GlossaryAI search
What is vector search?
Also called: semantic search, similarity search, vector database, nearest neighbor search
Definition
Vector search finds content by meaning. It turns the query and every document into embeddings, then returns the documents whose vectors are closest to the query's, instead of matching exact words.
Vector search, explained
Classic keyword search builds an index of words and returns documents that contain the query terms, ranked by signals like how often and where the words appear. Vector search builds an index of embeddings and returns the nearest neighbors to the query's embedding. The first is precise about wording. The second is forgiving about wording and precise about topic.
Doing this fast at scale is the hard part. Comparing a query with millions of vectors one by one is slow, so vector databases use approximate nearest neighbor indexes that trade a tiny bit of accuracy for large speedups. Products like pgvector, Pinecone, Weaviate and the vector features in most cloud databases all do versions of this.
Most modern AI search is hybrid. The system runs a keyword search and a vector search, merges the candidates, and often reranks them with another model before handing the best passages to the LLM. Hybrid setups exist because each method fails differently: vectors miss exact identifiers, keywords miss paraphrases.
For a site owner, this explains a few things you might notice. Your page can be cited for a question that shares almost none of its words. Precise names, numbers and error messages still matter, because the keyword half of the system matches them exactly. And pages that wander across topics tend to lose out to focused ones, because their passages don't sit close to any particular question.
If you're adding search to your own product, start hybrid. Pure vector search feels magical in a demo and then fails on the first user who types an exact SKU or error code.
Why it matters for founders
AI assistants find passages this way. Writing focused sections that each answer one question, while keeping exact names and numbers in the text, serves both halves of hybrid retrieval.
Example
A user asks, "tool that turns podcast episodes into short clips". Vector search surfaces a page titled "Repurpose long audio into social videos" even though it never says "podcast clips", because the meanings are close.
Common mistakes
- Assuming vector search makes keywords irrelevant. Hybrid systems still match exact terms.
- Shipping pure vector search in your product and failing on exact codes and names.
- Writing long sections that cover many topics, which dilutes each passage's meaning.
Related terms
- EmbeddingsEmbeddings are lists of numbers (vectors) that represent the meaning of a piece of text. Texts with similar meaning get vectors that sit close together, which lets software search by meaning instead of exact words.
- Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) is a technique where a system first retrieves relevant documents, then passes them to a language model so it can answer from that material instead of memory alone.
- Search intentSearch intent is the goal behind a search: what the person actually wants when they type a query. Pages rank best when their format and content match that goal, not just the words.
- Content chunkingContent chunking is structuring a page as self-contained sections, each answering one question, so search engines and AI retrieval systems can find, understand and quote a single passage without the rest of the page.