Skip to main content
AI search picks sources through a pipeline, and each stage fails differently. Peec’s reranker guide (full guide, Aug 2026) describes the common documented pattern:
  1. Query planning: the prompt is rewritten and split into several searches — see What is a query fan-out.
  2. Retrieval (wide net): keyword search (BM25) and vector search run together; result lists merge via Reciprocal Rank Fusion, rewarding documents that rank well in both. OpenAI’s documented File Search works this way (800-token chunks, 400 overlap, hybrid search, optional reranker).
  3. Reranking (fine sieve): a cross-encoder-style model reads each candidate passage against the query and re-scores the shortlist. Your page is often scored as passages, not one URL — a strong page can contain a weak candidate passage, which is why page-level SEO metrics cannot explain every citation win or loss.
  4. Generation: the LLM writes from the surviving passages and may cite some.
The diagnostic rule (Aug 2026): not retrieved → indexing, coverage, authority, or freshness problem; retrieved but losing → answer shape or passage relevance; strong passage but no citation → source quality, diversity rules, or generation-stage behavior. Fix the failed stage first — rewriting a paragraph for the reranker cannot fix a page that was never retrieved. What to actually write once diagnosed: Answer shape for AI citations. Source: Rerankers for GEO/AEO: How AI search chooses passages and sources, Peec (Metehan Yesilyurt), Aug 2026.