A new Pew Research Center analysis suggests that AI-generated or AI-assisted writing is becoming a significant part of the modern web. Researchers examined nearly 500,000 English-language webpages from the Common Crawl archive and found strong signals of AI authorship on about 10% of a random sample of pages in July 2026. Among pages published after ChatGPT's November 2022 launch, the share was considerably higher—more than one-third.
The numbers vary sharply by type of website. Commercial .com domains showed AI authorship signals on roughly 10% of pages, compared with about 4.6% of .org pages and around 1% of .edu and .gov pages. This suggests that businesses may be adopting generative AI particularly quickly for producing large volumes of search-oriented content, product information and other commercial web material where writing can be produced cheaply and at scale.
Pew's methodology is also important because AI detectors are imperfect at judging individual documents. The researchers used Open Pangram to identify statistical patterns associated with machine-written text across hundreds of thousands of pages. The study therefore does not mean that every page classified as showing AI authorship was necessarily written entirely by AI; rather, the patterns become more meaningful when examined across a very large dataset. The analysis also identifies stylistic changes associated with AI writing, including increased use of em dashes, Oxford commas and words such as “delve” and “pivotal.”
The larger implication is that AI is changing the composition of the web itself. As generating articles becomes cheaper, the amount of machine-assisted content can grow dramatically, potentially creating a feedback loop in which AI-generated material becomes part of the information environment used by future AI systems. Yet more AI-written content does not necessarily mean more visibility: separate research cited by TNW suggests that human-written material can still dominate search results and AI citations. The emerging challenge is therefore not simply detecting AI authorship, but preserving quality, originality, provenance and trust as machine-generated writing becomes a normal part of the internet.
