LLM · Large language model
An LLM (Large Language Model) is a language model trained on very large corpora, capable of generating text. ChatGPT, Gemini, Claude and Perplexity all rely on one or more LLMs, often paired with web search.
Standalone definition · 35 words
How does an LLM work?
The model predicts the next word from context, using a Transformer-type architecture (Vaswani et al., 2017). Training on vast corpora encodes knowledge in its parameters: this is parametric memory. At query time, it generates an answer token by token.
LLM and answer engine: what is the difference?
A standalone LLM answers from its memory. An answer engine pairs it with a search step (RAG, grounding) to retrieve sources and produce citations. For GEO, this second layer is what matters: what the engine retrieves and cites.
Which limits should you know?
Hallucinations, knowledge cutoff, sensitivity to prompt wording and variability from one run to the next. These limits explain why GEO measurement is done in repeated series and why retrieved sources matter.
Frequently asked questions
Are LLM and generative AI the same?
LLMs are a family of generative AI models, centered on text.
Does an LLM read my site?
It may have seen it during training or retrieve it live through an associated search engine.
Which LLM should I target for GEO?
All those that matter for your audience, measuring per engine.
Related terms
Further reading
- WebMCP : what is this new standard SEO/GEO for visibility in the LLM ?
- MCP Chrome DevTools & MCP Google Analytics : essential for GEO and LLMs
- Visibility on AI (Google mode AI, AI overviews, LLM) : the tools are lying to us ?
- ← Back to the GEO glossary (114 terms)