Semantic search retrieves results based on meaning rather than exact keyword matches. A semantic search for how to lower my electricity bill can return an article about reducing home energy costs even though the two share almost no words, because the system understands they are about the same thing.
This works by representing both the query and the documents as embedding vectors, then finding the documents whose vectors are closest to the query’s. Because the embeddings encode meaning, conceptually related text ends up nearby in vector space, allowing the search to handle synonyms, paraphrases, spelling variations, and even different languages — all without any manual list of synonyms or keyword rules.
The advantage over traditional keyword search is generalisation: semantic search works on queries it has never seen, matching intent rather than surface form. Its weakness is the mirror image — it can overlook exact matches for specific codes, names, or rare terms that embeddings represent poorly. For this reason semantic search is often combined with keyword search in a hybrid system, capturing both meaning and precise terminology.