Vector search retrieves items by comparing their numerical representations, called vectors, with a query vector. In text systems, an embedding model creates those representations so related meanings can be close together. Retrieval may use cosine similarity or another distance measure, with exact or approximate matching.
Vector search is a technique, while semantic search is the broader goal of finding information by meaning. A semantic search system can combine vectors with keywords, filters, and other ranking signals.
Practical relevance
A customer asks about "keeping sensitive files inside the country." A system using suitable embeddings might retrieve a passage about data residency even though it uses different words. Clear explanations of specific customer needs give retrieval systems useful material to compare.
A vector match only selects a candidate. It does not guarantee an AI citation, a favorable recommendation, or a search ranking. The embedding model, indexing, filters, and later answer generation all affect results.