35% of Post-ChatGPT Web Pages Show Signs of AI Authorship

More than one-third of English-language web pages published since ChatGPT became publicly available show signs of AI authorship or substantial AI editing, according to a new Pew Research Centre analysis.

The findings offer another measure of how quickly generative AI has become part of online publishing. They also closely match a separate research project that estimated roughly 35% of newly published websites were AI-generated or AI-assisted by mid-2025.

AI authorship rises sharply on newer web pages

Pew analysed about 490,000 English-language pages collected through the Common Crawl web archive. The dataset included snapshots from January 2021 through July 2026, providing researchers with material from both before and after ChatGPT’s 2022 launch.

The pages were analysed using Open Pangram, a detection model designed to identify statistical patterns associated with AI-generated or heavily AI-edited text.

Across a random sample of 10,000 pages collected in July 2026, 10% showed significant signs of AI authorship. That figure includes older pages created before modern generative AI tools became widely available.

When researchers limited the sample to pages published after ChatGPT’s release, the share climbed to more than 35%.

Commercial websites show more AI signals

The study found major differences between types of domains. In 2026 samples, around one in 10 pages on .com domains showed signs of AI authorship. The rate was about 4.6% for .org pages and roughly 1% for both .edu and .gov domains.

Researchers also tracked writing patterns that have become more common online. Compared with 2023, the use of em dashes roughly doubled, Oxford commas increased 63%, and several words frequently associated with AI-generated writing more than doubled in frequency. A comparison structure sometimes described as negative parallelism also became more common.

Those features alone do not prove that AI wrote a page. Pew stressed that AI detectors can produce false positives and false negatives, meaning individual pages can be incorrectly classified. The researchers said large datasets are more useful for identifying broader trends than for making definitive judgments about a single article.

Automation is changing both sides of the web

The rise of AI-generated material comes as automated systems are also responsible for a growing share of internet activity. Cloudflare has reported that automated traffic has surpassed human traffic in some measurements, with bots accounting for a majority of web requests. The shift was also detailed in a recent report on the growth of bot traffic.

Together, the findings illustrate how automation is increasingly shaping both the creation and consumption of online information. AI tools are now producing a substantial share of new web content, while automated crawlers, agents, and other bots are becoming an increasingly important part of the audience that accesses it.