Large language model (LLM)
A large language model is a program trained on very large amounts of text to predict what text should come next, which is what lets it hold a conversation.
How it works, in one paragraph
A language model reads what has been said so far and produces a likely continuation, a small piece at a time. Answering, drafting, summarising and translating all come from that ability. Newer models can work through a problem in steps before they answer, but they still produce text, and what they know comes from training plus whatever they are given with the question.
What it means in practice
It sounds equally sure when it is wrong. The same process produces fluent text and accurate text, so a wrong answer reads as confidently as a right one. You can’t tell them apart by reading. This is where hallucination comes from, and why a business chatbot needs its answers tied to your own content through retrieval.
It doesn’t know your business. Your prices, stock, bookings and policies were not in its training data. Everything specific to you has to be supplied at the moment of the question.
It can’t act by itself. A model writes text. It doesn’t check an order, take a deposit or book a courier. For that, something has to connect it to the systems holding your orders, calendar and payments, and decide what it is allowed to do there. See chatbot vs AI agent.
Do customers accept its answers?
Mostly, if the answers are useful. In a Kantar survey of 11,056 adults in 22 markets, commissioned by Meta (fieldwork April to September 2025):
- 73.3% prefer messaging when contacting a business.
- 67.7% say a response from an AI chatbot is helpful.
Source: Meta, State of Business Messaging. Meta sells business messaging, so read it as vendor data.
Terms you will meet next to it
- Token. The small piece of text the model works in. Usage and price are counted in tokens, not words.
- Context window. How much text the model can hold at once. A very long conversation eventually falls out of it.
- Prompt. The instructions and material given to the model with the question.
- Fine-tuning. Further training on your own examples. It reliably changes style and format. It is a poor way to teach facts, because your facts change and the training doesn’t.
Does it matter which model you use?
Less than vendors suggest. Models change every few months, and the leading ones are close for the job of answering a customer about an order. The result depends more on what the model is given to work with and what it is permitted to do. That is the work done around the model, and it is what you are choosing when you choose a tool.
Related terms
Give the model your business to work with
Start your 7-day free trial. No credit card needed. Connect your website, products and policies, and AeroChat answers customers from your own content, with your limits applied.
On Shopify? Install AeroChat from the Shopify App Store.