Glossary
Large language model the model behind every AI answer.
An LLM is an AI model trained on huge amounts of text to predict and write language. How LLMs power AI search, and how Prefer tracks what they say about you.
Large language model A large language model (LLM) is an AI model trained on huge amounts of text to predict the next word, which lets it write answers. Prefer tracks what LLM-powered engines say about your brand.
A large language model (LLM) is an AI model trained on very large amounts of text to predict the next piece of text, which lets it understand questions and write fluent answers. Prefer tracks what the LLM-powered engines, ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode, say about your brand, with Claude on Enterprise. Every AI search answer your buyers read was written by one.
How an LLM works#
Text is split into small units called tokens, often pieces of words. During training the model sees billions of examples and learns to predict which token comes next. Modern LLMs are built on the transformer architecture from the 2017 paper Attention Is All You Need. The GPT-3 paper showed that scaling this up let one model handle many tasks from a few examples in the prompt.
When you ask a question, the model reads your prompt (and anything else placed in its context window) and generates an answer one token at a time.
Why LLMs matter for AI visibility#
An LLM’s memory stops at its training cutoff. What it learned from its training data is fixed until the next model. That is why AI search products add retrieval: the model searches the web, reads current pages and writes from them, a method called retrieval-augmented generation.
So a brand shows up in AI answers in two ways. It can be remembered, which is slow to change and hard to influence. Or it can be retrieved and cited, which you can work on today. In Prefer’s 2 September 2026 ChatGPT run, 47 of 50 answers used live web search, as shown on does ChatGPT use live search.
Fig. 01
Memory vs search
A concrete example#
Ask a model with search turned off “what is [a company founded last year]?” It may say it does not know, or guess. Turn search on and it can find the company’s site, read it and cite it. Same model, different answer, because of what it could read.
Common confusions#
- LLM vs chatbot. ChatGPT, Gemini and Claude are products. The LLM is the model inside, and the product adds tools like web search.
- LLM vs search engine. A search engine ranks pages. An LLM writes an answer. AI search combines the two.
- “The model knows everything online.” It does not. It learned from a sample of text up to a cutoff, and OpenAI runs separate crawlers for training (GPTBot) and for search (OAI-SearchBot).
What this means for your brand#
- Be easy to retrieve. Let AI search crawlers in and publish pages that answer buyer questions plainly.
- Be consistent everywhere. The same name, category and facts on your site and on third-party pages.
- Watch the answers, not just rankings. What an LLM writes about you is the new first impression.
How Prefer helps#
Prefer measures your LLM visibility: how often each engine mentions and cites you, the sources behind each answer, and your share of voice against named competitors. To see what LLMs say about you now, try the free AI Visibility Checker or read why ChatGPT doesn’t know your company.
In context
The term in a sentence.
Related questions
People also ask.
- What is an LLM in simple terms?
- Is ChatGPT a large language model?
- How do LLMs decide which brands to recommend?
Questions
Asked plainly.
What is a large language model in simple terms?
It is an AI system trained on a very large amount of text so it can predict the next word and, by doing that over and over, write full answers. Prefer tracks what the LLM-powered engines (ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode) say about your brand. Claude tracking is available on Enterprise.
Is ChatGPT a large language model?
ChatGPT is a product built on OpenAI's large language models, with extra tools such as web search added on top. Prefer tracks ChatGPT's answers and the sources it cites for your buyer questions. The distinction matters: the model alone answers from memory, while the product can search the web and cite pages.
How do LLMs decide which brands to recommend?
They draw on two things: what the model learned in training, and the pages it retrieves when it searches. Prefer shows the second part directly, listing the sources behind each answer and your share of voice against named competitors. You can rarely change the first quickly, but you can influence the second by being on the pages engines retrieve.
Can I get my brand into an LLM's training data?
Not directly; OpenAI, for example, does not take submissions to its training data. Prefer focuses on the layer you can influence, live search, by tracking whether engines cite you and writing content on every self-serve plan that is built to be cited. Being widely and consistently described on the open web is the indirect route into future training data.
Keep reading