What are embeddings in AI?
Updated · By Robert Breen
Embeddings are lists of numbers that represent the meaning of a piece of text, an image or other content. Texts with similar meanings get similar numbers, so software can find related content by meaning, not just matching words. They are the basis of AI search over your documents.
Why it matters for a small business
Keyword search fails the moment people use different words. A customer asks about a "hot water tank" and your price list says "water heater". Embeddings fix that: both phrases land close together in meaning, so a search finds the right section. That is what lets an AI assistant answer questions from a long handbook, a folder of past quotes or years of support emails.
You will rarely create embeddings by hand. They work behind the scenes in tools that search your files, and in n8n they show up as embeddings nodes that feed a vector store. Knowing the idea helps you judge when a project needs that machinery (a large, changing pile of documents) and when a single uploaded file or a pasted list is enough.
In a real lesson: Build a Custom GPT Plumbing Quote Writer
None of Stepthrough's live lessons build embeddings directly. The closest you can practice is giving an assistant a document to search. In the plumbing quote writer lesson, you create a Custom GPT called Plumbing Quote Assistant for Blue Maple Plumbing, a made-up residential plumber, and use Upload files to attach Blue-Maple-Services-Pricing.pdf, with the service call fee, water heater, drain, leak and sump pump prices, the warranty and a sample quote.
When you then ask for "a quote description for a water heater replacement", the GPT has to find the part of that document that matters. OpenAI describes GPTs as searching uploaded knowledge for relevant content; matching a request to the right passage by meaning is the job embeddings are designed for.
For one short price list, you never see any of this. It becomes your problem when you outgrow a single file, for example hundreds of past quotes, and want an n8n agent to find the three most similar jobs. Then you would choose an embeddings model and store the results in a vector database.
Try this lesson free or read the step-by-step guide.
Common confusions
Embeddings vs keywords
Keyword search matches exact words. Embeddings match meaning, so "leaky faucet" and "dripping tap" are recognized as close. Many search systems combine both.
Embeddings vs the AI's answer
An embedding is not text a person reads. It is a numeric fingerprint used to find relevant material, which a large language model then reads to write the answer.
Tips
- For a few pages of reference material, pasting it or uploading a file is simpler than building embeddings.
- If you build document search, keep source documents clean and current; embeddings of outdated text find outdated answers.
Related terms
Where to learn more
Frequently asked questions
- Do I need embeddings to use ChatGPT with my documents?
- Not directly. Uploading a file to a chat or a Custom GPT lets ChatGPT handle the searching for you. You only work with embeddings yourself when you build your own document search.
- Are embeddings the same as tokens?
- No. Tokens are the chunks of text a model reads. An embedding is a list of numbers describing what a whole chunk of text means.