Glossary
What is Embeddings?
Lists of numbers that represent the meaning of text, images or other data, so that similar things can be found by comparing numbers.
In plain words
An embedding turns a piece of content into a list of numbers in such a way that content with similar meaning gets similar numbers. "Cancel my order" and "I want to stop my purchase" end up close together, while "track my parcel" lands elsewhere.
What they are used for
- Search by meaning rather than by exact words
- Finding the right passages to give a language model (see RAG)
- Recommending similar products, articles or people
- Grouping similar support tickets or reviews
- Spotting duplicates or near-duplicates
How they are made
A special model, trained for the purpose, reads the content and outputs the numbers. Different models make different numbers, so the content you store and the questions you ask must be turned into numbers by the same model. If you change the model, everything stored has to be redone.
Common mistakes
- Mixing embeddings from two different models
- Choosing a model that was not tested on your language or subject
- Forgetting to refresh embeddings when the source content changes
- Assuming "similar meaning" always means "correct answer"
Questions to ask a company that offers this
- Which embedding model will you use, and has it been tested on our language and subject?
- What happens to the stored data if we change the model later?
- How do changes to our content reach the index?
- How will you measure whether search results are actually relevant?
A good company will answer these plainly and in writing. Vague or evasive answers are worth noting.
Related terms
Find companies
Explanation last read through on 13 October 2026. It is general guidance in plain language, not advice for your particular project. Tell us if something is wrong or unclear.