An LLM (Large Language Model) is a large language model trained on massive volumes of text, capable of understanding and generating natural language. The main LLMs are GPT-4 (OpenAI/ChatGPT), Gemini (Google), Claude (Anthropic), Mistral and LLaMA (Meta).
An LLM learns the statistical patterns of language from billions of documents. It doesn't "understand" in the human sense: it predicts the most probable word in a given context. But this prediction capability is advanced enough to produce coherent, structured and often accurate responses.
For queries requiring up-to-date information, LLMs use RAG: they retrieve web sources in real time before generating their response.
LLMs are the engines behind ChatGPT, Google AI Overviews, Perplexity and Claude. In GEO, the goal is to optimize your content so these models select and cite it. LLMs don't rank pages like Google - they synthesize, cite and recommend, which fundamentally changes the rules of visibility.
On average, an LLM cites only 2 to 7 sources per response, compared to 10 results displayed by Google. Competition to be cited is therefore more intense, but the value of each citation is higher.