Semantic search: searching by meaning, not just keywords
Semantic search finds content whose meaning is close to your question, even if it uses none of the words you typed. It relies on numerical representations of what texts mean, and it is how assistants find the right passage in your documents.
Data as of October 22, 2026
Key figures
| Level | Intermediate |
|---|
How it works
Semantic search is built on embeddings: each text is turned into a list of numbers that captures its meaning. Two texts about the same thing get vectors that sit close together, even if they share no words. To search, you compute the query’s vector and retrieve the texts whose vectors are nearest. At scale, these vectors live in a vector database, designed to find nearest neighbors quickly.
A milestone was the Sentence-BERT paper by Nils Reimers and Iryna Gurevych, posted to arXiv in August 2019, which showed how to produce sentence embeddings that can be compared efficiently.
A concrete example
An energy supplier’s support team has several hundred help articles. A customer writes: “my bill doubled even though nothing changed at home”.
- Keyword search looks for “bill” and “doubled” and mostly returns generic billing articles.
- Semantic search also finds “Understanding a catch-up bill after a meter reading”, which never uses the word “doubled” but answers the problem exactly.
Why it matters
It is the foundation of RAG: before an assistant can answer from your documents, it has to find the right passages, and semantic search usually does that job. It also powers recommendations (“similar articles”), duplicate detection and automatic sorting of incoming messages.
Common mistakes and myths
- “Close in meaning means relevant.” A passage can be on the same topic without answering the question. A reranking step often helps sort the results.
- “It replaces keywords.” Codes, names and exact figures are still better handled by classic search; hybrid search combines both approaches.
- “Just index whole documents.” Quality depends heavily on how the text is split into chunks: a passage that is too long dilutes the meaning, one that is too short loses its context.
Related terms: embedding, vector database, RAG, chunking.
Related terms
- Embedding · A list of numbers (a vector) that captures the meaning of a text or image, so content can be compared by how close in meaning it is.
- Vector database · A database that stores embeddings and quickly finds the ones closest to a query, which means the most relevant pieces of content.
- RAG (retrieval-augmented generation) · A technique that first retrieves relevant passages from your documents, then hands them to the model so its answer is based on them.
- Hybrid search · Search that combines exact keyword matching with meaning-based (embedding) search, then merges both result lists into a single ranking.
- Reranking · A second search stage in which a more precise model rereads the top candidates and puts them back in the right order of relevance.
- Chunking · Splitting long documents into small, coherent passages so they can be indexed and only the useful pieces are sent to the model.
Full glossary
Frequently asked questions
How is it different from keyword search?
Keyword search looks for the exact words (or close variants) in your query. Semantic search compares meaning: “cancel my plan” can find a page titled “ending your subscription”.
Is semantic search always better?
No. For a product code, a name or a contract number, keyword search is often more precise. That is why many systems combine the two.
Do I need an AI model to run semantic search?
Yes, you need an embedding model to turn texts into vectors. But that model is usually much smaller and cheaper to run than a conversational large language model.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.