Hybrid Search
Also called Hybrid retrieval, Keyword plus vector search
Search that combines keyword matching with meaning-based (vector) search, so it finds both exact terms and differently worded matches.
What is hybrid search?
Hybrid search runs two kinds of search at once and merges the results:
- Keyword search finds passages containing the exact words in the query. It's excellent for names, product codes, error messages and jargon.
- Semantic search finds passages with the same meaning, even if the wording is different, by comparing embeddings.
Each covers the other's blind spots. A search for "can't log in" should find a help article titled "Password reset", which keyword search misses. A search for "error E-4012" needs that exact code, which meaning-based search can blur.
How does it work?
The query goes to a keyword index and a vector database in parallel. Each returns a ranked list, and the two lists are combined into one, usually by giving more weight to results that rank highly in both. The weighting between keyword and meaning can be tuned. Many systems then pass the merged list to a reranker.
What is it commonly confused with?
- Semantic search. Semantic search is one half of hybrid search, not the same thing.
- Searching twice. The value isn't two result lists; it's the single merged ranking.
Why does it matter?
Real questions mix everyday language with exact terms, so hybrid search usually finds the right passage more often than either method alone. That makes it a common default for RAG systems and site search, and most vector databases now support it.
