Glossary
What is a large language model (LLM)?
A large language model (LLM) is an AI model trained on very large amounts of text to predict and generate language. ChatGPT, Claude and Gemini are built on LLMs. They answer questions, write and summarise, and, when connected to search or tools, look things up and take actions.
How they work, briefly
An LLM is trained to predict the next piece of text across a huge collection of documents. That training gives it broad knowledge and the ability to write. A further training stage teaches it to follow instructions and be helpful. When you ask a question, it generates an answer token by token.
Training knowledge vs retrieval
| Training knowledge | Retrieval | |
|---|---|---|
| Source | Text seen during training | Pages fetched at answer time |
| Freshness | Up to a cutoff date | Current |
| Citations | None | Usually shown |
| How to influence | Be widely and consistently described online | Rank, be crawlable, answer clearly |
A product launched after a model's training cutoff can only appear in its answers through retrieval. That's why search visibility and crawlability matter for new products in AI answers.
What it means for brands
Models describe products the way the web describes them. Consistent facts across your site, directories, review sites and databases make it more likely an LLM describes you correctly. Inconsistent or thin information leads to vague or wrong answers.
Example
Asked about a startup launched last month, a model without search says it doesn't know it. With search on, the same model finds the startup's directory listings and site, and describes it accurately.
FAQ
Are LLMs the same as AI chatbots?
Chatbots like ChatGPT are products built on LLMs, with extra features such as search and tools.
Do LLMs remember my website?
Only what was in their training data, up to a cutoff. Fresh information comes from retrieval.
Why do LLMs make things up?
They generate likely text, which can be wrong; that's called hallucination. Retrieval and citations reduce it.