Semantic search finds results based on what a query means. Instead of looking for the same words, it converts the query and the content into embeddings and returns the content whose vectors are nearest, typically from a vector database.
It handles synonyms and paraphrases well: a search for "cancel my plan" can find a page titled "Ending your subscription". It can be weaker on exact terms such as product codes, names or error numbers, which is why many systems combine it with ordinary keyword search, a setup often called hybrid search.
It is the retrieval half of Retrieval-Augmented generation and is also used for site search, recommendations and finding duplicates. The quality of the results depends on how the source was split (document chunking) and on the embedding model used.
Related: Knowledge Base, AI Agent Memory.