# What are embeddings?

> An embedding is a list of numbers that stands for the meaning of text, an image or an ad, so software can find similar things. A plain example for ad teams.
- **Author**: Marxx
- **Published**: 2026-11-17
- **Category**: AI
- **URL**: https://marxx.ai/library-glossary/embeddings

---

An embedding is a list of numbers that stands for the meaning of a piece of text, an image or an ad. Things with similar meaning get similar numbers, so software can find "ads like this one" by comparing numbers instead of matching words.

OpenAI describes an embedding as a vector (a list) of floating point numbers. It says the distance between two vectors measures how related they are: small distance means high relatedness, large distance means low relatedness (OpenAI documentation, checked 2026-10-20).

## An everyday picture

Think of a map of a city, where shops that sell similar things sit near each other. Embeddings do this for meaning. Each phrase gets a spot on a very large map, with far more than two directions.

## A worked example

Here is a toy version with two numbers per phrase. Real embeddings use many more numbers. The values are made up.

| Phrase | Number 1 | Number 2 |
|---|---|---|
| running shoes | 0.90 | 0.10 |
| jogging sneakers | 0.85 | 0.15 |
| wedding cards | 0.10 | 0.90 |

"Running shoes" and "jogging sneakers" share no words, but their numbers are close. The distance between them is about 0.07. "Running shoes" and "wedding cards" are far apart, at about 1.13. Software reads the first pair as related and the second as unrelated.

## What OpenAI says embeddings are used for

Ranking search results, grouping similar texts, suggesting related items, spotting outliers, measuring how varied a set of texts is and sorting text into labelled groups.

## Why this matters for ad teams

- **Search by meaning.** A search for "tired skin" can find ads that say "dull face".
- **Grouping.** Thousands of ads or comments fall into themes without hand tagging. See [creative tagging](/glossary/creative-tagging).
- **Finding repeats.** Near-duplicate ads sit close together.
- **Feeding AI tools.** A tool can pick the most relevant pieces of your data for an [AI agent](/glossary/ai-agent) to read.

You do not build embeddings yourself. Tools do it behind the scenes. See also [competitive intelligence](/glossary/competitive-intelligence) and [creative intelligence](/glossary/creative-intelligence).

Marxx searches your own ads by meaning, such as "every ad that opened on a founder". We make Marxx. See [creative intelligence](/features/creative-intelligence).

## FAQ

### Are embeddings the same as keywords?

No. Keywords match exact words. Embeddings match meaning, so different words can still land close together.

### Is an embedding the same as a chatbot like ChatGPT?

No. A chatbot writes text. An embedding model turns text into numbers. A tool may use both.

### Do I need to know maths to use them?

No. You only need to know that "similar" is measured by the distance between lists of numbers.

---
- [More AI articles](https://marxx.ai/library-glossary/category/ai)
- [All articles](https://marxx.ai/library-glossary)