What Is a Large Language Model (LLM)?

A large language model (LLM) is a neural network — typically built on the transformer architecture — trained on vast amounts of text to predict the next token in a sequence. From that single learned skill, prediction at scale, emerge the capabilities behind modern AI assistants: answering questions, summarizing, reasoning, writing code, and recommending products. GPT, Claude, Gemini, and Llama are all LLMs.

How is an LLM built?

Training proceeds in stages. Pretraining feeds the model trillions of tokens of web text, books, code, and licensed data — GPT-3, an early reference point, had 175 billion parameters trained on roughly 300 billion tokens (Brown et al., 2020); frontier models since have grown far beyond it. Post-training then shapes raw prediction into useful behavior: instruction tuning teaches the model to follow directions, and RLHF aligns its answers with human preferences. The finished artifact is a set of weights — numbers encoding everything the model absorbed, including every brand it ever read about.

What can and can't an LLM do?

An LLM generates fluent, statistically plausible text; it does not look facts up unless connected to retrieval. Its knowledge freezes at a training cutoff, it can hallucinate confident falsehoods, and its outputs vary run to run because generation is probabilistic. Modern assistants patch these limits with tool use — web search, code execution — which is why the same model gives different answers with browsing on versus off.

Why do LLMs matter for brand visibility?

Every AI answer about your category is an LLM output shaped by two inputs you can influence: the training corpora (your durable web presence, third-party mentions, Wikipedia) and the retrieved context (your indexable, extractable pages). That two-path structure is the foundation of GEO — and the reason this glossary exists.

Example

Ask Claude, ChatGPT, and Gemini "what is the best CRM for startups" with browsing off, and each produces a different list drawn purely from what its weights encode about the CRM market as of its cutoff. Same question, three parametric memories, three answers — a controlled demonstration of why per-model tracking matters. The supporting concepts — tokens, context windows, fine-tuning — are defined throughout this glossary.