More than one-third of English-language webpages published since ChatGPT launched show significant signs of artificial intelligence involvement.
The finding comes from a new analysis by Pew Research Center. The study examined how generative AI is changing the way people write and publish content online.
Pew analysed around 490,000 English-language webpages collected through Common Crawl. The webpages covered the period from January 2021 through July 2026.
Researchers used the Open Pangram AI-detection model to examine the content. The analysis focused on signs suggesting that AI had written or substantially edited the text.
AI Signals Rise Sharply After ChatGPT
The figures look different when researchers consider the entire sample. Across webpages captured in July 2026, about 10% showed significant signs of AI authorship. This figure included older webpages published before modern generative AI became widespread.
However, the picture changed considerably when Pew examined pages published after ChatGPT launched in November 2022.
More than one-third of those webpages showed significant AI signals in the July 2026 snapshot. Therefore, the findings point to a notable rise in AI involvement across online publishing.
Commercial Websites Lead AI Writing Trend
AI-generated and AI-edited content does not appear evenly across different types of websites.
Pew found that roughly one in 10 webpages on .com domains showed signs of AI authorship in its 2026 samples.
The figure was considerably lower among other domain types. About 4.6% of .org webpages showed AI signals.
Meanwhile, approximately 1% of webpages on both .edu and .gov domains showed similar signs.
The findings suggest that commercial websites have adopted AI-assisted writing more extensively than several other types of online platforms.
AI Writing Is Also Changing Online Language
Pew’s analysis found another noticeable change across online writing.
Several writing patterns commonly associated with AI models have become more frequent since 2023. For example, the use of em dashes has roughly doubled.
Similarly, the use of Oxford commas has increased by 63%.
Certain words also became considerably more common. These included “delve,” “interplay,” and “testament,” which more than doubled in frequency.
Another writing pattern has also become more common. Known as negative parallelism, it includes constructions such as “it’s not just X, it’s Y.”
Pew found that this pattern has nearly tripled. However, researchers noted that it remains relatively uncommon overall.
AI Detection Still Has Limitations
The findings do not mean that every webpage containing these patterns was written by AI. Pew cautioned that individual writing habits cannot reliably establish whether humans or AI produced specific content.
People naturally use em dashes, Oxford commas and certain vocabulary. Likewise, AI detection systems can incorrectly classify both human-written and machine-generated material.
However, large-scale analysis can reveal broader changes across online writing.
As generative AI becomes more widely used, the study suggests that its influence is increasingly visible in how the internet sounds and reads.
