Pew Research finds a third of post-Nov 2022 English web pages were AI-written

Pew Research published on Thursday a figure illustrating how the web changed over two and a half years: more than a third of English-language web pages that went live after November 2022 contain signs of writing or editing by large language models. The analysis used the Common Crawl archive and covered close to half a million pages from the past five years.

The researchers ran the sample through Open Pangram, a detection tool that flags text that was written or substantially edited by artificial intelligence. In a random sample of 10,000 pages from July 2026, about 10% of pages showed clear AI signatures, but the sample necessarily included older pages from before the chatbot era. After removing pages published before November 2022, the rate jumped to 35%, more than one in three.

Splitting the results by domain suffix reveals an uneven picture: .com domains showed a detection rate ten times higher than .edu and .gov domains, which both stood at roughly 1%. .org domains recorded 4.6%. The data suggest that the commercial and open-source space absorbs most synthetic content, while academic and government institutions remain largely human-generated.

Pew notes that Open Pangram, like any detection tool, may mistakenly label human text as artificial, but at scale the trend is “at least directionally correct”. At the same time, increases were identified in the frequency of stylistic markers associated with models: long dashes, Oxford commas, and constructions of the form “it’s not X, it’s Y”. These markers have accumulated over the years alongside the rise of generative tools.

The report arrives shortly after Cloudflare reported that bot traffic has overtaken human traffic, a milestone that occurred earlier than forecasts. The practical implication is that a growing share of the web consists of bots reading pages written by other bots, while the human layer is shrinking to institutional and closed corners.