How LLMs decide which pages to cite — and how to optimize for it

When ChatGPT or Perplexity answers a question, it runs RAG: retrieves top candidates from a crawled index, then scores them. The scoring criteria are public knowledge from the Princeton GEO paper (arxiv.org/abs/2311.09735). Key signals: answer directness, cited statistics, structured data (JSON-LD), crawl access, and content freshness. What surprised me most in the research: schema markup alone shifts precise information extraction from 16% to 54%. That's not a marginal gain — that's the diffe