What is an embedding? Turning meaning into numbers
An embedding turns text or an image into a list of numbers that keeps its meaning. Similar content gets similar numbers.
What is an embedding?
Computers cannot grasp meaning directly; they work with numbers. An embedding turns a piece of text (or an image or audio clip) into a list of hundreds or thousands of numbers, called a vector. The model that does this is an embedding model. The individual numbers mean nothing to a person. What matters is where vectors sit relative to each other.
How does meaning become numbers?
Think of it as a map. The embedding model gives every text a position in a space with many dimensions, and texts with similar meaning land close together. "Terminating a lease" and "ending the agreement with my landlord" share no words but sit side by side, while "rent" and "bank interest" stay far apart.
Similarity between two vectors is usually measured with cosine similarity. The closer the score is to 1, the closer the two pieces of content are in meaning.
Where are embeddings used?
- Semantic search: finding the right document even when it does not contain the user's exact words.
- RAG: choosing the document passages that best fit a question and handing them to a language model.
- Recommendations: finding similar products and similar content.
- Clustering and classification: grouping thousands of customer reviews by topic or spotting repeated support requests.
AI powered search also matches content by meaning, which is why pages that answer a question clearly and directly are picked more easily. That is the logic behind GEO work.
Where are vectors stored?
For a few hundred documents a simple file is enough. Once you reach millions, a vector database is used to find the nearest vectors quickly.
One caution: queries and documents in the same system must be embedded with the same model. Vectors from different models live on different maps and cannot be compared.