Training a model to identify AI-generated web content from structure alone

Researchers from Sitefire, a company that helps marketing teams get recommended by AI search, trained a model to identify AI-generated web content from its structure alone. The model was trained on 2,250 blog posts from 268 B2B company websites and was able to distinguish between AI-generated and human-written posts with 98% accuracy. The model found that AI-generated content often follows a 'tidy, self-announcing' structure, where the main point is repeated in the title, intro, and conclusion. The model was also able to identify which AI model wrote a post with 79% accuracy.