Embedding
An embedding is a numeric vector that captures the meaning of a piece of text, letting a docs search system match questions to relevant passages by semantic similarity rather than keywords.
In docs search, every chunk of documentation is converted into an embedding and stored in a vector index. A user's question is embedded the same way, and the system returns the chunks whose vectors sit closest to the question's.
This is what lets semantic search connect "how do I revoke a token" to a page titled "Deleting API credentials". Exact-word matching can miss that connection. Embeddings help a system find passages with related meaning, although retrieval still needs evaluation.
Embeddings are the retrieval half of RAG, and their usefulness depends on chunking. Embed a whole page and the vector blurs many topics together; a focused section can improve the match, provided it keeps enough context.
Frequently asked questions
What does an embedding actually look like?
A list of hundreds or thousands of floating-point numbers produced by an embedding model. Texts with similar meaning produce vectors that sit close together.
Do embeddings go stale when docs change?
Yes. When a page changes, its chunks must be re-embedded and re-indexed, which is why docs AI systems recrawl the site periodically.