What is Vector Search

Definition

Vector search finds items most similar in meaning to a query by comparing their embedding vectors and returning the nearest neighbours in vector space, enabling semantic retrieval where results match intent rather than exact keywords.

« Back to Glossary Index

Retrieves results by meaning rather than exact keyword match
Handles synonyms and paraphrasing that keyword search misses
Powers semantic search, recommendations, and RAG retrieval
Scales similarity queries over large embedding collections

Real World Example

A help centre uses vector search so a user asking 'my card won't work' surfaces the article titled 'troubleshooting declined payments', even though the two share almost no keywords, because their meanings are close.

FAQs

How does vector search differ from keyword search?

Vector search matches by semantic similarity of embeddings, while keyword search matches literal terms, so vector search captures meaning and synonyms.

What is nearest-neighbour search?

It is finding the vectors closest to a query vector in high-dimensional space, which represents the most semantically similar items.

Where is vector search used?

In semantic search, recommendations, deduplication, and as the retrieval step in RAG pipelines.