What is an LLM?
A large language model (LLM) is an AI model trained on large amounts of text to predict and generate language. Most are based on the Transformer architecture introduced in 2017. On their own they answer from training data; with web search, they retrieve pages and cite them. ChatGPT, Gemini, Claude and Grok are built on LLMs.
How does an LLM work?
LLMs are neural networks, mostly built on the Transformer architecture from “Attention Is All You Need” (first posted June 12, 2017).[1] They generate text one token at a time based on patterns learned in training.
How do LLMs use search?
Search-enabled assistants retrieve web pages and ground answers in them; see RAG.[2]
Why do LLMs make mistakes?
OpenAI researchers argue LLMs hallucinate partly because training and evaluation reward guessing over acknowledging uncertainty.[3] See hallucinations about brands.
What does this mean for visibility?
Two routes: being represented in training data, and being retrieved at answer time. Only the second is directly measurable; see LLM visibility.
Frequently asked questions
Is ChatGPT an LLM?
ChatGPT is a product built on OpenAI’s LLMs.
What does LLM stand for?
Large language model.
Do LLMs browse the web?
Some can, through search tools; otherwise they rely on training data.
See also
References
Sources accessed October 14, 2026 unless another date is given. See all sources.
- “Attention Is All You Need (Vaswani et al.)”. arXiv. v1 June 12, 2017.
- “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al.)”. arXiv. v1 May 22, 2020.
- “Why Language Models Hallucinate (Kalai, Nachum, Vempala, Zhang)”. arXiv / OpenAI. September 4, 2025.