Skip to content
Module
AI

What is semantic search? Definition and examples

Semantic search finds content based on meaning rather than on the exact words in the query. The query and the documents are turned into embeddings, vectors that capture what they mean, and the system returns the documents whose vectors sit closest to the query's. A search for "refund" can therefore surface a page about "getting your money back" or "returning a purchase."

Definition written by Alex Dionisio, Senior Product Manager. Updated November 13, 2026.

Why it matters for a PM

Users rarely describe their need with the words the documentation uses. Semantic search cuts down on empty result pages and is the retrieval step in most RAG assistants. It also has blind spots that a PM needs to know in order to specify the expected behavior: product codes, exact references, proper names and negations.

Example

On an online bank's help center, a customer types "my card stopped working abroad." Keyword search finds nothing, because the article is titled "Enable payments outside the eurozone." Semantic search ranks it first. The query "error E-402," however, returns poor semantic results, so the team combines both approaches.

Key points

  • It relies on an embedding model and often on a vector database.
  • It handles synonyms and rephrasing well, but struggles with exact identifiers and numbers.
  • Hybrid search, which blends meaning and keywords, is often the best option in production.
  • Reranking the top results frequently improves relevance.

Common mistakes

  • Dropping keyword search entirely when users also look for exact references.
  • Judging quality on a handful of hand-picked queries instead of a set of real searches.
  • Ignoring language: an embedding model that is weak in your users' language hurts results.

Go further with Module

The courses and lessons that cover this concept:

Guide on this topic