Lexical search retrieves records by matching query words or phrases in standard text search. For a query such as “Dubious Parenting Advice,” it can find literal or partial matches that semantic search may overlook. Production retrieval commonly combines both strategies: lexical matching preserves exact wording, while vector search captures related meaning. section 3.2.7 section 1.1
A hybrid retrieval workflow uses lexical search as one stage in a broader search pipeline:
- Rewrite an informal question into focused search terms, such as converting “What’s the warranty on my X24 blender?” into “X24 blender warranty.” section 1.1
- Submit the refined text query alongside a vector query so literal matches and semantically related records are retrieved together. section 3.2.7
- Merge and rank the returned hits before passing relevant results to downstream answer generation. section 3.2.7 section 1.1
- Prompt for clarification or rephrasing when retrieval returns nothing relevant. section 1.1
Finding a closely titled product record
Searching for “Dubious parenting advice” with both text and vector retrieval returns a record titled “Dubious Parenting Tips,” demonstrating how lexical matching and semantic similarity can complement one another. section 3.2.7













