A recent study conducted jointly by Stanford University, Imperial College London, and the Internet Archive projects that by mid-2025, around 35% of newly created websites will be either fully AI-generated or produced with AI assistance. The estimate marks a sharp rise from the period before ChatGPT launched in November 2022, when such content was described as nearly nonexistent.
How the researchers measured AI’s web presence
The study, titled The Impact of AI-Generated Text on the Internet, relied on Internet Archive data and used the AI text detection system Pangram v3 to classify web pages. The researchers focused on how AI-generated pages differ from human-written content in terms of semantic patterns and emotional tone.
According to the findings, AI-generated websites show 33% higher semantic similarity than human-authored pages, suggesting a decline in semantic diversity across the web. The study also found that AI-generated content scored 107% higher in positive sentiment, which the authors linked to the tendency of language models to produce polished and optimistic text.
No clear evidence of factual decline yet
Despite growing concerns that large volumes of AI-written material could weaken information quality online, the researchers said they found no significant evidence that AI-generated content is currently harming factual accuracy on the internet. In other words, the strongest measurable effects so far appear to be related to tone, repetition, and stylistic convergence rather than proven large-scale factual distortion.
The research team added that it is now working with the Internet Archive to build a monitoring tool capable of tracking AI’s online presence in real time. If successful, that system could provide ongoing data for researchers, platforms, and policymakers seeking to understand how generative AI is reshaping the structure and character of the web.