A large language model (LLM) is software that has learned patterns from a very large amount of text and uses them to predict what text should come next. ChatGPT, Claude and Gemini are products built on LLMs. Google's own course describes a language model as one that "estimates the probability of a token or sequence of tokens occurring within a longer sequence of tokens", where a token is a word, part of a word or a single character.
Why it matters for your business
An LLM is the engine inside almost every AI tool you are being sold this year: the chatbot, the "AI assistant" in your CRM, the email writer. Knowing what the engine can and cannot do tells you which pitches are real.
It is very good at language work: drafting replies, summarising long documents, pulling fields out of messy emails, translating, rewriting in a different tone. It is not a database. It does not look facts up unless someone builds that in, and it can state something false with full confidence. That is why the useful question is never "can AI do this?" but "what does it read, and who checks it?"
An example
A property brokerage gets dozens of enquiry emails a day, each written differently. An LLM can read every one and return the same five fields: name, budget, area, property type and move-in date. Someone used to copy those by hand. The LLM does not decide which lead is worth a call. It turns messy text into tidy data, and a person or a simple rule takes it from there.
How to use it
Start with jobs where the input is text, the output is text and a mistake is cheap to catch:
| Good first use | Why it works |
|---|---|
| Drafting replies a person approves | A human checks before it goes out |
| Summarising calls, documents or tickets | Easy to compare with the original |
| Extracting fields from emails or forms | Output can be checked against rules |
| Answering FAQs from your own documents | Answers are tied to text you control |
Keep it away from anything where a wrong answer costs money or trust and nobody reviews it, such as quoting prices or promising refunds. For a longer, plain explanation of tokens, the context window and why LLMs get things wrong, read our guide What is an LLM?. When you want one working inside your own systems, that is what we build as custom AI agents.
Frequently asked questions
Is ChatGPT an LLM?
Not exactly. ChatGPT is a product, and the LLM is the model underneath it. The product adds a chat screen, memory, file uploads and safety layers on top of the model.
Can an LLM learn from my business data?
It can use your data without being retrained on it. The usual method is to hand the model the right documents at the moment it answers, which is called retrieval-augmented generation (RAG). Retraining a model is rarely needed for a small business.
Which LLM is best for a business?
There is no single best one, and the rankings change every few months. Test two or three models on your real examples, then compare accuracy, speed, cost per use and where your data is processed.
Last reviewed 22 Oct 2026 by the Astrafolio team.