Definition

Embeddings

Dense numerical representations of text, images, or other data that capture semantic meaning in a way that machines can compare and search.

The Full Definition

An embedding is a vector — typically 768, 1024, or 1536 numbers — produced by a machine learning model that captures the semantic meaning of a piece of text (or an image, or audio). Two pieces of text with similar meaning produce embeddings that are close together in this high-dimensional space, even when they share no exact words. This is the fundamental representation that powers semantic search, RAG, clustering, and recommendation.

Why It Matters

Embeddings are the bridge between human meaning and machine comparison. Any AI system that retrieves, ranks, or clusters content meaningfully — search, RAG, dedup, recommendation, anomaly detection — is built on embeddings under the hood.

How This Shows Up in Practice

A support team embeds every historical ticket. When a new ticket comes in, the system embeds it and instantly surfaces the 5 most semantically similar past tickets — including the resolution paths that worked — even if the customer used completely different language.

Common Questions

Which embedding model should I use?

OpenAI's text-embedding-3-large and open-weight models like BGE and Nomic Embed are strong general-purpose choices. For specialized domains, fine-tuning your own embedding model often pays off.

Do embeddings need to be re-generated when the source model updates?

Yes — embeddings from different models aren't comparable. Plan for re-embedding when you change models, and budget for that operationally.

Related Terms

Want to put this to work?

A free process audit maps where embeddings — and the rest of the modern AI stack — actually move the needle in your business.

Survey My Business