A significant portion of English-language webpages now show signs of AI authorship, with a notable increase observed since the launch of ChatGPT in November 2022, according to a study by the Pew Research Center. The research indicates that approximately 10% of all English-language webpages exhibit significant indicators of being written or assisted by artificial intelligence.
This trend is particularly pronounced on commercial (.com) domains. Since ChatGPT’s release, the percentage of webpages published after November 2022 that display significant AI authorship signs has risen to 35%. In contrast, academic (.edu) and government (.gov) domains show AI authorship signs at a much lower rate, around 1% each. Organizational (.org) sites fall in between, with a 4.6% rate.
The study analyzed approximately 490,000 pages from the Common Crawl web archive, spanning from January 2021 to July 2026. Researchers utilized Open Pangram, an AI detection model, to identify patterns indicative of AI authorship. The model weighs statistical patterns across large batches of text rather than flagging specific words or phrases.
Linguistic shifts have also become more apparent since 2023, correlating with the rise of AI-generated text. These include a roughly twofold increase in the use of em dashes, a 63% rise in Oxford commas, and more than a doubling in the frequency of words often favored by AI, such as “delve,” “interplay,” and “testament.” Additionally, a grammatical structure known as “negative parallelism” has nearly tripled in frequency since 2023, though Pew notes it remains relatively uncommon overall.
Researchers caution that current detection models are not infallible. “Detection models can misclassify individual pages in both directions,” the study notes. Furthermore, the presence of “significant signs of AI authorship” does not necessarily mean a page was written entirely by a machine; much of the text may have been AI-assisted rather than fully AI-generated.
The disparity in AI content across different domain types is attributed to publishing cycles and editorial oversight. Commercial domains, which encompass a wide range of content from newsrooms to high-volume content farms, can produce pages at a pace that human editors may struggle to match. Academic and government sites, conversely, typically undergo more rigorous editorial review and institutional approval processes, leading to slower publishing cycles.
Separate research using similar detection methods has found AI-generated text in approximately 9% of U.S. newspaper articles this year, including content on opinion pages of publications like The New York Times. Looking ahead, companies like Anthropic are developing model-level text fingerprinting, which could make AI text easier to identify with greater accuracy.
The Pew Research Center’s findings highlight a significant and ongoing transformation in how web content is created, with AI playing an increasingly prominent role, particularly in the commercial online space.
Why This Matters
The materials describe a narrow update: A study by the Pew Research Center found that 10% of English-language webpages show significant signs of AI authorship, a figure that rises to 35% for pages published after November 2022. Detection models can misclassify individual pages in both directions.
Broader Context
Source materials place the factual news in this context: AI-written content is everywhere, literally.
Leave a Reply