Semantic search
Also known as: vector search, meaning-based search, AI search, dense retrieval
Traditional keyword search (like a database search or a basic Ctrl+F) looks for documents containing the exact words in your query. Semantic search instead uses embeddings: both the query and the documents are converted to numerical vectors, and the search finds documents whose vectors are close to the query vector in meaning-space. It finds conceptually similar content even when the vocabulary differs.
Semantic search is what makes 'talk to your documents' products work. You upload your company wiki, embed all the pages, and when a user asks a natural language question, the system finds the most semantically relevant pages, not just pages that contain the exact query keywords. This is meaningfully better for ambiguous queries, synonym-heavy domains, and conversational inputs.
The tradeoff: semantic search is better for conceptual matches but can miss exact terms, product names, or codes that keyword search would catch reliably. Production systems often combine semantic search with keyword search (hybrid retrieval) to get the benefits of both. Knowing the difference helps you debug retrieval issues: if your semantic search is missing exact product codes, the fix is to add a keyword search layer, not to change the embedding model.