What is semantic search?
Search by meaning instead of exact wording, so a vague description still finds the right record.
Semantic search — Semantic search finds records by meaning rather than exact words, by comparing numeric representations (embeddings) of the query and the content. It is used to find "the client who asked about pricing last month" without knowing which words they used.
Why it matters
A keyword search only finds text that matches the words typed in. Someone searching "the client who asked about pricing last month" using keyword search would need the word "pricing" to actually appear in that record; if the note said "wanted to know the cost" instead, the search would miss it. Semantic search works differently: it converts both the query and the content into embeddings, numeric representations that place similar meanings near each other, and finds content close to the query in that space rather than matching literal words.
This makes semantic search well suited to the kind of vague, natural questions a person actually asks, and to retrieval-augmented generation, where a model needs the right passage from a document store without knowing its exact wording in advance. It is not a full replacement for keyword search, though. An invoice number, an email address or an exact order ID needs an exact match, and semantic search can return near misses for queries like that where only the precise identifier is correct.
How semantic search works
- 1
Content is converted to embeddings
Each record or document gets a numeric representation of its meaning.
- 2
A query comes in
A natural question or description, in whatever words the person used.
- 3
The query is converted the same way
Into an embedding in the same numeric space as the content.
- 4
The closest matches are found
By comparing the query's embedding to every content embedding.
- 5
Results are ranked and returned
By closeness of meaning, not by literal word overlap.
The mistake to watch for
Questions
- How is semantic search different from keyword search?
- Keyword search matches literal words in the query against literal words in the content. Semantic search compares meaning using embeddings, so a query worded differently from the source can still find it.
- What is an embedding?
- A list of numbers that represents the meaning of a piece of text, generated by a model, positioned so that texts with similar meaning have numbers close to each other.
- Where is semantic search used in a CRM?
- Finding records or knowledge-base passages from a natural description, and as the retrieval step in retrieval-augmented generation, where the right document has to be found before a model can answer.