Large Language Model (LLM)
A large language model (LLM) is an AI system trained on vast amounts of text to predict and generate language, powering tools like ChatGPT, Claude, and Gemini.
LLMs generate answers by predicting likely sequences of words, informed by patterns learned in training and, increasingly, by retrieving current information from the web at query time. That retrieval step is where GEO applies: the clearer and more trustworthy your content, the more likely a model is to use and cite it.
For a complete, plain-English explanation — how LLMs work, from tokens and transformers to training and inference, the major models in 2026, what they get wrong, and how they are reshaping search — read our full guide: What Are Large Language Models (LLMs)?
How it works
Text is split into tokens, and a transformer network learns, across enormous amounts of text, which token is likely to come next given everything before it. Training at that scale produces a model that has absorbed grammar, facts, styles and patterns of reasoning. Further training on examples and human feedback turns it into an assistant that follows instructions.
At use, the model generates one token at a time within its context window, optionally drawing on documents or search results supplied alongside the question.
Example
When someone asks an assistant to compare two web hosts, a model with search may look up both providers' current plans, read the pages it finds and write a comparison, citing them. Without search it answers from training data, which may be out of date on prices.
Common mistakes
- Treating the model's confident tone as a signal that it is right.
- Assuming it knows about recent events or your internal documents without being given them.
- Using one model for everything instead of matching model size to the task.
