What are embeddings?
How AI turns meaning into numbers it can actually compare — the quiet mechanism behind search that understands what you meant, not just what you typed.
An embedding is a way of representing a word, sentence, or document as a list of numbers — positioned so that things with similar meaning end up mathematically close together, and unrelated things end up far apart. It's how a system can tell "puppy" and "dog" are related without either word ever literally appearing in the other, because meaning gets compared as position, not as matching text.
How this actually works
Picture a map where every word, sentence, or document has a specific location — not on a 2D map, but in a space with hundreds or thousands of dimensions. Words used in similar contexts during training end up positioned near each other on that map. "King" and "queen" sit close together; "king" and "banana" sit far apart. Comparing two pieces of text then becomes a matter of measuring the distance between their two points, rather than checking whether they share the same words.
Why this matters in practice
This is what lets AI-powered search understand a question rather than just matching keywords — a search for "how to fix a slow laptop" can surface a document titled "improving computer performance" even though not one word overlaps, because the two are close together in meaning-space. It's also the core mechanism behind RAG, which relies on embeddings to find genuinely relevant passages before generating an answer.
Related terms
FAQ
Do I ever interact with embeddings directly?
Rarely, if you're just using an AI chat tool — embeddings mostly work behind the scenes in search and retrieval features.
Are embeddings the same across every AI model?
No — different models produce different embeddings, so comparisons generally need to use embeddings from the same model or system consistently.
Is this the same thing as a token?
No — a token is a chunk of text; an embedding is a numeric representation of meaning, often built from many tokens at once.
Back to the full AI Glossary, or see this mechanism in action in What is RAG?