual pages, so the findings are more reliable for big groups of pages than for any one page.

The analysis is also limited to English-language webpages found in Common Crawl, so it is not the whole internet.