35% of new websites by mid-2025 were AI-generated
Stanford-led study finds 35% of newly published websites by mid-2025 were AI-generated or AI-assisted, with higher semantic similarity and a much more positive tone.
A study by researchers at Stanford University, Imperial College London and the Internet Archive found that about 35% of newly published websites were AI-generated or AI-assisted by mid-2025. The share was essentially zero before the November 2022 launch of ChatGPT. The team analyzed 33 months of archived webpages from the Internet Archive’s Wayback Machine and used an AI text detector called Pangram v3 to classify pages.
The researchers tested six hypotheses about how AI content affects the web and found evidence for two. Pages identified as AI-generated showed pairwise semantic similarity scores roughly 33% higher than human-written pages. The study describes this as more frequent repetition of similar ideas and phrasing across pages. AI-identified content also registered more than 107% higher positive sentiment than human content; the authors link that pattern to models’ tendency to produce agreeable, approval-seeking language.
The analysis did not find a statistically significant link between the share of AI content and factual error rates. Character-level tests did not show a clear rise in stylistic homogeneity tied to AI prevalence. The paper is titled “The Impact of AI-Generated Text on the Internet.”
The researchers flagged a training-data risk. At roughly 35% prevalence of AI-generated pages, the study says the theoretical risk of model collapse-where future language models could degrade after being trained on data that includes outputs from earlier models-becomes an empirical concern for the next generation of foundation models. The team notes that future models trained on contemporary web crawls will ingest substantial amounts of AI-produced text.
The investigators are working with the Internet Archive to build a continuous monitoring system to track the share of AI content on the web in real time. A companion U.S. survey conducted with the study found most Americans believe various negative effects from AI content; people who use AI infrequently were about 12% more likely to endorse those harms than frequent users.
Jonáš Doležal, a co-author and researcher at Imperial College London, described the speed of the change as “staggering.”
The material on GNcrypto is intended solely for informational use and must not be regarded as financial advice. We make every effort to keep the content accurate and current, but we cannot warrant its precision, completeness, or reliability. GNcrypto does not take responsibility for any mistakes, omissions, or financial losses resulting from reliance on this information. Any actions you take based on this content are done at your own risk. Always conduct independent research and seek guidance from a qualified specialist. For further details, please review our Terms, Privacy Policy and Disclaimers.
Articles by this author
