# What is an LLM (large language model)?

> An LLM is an AI model trained on huge amounts of text so it can read and write language. What it does in ad work, where it fails and a worked example.
- **Author**: Marxx
- **Published**: 2026-11-20
- **Category**: AI
- **URL**: https://marxx.ai/library-glossary/llm

---

An LLM (large language model) is an AI model trained on huge amounts of text so it can read and write language, as in chatbots such as ChatGPT. You type a request, and it predicts a useful reply one piece of text at a time.

## What it does in ad work

- Writes first drafts of hooks, headlines and scripts
- Sums up long text, such as 500 customer reviews
- Sorts and labels things, such as grouping comments by topic
- Answers questions about data you give it

## What it does not know

An LLM does not know your ad account unless you connect it. It does not know today's prices or your last campaign unless you paste them in. It can also state something wrong in a confident voice. Always check names, numbers and claims.

## A worked example

Say you paste 40 reviews of your face wash and ask for the five most common reasons people bought. The LLM returns five reasons in a minute.

Now check it. You count by hand:

| Reason it gave | Reviews that mention it |
|---|---|
| Does not dry the skin | 22 of 40 |
| Light smell | 11 of 40 |
| Good for oily skin | 9 of 40 |

The numbers are made up. The top reason appears in 22 of 40 reviews, which is 55%. That is your first hook to test. Use the LLM to find patterns fast, then confirm them yourself.

## Safe habits

- Keep customer details out of public tools unless your tool's settings allow it.
- Give it your real facts, not guesses.
- Edit the output in your own voice. See [ChatGPT prompts for Facebook ads](/templates/chatgpt-prompts-for-facebook-ads).

## Connecting it to your ads

An LLM gets far more useful when it can read your accounts. That is what an [MCP server](/glossary/mcp) does: it lets the model call tools and read data. A model that can act on tools is called an [AI agent](/glossary/ai-agent). Marxx has a hosted MCP server. Claude, or any MCP client, can read Meta and Google Ads performance through it. We make Marxx. See [the Marxx MCP server](/mcp). Some search engines use LLMs to write answers too. See [AI Overviews](/glossary/ai-overviews).

## FAQ

### Is ChatGPT an LLM?

ChatGPT is a chat product that runs on LLMs. People often use the two names for the same thing.

### What is a token?

A token is a small chunk of text, often part of a word. Many LLM tools price and limit usage by tokens.

### Can an LLM replace my media buyer?

It can draft and summarise. It cannot be accountable for your budget. Keep a person in charge of what ships and what gets spent.

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