AI glossary Data & Retrieval
What is hybrid search?
← All glossary termsA retrieval approach that combines keyword search (such as BM25) with semantic vector search and merges both result lists.
Explained
Hybrid search runs two kinds of retrieval over the same content and merges the results. Keyword search, usually with a ranking function such as BM25, finds documents that contain the exact words of the query, which matters for product codes, policy numbers, names and jargon. Vector search finds documents with a similar meaning even when they use different words. Combining them gives better coverage than either alone, which is why hybrid search is a common default for enterprise search and retrieval-augmented generation (RAG).
How it works
- The query is sent to a keyword index (for example BM25 over an inverted index) and to a vector index built from embeddings.
- Each returns a ranked list of candidates.
- The two lists are merged into one, either by combining normalised scores or by rank-based methods such as reciprocal rank fusion (RRF).
- Optionally, a reranker model reorders the top results before they are shown or passed to a language model.
Why combine keyword and vector search
Vector search is good at paraphrases and vague questions but can miss exact identifiers or rare terms, because similar-looking codes end up close together. Keyword search is precise on exact terms but misses documents that say the same thing with other words.
In business content, both kinds of query are common: a user may ask about 'the refund rules for damaged goods' and, in the next query, type a specific contract number.
Practical points
- Use the same filters and permissions on both indexes so the merged list does not leak documents.
- Rank-based fusion such as RRF is simple and does not require the two scores to be on the same scale.
- Tune with your own queries: the right balance between keyword and semantic results depends on the content and on how people search.
- Many search engines and databases now offer hybrid search natively, so it rarely requires building the merge yourself.
Frequently asked questions
What is the difference between hybrid search and semantic search?
Semantic search uses only vector similarity to find content with a similar meaning. Hybrid search adds keyword matching on top and merges both, so it also finds exact terms such as codes and names that semantic search can miss.
What is BM25 in hybrid search?
BM25 is a classic ranking function for keyword search. It scores documents by how often the query terms appear, adjusted for how rare those terms are and for document length. In hybrid search it provides the lexical half of the results.
What is reciprocal rank fusion?
Reciprocal rank fusion (RRF) is a simple way to merge ranked lists. Each document gets a score based on its position in each list, and the scores are added. It works well without needing to calibrate the original scores against each other.
Is hybrid search better for RAG?
It is often a safer default, because enterprise questions mix natural language with exact references. Whether it helps in your case should be checked with an evaluation set of real queries and the documents that should come back for each.
Decision guides that use this term
Where it shows up at Thinkia
Further reading
Related terms
Sources
- Robertson and Zaragoza · The probabilistic relevance framework: BM25 and beyond
- Cormack, Clarke and Büttcher · Reciprocal rank fusion (SIGIR 2009)
- Elastic · Hybrid search: combined full-text and kNN results
- Microsoft · Reciprocal rank fusion (RRF) in Azure Cosmos DB queries
Updated: 2026-10-21