An LLM, or large language model, is software that predicts text. Feed it a question and it produces the words most likely to follow, based on patterns learned from an enormous volume of writing. ChatGPT, Gemini, Claude and Copilot are all built on one.
How a language model actually works
There is no database of facts inside it. During training, the model reads billions of documents and learns statistical relationships between words and the ways they form sentences. When you ask it something, it generates an answer one token at a time, each one chosen because it fits the pattern.
That design explains both the strengths and the failures. The fluency is real. So is the tendency to state something confidently that is not true, which the industry calls hallucination. The model is not lying. It is producing plausible text.
Why marketers need to care
Buyers increasingly ask an assistant before they search. If the assistant names three suppliers in your category and you are not one of them, you never entered the consideration set. There was no impression to measure and no click to attribute.
Two things decide whether you get named. What the model absorbed during training, which you cannot change retrospectively. And what it retrieves live from the web when it answers, which you can influence. Most modern assistants do both.
What you can influence
- Be crawlable. Assistants fetch pages when they answer. A page behind a script that never renders is a page that cannot be cited.
- Be quotable. Clear definitions, plain headings and direct answers get lifted. Marketing prose does not.
- Be corroborated. Models weight claims that appear consistently across independent sources. One page saying it is rarely enough.
This is the same discipline as ranking well, applied to a different surface. Our AI search team works on both together because separating them makes no sense.
Common questions
Is an LLM the same as AI?
No. AI is the broad field. A large language model is one type of system within it, built specifically to work with text. Image generators and recommendation engines are AI too and work nothing like an LLM.
Do LLMs use Google rankings?
Some do, indirectly. Assistants that search the web often draw on a search index to find sources before summarising them, so ranking still helps. Answers drawn purely from training data do not touch rankings at all.
How do I find out if AI assistants mention my brand?
Ask them, repeatedly, with the questions your buyers would use, and record what comes back. Tools now track this at scale, but manual checks across ten real buying questions will tell you most of what you need in an afternoon.
Short definition instead? See LLM in our glossary.
