# Vector search

> Vector search represents text (and other data) as numerical embeddings and finds results by similarity in that vector space, which is what lets search match on meaning instead of literal words.

_Source: https://nobi.ai/glossary/vector-search_

## What does "vector search" mean?

Vector search is a technique that represents text as **embeddings** - lists of
numbers that capture meaning - and retrieves results by finding the embeddings
most **similar** to the query's. Because closeness in this vector space
reflects closeness in meaning, vector search can match a query to a product
even when they share no exact words.

It's the engine under most "search by meaning" experiences, including
[semantic search](/glossary/semantic-search).

## How does vector search work?

1. **Embed the catalog.** Each product's text (title, description, attributes)
   is converted into an embedding and stored.
2. **Embed the query.** When a shopper searches, their query is converted into
   an embedding the same way.
3. **Find nearest neighbors.** The system returns the products whose embeddings
   are closest to the query's, ranked by similarity.

Because the comparison is mathematical, synonyms, paraphrases, and descriptive
phrases all land near the right products without hand-written rules.

## Why does vector search matter?

Vector search is what makes meaning-based retrieval possible at scale. For a
store, the practical payoff is fewer dead-end searches on descriptive or
long-tail queries and less time spent maintaining synonym lists. It's also the
retrieval step behind grounded AI answers: a system can pull the most relevant
facts by vector similarity before generating a response, which keeps answers
tied to your real content.

In [Nobi](https://dashboard.nobi.ai), embeddings power both search and the
shopping assistant, so a shopper can describe what they want and get matched to
the live catalog.

## Frequently asked questions

**Is vector search the same as semantic search?**
They're closely related. Vector search is the underlying technique; semantic
search is the meaning-based outcome it enables.

**What is an embedding?**
A list of numbers representing the meaning of a piece of text. Similar meanings
produce embeddings that sit close together.

**Do I need a separate vector database?**
Not necessarily - many teams use a search product that handles embedding and
similarity for them, with no separate infrastructure to run.