Research Reveals 35% of New Websites Are AI-Generated or AI-Assisted by Mid-2025
Key Takeaways
- ▸35% of newly published websites were AI-generated or AI-assisted by mid-2025, up from zero before ChatGPT's launch
- ▸AI-generated text correlates with decreased semantic diversity and increased positive sentiment on the internet
- ▸Contrary to public perception, factual accuracy and stylistic diversity were not found to be significantly harmed by AI-generated content
Summary
A comprehensive academic study analyzing representative samples of newly published websites from the Internet Archive found that by mid-2025, approximately 35% of new websites were classified as AI-generated or AI-assisted, a dramatic increase from zero before ChatGPT's launch in November 2022. Researchers from Imperial College London, Stanford University, and the Internet Archive employed multiple AI detection methods and robust sampling techniques to construct a statistically representative snapshot of internet content trends.
The study presents nuanced findings that diverge from widespread public concerns about AI's impact on online discourse. While the research documents a correlation between increased AI-generated text and decreased semantic diversity, as well as increased positive sentiment, it found no statistically significant evidence that AI-generated content has reduced factual accuracy or stylistic diversity—contrary to public perception. The findings were based on analysis of web pages, surveys of 853 US adults about their AI usage and perceptions, and validation across multiple AI detection models (Binoculars, Desklib, DivEye, and Pangram v3).
The rapid proliferation of AI-generated text reflects how quickly AI tools have become embedded in content creation workflows. The research highlights a significant gap between public fears about AI contamination of the internet and empirical evidence about actual impacts, raising important questions about which concerns merit prioritization and policy attention.
- Researchers used representative sampling and multiple detection methods to study internet-wide impacts from 2022-2025
Editorial Opinion
This research provides essential empirical grounding to an increasingly heated debate about AI's impact on information quality. While the 35% figure underscores how rapidly AI tools have transformed content creation, the more important finding—that AI hasn't significantly degraded factual accuracy or stylistic diversity—challenges narratives of inevitable internet contamination. The study's most critical insight may be that public perception of AI's harms often outpaces evidence, suggesting policymakers should focus on proven risks rather than feared ones.
