AI-Generated Content Dominates New Websites by Mid-2025
- By mid-2025, AI-generated or AI-assisted content accounted for 35% of newly published websites.
- The prevalence of AI content has led to a semantic contraction, with pairwise semantic similarity scores being 33% higher than human-written content.
- AI-generated sites exhibited positive sentiment scores over 107% higher than those written by humans.
- Research indicates no significant increase in factual inaccuracies or stylistic homogeneity due to AI content proliferation.
- The risk of model collapse becomes an empirical concern as future models may train on less semantically diverse, AI-generated data.
A study conducted by Stanford University and Imperial College London reveals that a substantial portion of new web content is now generated or assisted by AI, significantly altering the internet’s landscape in just three years since ChatGPT’s launch in November 2022. The findings highlight a shift towards more semantically similar and positively skewed content without increasing factual errors or uniformity in writing styles.
With AI-generated content comprising a third of new websites, the risk of future model degradation becomes tangible as they train on increasingly homogeneous data sources. Source
