Over 35% of New Website Content Will Be AI-Generated by 2025, Semantic Similarity Up 33%

When browsing websites, more content you see may come from AI. A study shows that by mid-2025, about 35% of newly published website content globally will be fully or partially AI-generated, compared to almost zero before ChatGPT's launch.

Researchers from Imperial College London, Internet Archive and Stanford University analyzed 33 months of web samples. They found AI-generated content has 33% higher semantic similarity than human content and 107% higher positive emotion scores, showing semantic contraction and excessive optimism.

This could narrow the range of ideas online, marginalize unique human views, and increase the risk of model collapse. Researchers warn that content homogenization may make users doubt information credibility, and the industry needs to rethink search algorithms, focusing on semantic diversity and encryption traceability.