semantic search

Semantic search is search by meaning rather than by matching letters. Traditional keyword search returns the documents that literally contain your words; semantic search returns the documents that are about the same thing, even when they are worded completely differently. Ask for "ways to lower my electricity bill" and it can surface an article on "reducing household energy costs" that shares almost no words with your query.

It works by embedding both the query and every document into the same vector space and then ranking by closeness, so synonyms and paraphrases naturally end up near one another. Inside a RAG pipeline, semantic search is the usual first-stage retriever. Its weakness is the exact mirror of its strength: chasing meaning, it can glide right past an exact part number, surname, or acronym that a plain keyword match would have caught instantly.