Hybrid search combines two complementary retrieval methods — keyword search and semantic search — and merges their results so that neither the exact term a user typed nor the underlying meaning they intended gets missed.
Why it matters
No single retrieval method handles every case well. Keyword search is reliable when someone types a precise term: a product code, a legal clause number, a person's name. It finds those exactly, without the fuzziness that semantic search sometimes introduces. But it fails completely when the query and the answer are phrased differently — synonyms, paraphrases, and conceptual matches slip through an exact-word index invisible.
Semantic search inverts that strength and weakness. It finds passages that mean the same thing as the query even when no words overlap, but it can miss a precise identifier that happens not to be common enough to cluster strongly in the embedding space.
Hybrid search runs both at once and fuses the scores — often weighted, sometimes via a reciprocal rank fusion method — so the final ranked list surfaces passages that are relevant by either measure. In practice, this matters for document Q&A: a user asking about "Section 4.2 liability caps" needs both the exact section reference and the ability to match surrounding context that uses different words to discuss the same restriction.
A reranker often sits on top of the fused list, re-scoring the top candidates by reading query and passage together for a final precision boost before the model sees them.