Polp
← Back to glossary
RAG and search

Hybrid search

A retrieval method that combines lexical keyword matching with semantic vector search and merges both result sets into one ranking.

Business example

An employee asks for the travel policy using everyday language. Semantic retrieval finds the concept, while keyword retrieval preserves an exact policy code and the hybrid ranker combines both signals.

How it works

  1. 1.Run a lexical query that rewards exact terms and identifiers.
  2. 2.Run a vector query that retrieves conceptually similar passages.
  3. 3.Normalize or fuse both ranked result lists into one ordered set.
  4. 4.Apply permissions, metadata, and optional reranking before sending context to the answer model.

Common misconceptions

Hybrid search simply means using a vector database.
A hybrid pipeline keeps at least two retrieval signals—typically lexical and vector—and defines how their results are combined.
Semantic search makes exact keyword matching obsolete.
Exact names, codes, acronyms, and legal references often remain strong lexical signals.

Related concepts

Semantic searchEmbeddingsVector databaseRetrieval-Augmented Generation (RAG)

Related reading

RAG is not uploading PDFs: permissions, freshness, traceability, and reliable sources
Goodbye file search: how conversational AI is replacing enterprise search

Sources

Reviewed:

Reviewed by: Javier Chulvi Bernad · LLM Engineer · Madrid