What is semantic search?
Search by meaning instead of exact wording, so a vague description still finds the right record.
Want to see it first? Book a demo
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. Say you search for "the client who asked about pricing last month." Keyword search needs the word "pricing" to be in that record. If the note said "wanted to know the cost" instead, the search would miss it. Semantic search works another way. It turns both the query and the content into embeddings: lists of numbers that place like meanings near each other. Then it finds content close to the query in that space, rather than matching the exact words.
That makes semantic search a good fit for the loose, plain questions people really ask. It also suits retrieval-augmented generation, where a model needs the right passage from a store of documents without knowing its wording ahead of time. It won't fully replace keyword search, though. An invoice number, an email address or an order ID needs an exact match. For queries like that, semantic search can return near misses when only the exact ID is right.
How semantic search works
- 1
Content is converted to embeddings
Each record or document gets a list of numbers that stands for its meaning.
- 2
A query comes in
A plain question or description, in whatever words the person used.
- 3
The query is converted the same way
It becomes an embedding in the same number space as the content.
- 4
The closest matches are found
The query's embedding is checked against each content embedding.
- 5
Results are ranked and returned
They're sorted by how close the meaning is, not by shared words.
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?
- An embedding is a list of numbers, made by a model, that stands for the meaning of a piece of text. Texts with like meanings get numbers close to each other.
- Where is semantic search used in a CRM?
- It finds records or knowledge-base passages from a plain description. It's also the lookup step in retrieval-augmented generation, where the right document has to be found before a model can answer.