Glossary

What is Large Language Model (LLM)?

The neural networks behind ChatGPT, Gemini, and Claude — trained on web-scale text, now mediating how buyers discover brands.

Definition

A large language model is a neural network trained on web-scale text to predict and generate language. GPT-5, Gemini, and Claude are LLMs; ChatGPT, AI Overviews, and Perplexity are products wrapping them. LLMs answer from two sources: parametric memory (patterns absorbed in training) and retrieval (live documents fed at answer time).

Why it matters

LLMs are becoming the interface between buyers and the web: AI platforms process 3.5B+ queries weekly, and half of B2B software buyers start vendor research in a chatbot (G2, 2026). For marketers the operational takeaway is the two-source model — what the model remembers about you (shaped over years by corroborated coverage) and what it retrieves about you (shaped this week by crawlable, answer-shaped pages). GEO works both.

Frequently asked

How do LLMs decide which brands to mention?

A blend of training-data association (which brands the corpus links to which problems) and retrieval-time evidence (which pages answer the prompt credibly). Corroboration across independent sources moves both.

Do different LLMs see my brand differently?

Often, yes — different training corpora, cutoffs, and retrieval stacks. That's why visibility tracking must run per-assistant rather than assuming one score.

Related terms