A recent study by the Pew Research Center shows that AI-generated content on the English internet is rapidly increasing. The research team sampled approximately 490,000 English web pages from the Common Crawl archive, covering January 2021 to July 2026, and analyzed them using the AI detection model Open Pangram.
The share rose to 35% after ChatGPT.
The results show that in web samples from July 2026, approximately 10% exhibited clear signs of AI-generated content. When only considering pages published after ChatGPT's launch in November 2022, this proportion rose to 35%.
Differences between domains are increasing
Research shows that AI writing traces are not evenly distributed. Pages with .com domains show related indicators at a rate of approximately 10%, while .edu and .gov domains each hover near 1%, and .org domains are around 4.6%. Pew notes that in 2021, the performance of these domain types was roughly similar.
- Approximately 10% of the .com page shows signs of AI-generated content.
- .edu and .gov are both close to 1%.
- The .org page is approximately 4.6%.
Identification is based on text patterns.
This study does not determine whether content is AI-generated by analyzing individual words, but rather by identifying statistical patterns across large volumes of text. The research notes that the use of dashes has approximately doubled, Oxford commas have increased by 63%, and certain words commonly found in AI-generated text have become significantly more frequent.
In addition, phrases similar to “not X, but Y” have increased nearly threefold since before 2023, though they remain uncommon overall. Pew also cautions that detection models may still produce false positives, so “clear signs of AI writing” does not mean the entire content is fully generated by a machine.
Business websites load faster.
Research suggests that this difference is related to the publishing process. University and government websites typically require longer review cycles, while commercial websites covering news, marketing, and content farms are more likely to adopt AI-assisted writing at scale.
From a temporal perspective, the proportion of AI writing indicators on .com pages has risen from approximately 1% in January 2021 to 9.35% in January 2026. The report also notes that companies such as Anthropic are advancing model-level text fingerprinting technologies to improve the accuracy of AI-generated text detection.