How does it work?
Your query is turned into an embedding and compared with the embeddings of your documents in a vector store. The system returns the most related results, not the ones with the exact same words.
Someone searches for "lower energy costs" and also finds your page about "saving on heating bills".
What is it for?
- Better search on your website or across your documents
- Answering FAQ and knowledge-base questions, even when the visitor uses different words
- The retrieval step in RAG, which an AI assistant relies on
- Product search and recommendations
Keyword search vs. semantic search
A traditional search matches exact words: if the word isn't there, it finds nothing. Semantic search understands intent and also finds synonyms, paraphrases and related phrasings. For a ten-question FAQ, keyword search is fine; across hundreds of pages of content, semantic search makes the difference.
Related terms
- Embeddings: the meaning vectors that make semantic search possible.
- Vector store: the database that searches by meaning.
- RAG: uses semantic search to retrieve the right info.
