Document Archive & Local Full Text•Localized Archive
Ahrefs Research: Why ChatGPT Selects One Resource Over Another
Ahrefs monitored RAG processing layers inside ChatGPT to determine how citation networks choose references.
1. Key Metrics
- The 15% Filter: 85% of crawled web documents are skipped during LLM final generation.
- Title-Query Vector Alignment: Natural, query-focused H1 structures dominate citations.
- Slug Semantics: Crawler heuristics favor clean URL slugs over opaque query parameters.