Reranking
Reranking is the process of re-scoring retrieved documents using a more sophisticated model, ordering them by relevance before feeding them to the final answer generation step. It improves answer quality by filtering out noise.
Definition
Initial retrieval often returns thousands of candidates, many only tangentially related. A reranker is a specialized model that takes these candidates and produces a more accurate relevance score. Reranking removes documents that initial retrieval misjudged, ensuring that only truly relevant sources reach the generation step.
A typical RAG pipeline first does fast, approximate retrieval on embeddings, then a reranker evaluates the top candidates with greater precision. Only the top reranked results get passed to the LLM for citation and synthesis. This two-stage approach balances speed with accuracy.
Why it matters for AI visibility
Reranking determines whether retrieved documents actually make it into final answers. Your brand could be retrieved initially but filtered out by a reranker that judges your content less relevant than others. Writing content that clearly establishes relevance and authority for your target topics improves reranking scores.
Related terms
Retrieval-Augmented Generation (RAG)
RAG is the technique of retrieving relevant documents at answer time, then using them to ground an LLM's response. It enables AI search engines to cite sources while generating answers.
AIHybrid Search
Hybrid Search combines vector search with keyword search, leveraging both semantic similarity and exact term matching to retrieve relevant documents. It provides more robust results than either method alone.
GEOAI Grounding
AI Grounding is the process of constraining LLM generation to reference and cite specific source documents, reducing hallucination and ensuring generated responses are anchored to retrievable content.
AIVector Search
Vector Search is the technique of finding semantically similar documents by comparing their embeddings. It powers retrieval in AI search engines and RAG systems.