Suppose you search help articles for “can’t sign in.” A vector database might find “Reset your password,” even though the words differ. Each article has an embedding: a list of numbers, or vector, representing its content. The app makes one for your search too, and the database finds the closest matches.

This is one way to power semantic search across a large collection.

A match is not an answer

Similar items may still be irrelevant, outdated, or off limits to a user. Apps must check who can see each result and may use keyword search as well when exact names or codes matter.