What Is Vector Search? How AI Finds Content by Meaning
Definition
Vector search finds content by meaning instead of exact words. It turns text into embeddings, lists of numbers that represent what the text means, then returns the stored items whose embeddings sit closest to the query's. That's how a search for "cheap flights" can find a page about "budget airfare."
How Searchable measures this
No AI engine shows its search index, but its results are visible. Searchable records the pages each engine retrieved as sources for your tracked prompts and, on engines that expose them, the search queries it ran, so you can see which of your pages get found and for which questions.
How vector search works
Vector search runs in two stages. First, every document, or every chunk of a document, is passed through an embedding model. The model outputs a vector, a long list of numbers, that places the text in a mathematical space where similar meanings sit close together. The vectors are stored in an index.
When a query comes in, it's turned into a vector the same way. The system then finds the stored vectors nearest to it, usually by cosine similarity, and returns the matching text. A query and a passage can match without sharing a single word, as long as they mean similar things. Embeddings covers the first stage in more detail.
Vector search vs keyword search
Keyword search matches the words in a query to the words in a document, typically with a ranking formula such as BM25. Vector search matches meaning. Each one catches results the other misses.
| Keyword search | Vector search | |
|---|---|---|
| Matches on | Exact words and close variants | Meaning, through embeddings |
| Strong at | Product names, codes, rare terms | Paraphrases, synonyms, natural questions |
| Weak at | Different wording for the same idea | Exact identifiers and new terms |
| Example miss | "Budget airfare" for "cheap flights" | An exact model number |
Hybrid search
Hybrid search runs keyword search and vector search on the same query, then merges the two result lists into one. A common merging method is reciprocal rank fusion, which rewards documents that rank well on both lists.
The result matches both the exact terms and the meaning of a query. That's why hybrid retrieval is a common default in retrieval-augmented generation systems, where missing the one passage with the right product name can spoil the answer.
Where vector search shows up in AI search
AI engines retrieve before they answer, and some form of meaning-based matching is part of most modern retrieval. The exact setup inside any major AI engine isn't public, so treat any claim about a specific engine's index with care.
What is visible is the pattern in results. Engines often cite pages that answer the question well without repeating its exact words, which is what meaning-based retrieval would produce. They also split one question into several searches, a process called query fan-out, so a page can be retrieved for a sub-question the user never typed.
What vector search means for your content
Vector search usually works at the passage level, so each section of a page gets matched on its own:
- Write sections that answer one clear question each, so a section's embedding has a clear meaning.
- Use the words buyers use, including the exact names of products and competitors, so the keyword side of hybrid search can match too.
- Keep related facts together in one passage, since a retrieved chunk won't carry context from elsewhere on the page.
Content chunking explains how pages get split into the passages that are embedded.
Frequently asked questions
Is vector search the same as semantic search?
They overlap. Semantic search is the goal, finding results by meaning. Vector search is the most common way to do it, by comparing embeddings. A semantic search system can also use other signals, such as entities or a knowledge graph.
What is a vector database?
A vector database stores embeddings and is built to find the nearest ones to a query quickly, even across millions of items. Examples include Pinecone and the pgvector extension for Postgres.
What is approximate nearest neighbor search?
Comparing a query with every stored embedding is too slow at scale, so vector search usually uses approximate nearest neighbor (ANN) methods. They organize embeddings into structures, such as graphs, that find very close matches quickly while accepting a small chance of missing the exact best one.
Does vector search replace keyword search?
No. Keyword search is still better for exact terms like product names and part numbers. Many systems run both and combine the results, which is called hybrid search.
Related terms
Embeddings are lists of numbers that represent the meaning of a piece of content, such as a paragraph of text or an image. Content with similar meaning gets similar numbers, so a system can find a relevant passage even when it shares no words with the question. Embeddings are what make semantic search and AI retrieval work.
Semantic search is a way of finding information by matching the meaning of a query instead of its exact words.
Retrieval-augmented generation (RAG) is a technique where an AI model first searches a set of documents for passages that match a question, then writes its answer from those passages.
Content chunking describes writing pages in clear, self-contained sections.