What is a large language model (LLM), in plain terms?

Ascent Vision · Published October 24, 2026

A large language model, or LLM, is software trained on a huge amount of text to predict which words are likely to come next, which lets it write, summarize and answer questions in plain language. LLMs are the basis of chatbots such as ChatGPT, Claude and Gemini. Because an LLM predicts likely wording rather than looking facts up, it can sound certain and still be wrong, and the same question can get a different answer each time.

A plain definition

Google Cloud defines a large language model as a statistical language model, trained on a massive amount of data, that can be used to generate and translate text and other content. Wikipedia's entry describes it as a machine learning model trained on a vast amount of text, and says LLMs are the basis for many modern chatbots, such as ChatGPT, Claude and Gemini. Google says it uses LLMs to power its Gemini app.

"Large" refers to scale. Google's machine learning course says the Transformer is the most widely used design for building LLMs, and that Transformers contain hundreds of billions or even trillions of parameters. As a plain gloss, not the course's words, parameters are the internal values a model adjusts as it learns.

How it writes an answer

Google's Gemini overview says training teaches a model to pick up patterns in language and predict the next probable word: after "peanut butter and", jelly is likelier than shoelace. Google's course puts it plainly: "LLMs are essentially autocomplete mechanisms" that can predict thousands of tokens, which may be words, parts of words or single characters. The course suggests thinking of a user's question as the given text, and of the answer as what the model predicts should follow.

Why it can be wrong, and why answers vary

Google says that because LLMs work by predicting the next word or words, they are not yet fully capable of telling accurate from inaccurate information on their own, and that they are "neither information databases nor deterministic information retrieval systems". It adds that, like all LLMs, Gemini can sometimes confidently and convincingly give inaccurate or misleading responses. It also describes data voids: subjects with too little reliable information for a model to learn from, which can lead to low-quality or inaccurate answers.

Variation is built in. Google says models are often given room to pick slightly less probable words, and that the same prompt will not necessarily get the same response twice. Wikipedia adds that software around an LLM can give it memory and tools such as web search. The related answers on why ChatGPT gives different people different answers and on where ChatGPT gets its information cover those layers, and the related answer on whether AI assistants get facts wrong covers checking what they say about you.

What this explains for an owner

A model that predicts from patterns can fill a gap with a confident guess. As general advice, plain and current facts about what you do, where and for whom, stated on your own website, give it something accurate to work from. How customers use these assistants, and the work of being named in their answers, are covered in the related answers on how customers find businesses through AI assistants and on what GEO is.

Sources

GEO: recommended in AI answers

When a customer asks ChatGPT, Perplexity or Google's AI who to call, the answer names a few businesses. GEO is the work of making yours one of them.

See the service Work with us