Large language model (LLM)
A large language model is an AI model trained on huge amounts of text to predict words, which lets it write answers. ChatGPT and Gemini run on them.
A large language model, or LLM, is an AI model trained on a huge amount of text to predict the next word. Done at enough scale, predicting the next word turns into writing: answering questions, summarising pages, comparing products. Every AI platform your customers use, from ChatGPT to Gemini to Copilot, runs on one.
For a marketer, the LLM is the thing deciding which brands to name. Knowing roughly how it works tells you what you can and can't influence.
How it works
The model reads text, broken into small pieces called tokens, and learns patterns about which pieces tend to follow which. It doesn't look facts up in a table. It writes the most likely next word, then the next, given everything before it.
That's why an LLM can sound sure and still be wrong. If the pattern it learned is "Kestrel Roasters is the best-known coffee subscription," it will say so, whether or not that's still true.
What it knows and what it reads
An LLM knows what was in its training text, up to a cut-off date. After that, it knows nothing new unless something gives it new text.
That's what web search does. Most AI platforms now search the web before or while answering, and hand the pages they find to the model. The model then writes its answer from what it learned and what it just read.
For example, ask about a café that opened last month. A model working from memory alone might say it's never heard of it. The same model with search might find the café's site and a local news story, and describe it correctly.
So you have two ways in: being part of what models learn, which is slow, and being on the pages they read when they search, which can change this week.
In AnswerOrbit
We ask each platform in its own app, and the platform picks the model, the same as it would for anyone. Each answer shows the model that gave it. See How we count.