What is a large language model (LLM)?

A large language model (LLM) is an AI model trained on vast amounts of text to understand and generate language. Model families such as OpenAI's GPT, Google's Gemini and Anthropic's Claude run assistants like ChatGPT, Gemini and Claude. LLM stands for large language model.

Why large language models matter for your brand

Large language models now write the answers that buyers read. OpenAI says that every week, "1 billion people turn to ChatGPT for everything from quick questions and web searches to planning, research, advice, and complex decisions." When one of those questions is about your category, the answer can name your brand, a competitor, or nobody.

What an LLM says about your brand comes from two places: what it learned in training, and what it reads on the web when you ask. The two behave very differently, and OpenAI and Anthropic use a separate crawler for each.

Training data vs web search

	Training data	Web search at answer time
When the model learns it	Before its training cutoff	When the question is asked
Crawlers involved	GPTBot, ClaudeBot	OAI-SearchBot, Claude-SearchBot
How fast changes show	With the next model version	As soon as a page is retrieved
Sources shown	Usually none	Cited links

Training data is the text a model learned from, up to a cutoff date. Anthropic, for instance, publishes a reliable knowledge cutoff and a training data cutoff for each Claude model. OpenAI's GPTBot crawls content "that may be used in training our generative AI foundation models."

Web search fills the gap. Google calls it retrieval-augmented generation, or grounding: the system retrieves "relevant, up-to-date web pages" and writes from them. ChatGPT "may search the web automatically when your question would benefit from current information."

The two are independent. OpenAI notes that a site can allow OAI-SearchBot "to appear in search results while disallowing GPTBot" for training.

Example: a brand the model has never heard of

A software company launched after a model's training cutoff. A buyer asks: "What does [brand] do, and is it any good?" Without web search, the model may say it does not know the brand or confuse it with a similar name. With web search, it reads the company's site and the reviews that rank, and answers with links. If the site blocks OAI-SearchBot, ChatGPT search will not show it in that answer. And when a new review of the product starts to rank, the searched answer can change the same week, while the trained answer waits for the next model.

How to check what LLMs say about your brand

Ask the assistants the questions your buyers ask, about your category and about you by name, and note whether each answer cites sources. Then make sure both routes can reach you:

Let search crawlers in. Allow OAI-SearchBot and Claude-SearchBot in robots.txt and in your CDN or firewall rules.
Keep key facts in the HTML. In a December 2024 study, Vercel found that none of the major AI crawlers it observed, including GPTBot, OAI-SearchBot and ClaudeBot, rendered JavaScript. Prices or features that only appear after scripts run are invisible to them.

Hawkscan shows the full answers that ChatGPT, Gemini, Microsoft Copilot and Google AI Mode give to the prompts you track, every day. See also answer engine and ChatGPT search.