What is an embedding?
An embedding is a list of numbers that stands in for a piece of text's meaning, positioned so that texts which mean similar things end up with similar numbers even if they don't share any words.
To compare meaning rather than exact wording, AI systems convert text into an embedding: a long list of numbers produced by a model, where the position of those numbers reflects meaning rather than spelling. Two sentences that share no words in common, 'cut the price' and 'reduce the cost', can end up with very close embeddings, because they mean roughly the same thing, while two sentences with overlapping words but different meanings can end up far apart.
This is the mechanism behind vector search and much of how AI answer tools decide which passage of a page is relevant to a question, even when the question is phrased nothing like the page. It's worth being clear that this is not keyword matching with extra steps: repeating an exact target phrase doesn't reliably pull a page's embedding closer to a query, since the comparison is about meaning, not string overlap.
Related
- Vector searchVector search finds relevant text by comparing meaning rather than matching words, ranking passages by how close their embeddings are to the question's.
- Retrieval-augmented generation (RAG)Retrieval-augmented generation is the technique of retrieving relevant documents and giving them to a model as context before it answers.
- GroundingGrounding is when a model answers using documents retrieved at the time of the question, rather than only from what it absorbed during training.
- TokensTokens are the small chunks, often pieces of words rather than whole words, that a language model actually reads and produces, and they're the unit that usage limits, pricing, and the context window are all measured in.
Want to know where you actually stand on this? Run a free visibility check or try the free tools.