Lesson 8 of 28 · Language models and chat assistants

What is an AI chatbot?

An AI chatbot is software that holds a conversation with you in text or voice and uses artificial intelligence to understand your message and write a reply. Older chatbots followed scripts and menus; modern ones use large language models, so they can answer open-ended questions, follow instructions and keep track of the conversation so far.

From scripts to language models

Chatbots are older than modern AI. ELIZA, written by Joseph Weizenbaum and described in 1966, matched keywords in what you typed and turned your statements back into questions. Since then, chatbots have gone through three broad generations:

Rule-based: menus, buttons and scripted decision trees ("Press 1 for billing"). Predictable, but stuck as soon as you go off-script.
Intent-based: machine learning sorts your message into a known intent, such as "track my order", and returns a prepared answer.
LLM-based: a large language model writes each reply, so the bot can handle open-ended questions, follow instructions and change topic mid-conversation.

How a modern AI chatbot works

Each time you send a message, the chatbot app assembles a package for the language model: its instructions (a system prompt that sets the bot's role and rules), the conversation so far, and often relevant passages from the company's own documents, found through a technique called retrieval-augmented generation (RAG). The model writes its reply from that package.

Some chatbots can also call tools, such as looking up an order or booking a meeting. Their "memory" of your conversation is usually just the earlier messages being sent again with each turn, which only works while they fit inside the model's context window.

Chatbot, assistant or agent?

The words overlap, but a useful rule of thumb is this. A chatbot mainly talks. An AI assistant is a general-purpose chatbot that helps with many kinds of task, as ChatGPT does. An AI agent goes further: it takes actions on its own across several steps, such as researching a question, filling in a form and sending you the result.

What a chatbot is not

A chatbot is not a person, and you should be told when you are talking to one. The EU AI Act, for example, requires AI systems that interact directly with people to be designed so that people know they are dealing with AI, unless that is obvious.

A chatbot is also not automatically correct. A support bot that has not been connected to accurate company data may guess, and a guess can sound exactly like official policy. Good business chatbots answer from approved sources and hand over to a human when they are unsure.

Where you meet chatbots

You meet them on company websites and in banking apps, in messaging apps such as WhatsApp and WeChat, on internal HR and IT help desks, and as general assistants such as ChatGPT. For a business choosing one, the questions that matter are which channels it supports, whether it answers from your own documents, how it hands over to a human and how it treats customer data.

Example

The same question, two generations of chatbot

Worked example

Customer: "My parcel says delivered but it isn't here." Rule-based bot: "Please choose an option: 1) Track order 2) Returns 3) Talk to an agent" LLM chatbot connected to order data: "Sorry about that. Your parcel was marked as delivered at 2:14 pm today. Couriers sometimes leave parcels with a neighbour or in a safe spot. If it hasn't turned up by tomorrow, I can open a missing-parcel claim for you. Would you like me to do that?"

The LLM chatbot can quote the delivery time only because the business connected it to its order system. On its own, a language model would not know that detail, and it might even invent one.

Key points

An AI chatbot converses in text or voice and uses AI to understand and reply.
Chatbots evolved from scripted menus to large-language-model assistants.
A chatbot's "memory" is usually the conversation sent again with each message.
Business chatbots need access to company data to give specific, correct answers.
People should be told when they are talking to a bot rather than a human.

Try it on Mindber

Browse AI chatbots

Find customer-support chatbots

Product Match: describe your use case

Next lesson: What is a token in AI?

Frequently asked questions

What is the difference between a chatbot and ChatGPT?

ChatGPT is one example of an AI chatbot: a general-purpose assistant. "Chatbot" is the broader category, covering everything from a scripted help-desk widget to a support bot that answers from a company's documents. Many business chatbots use the same kind of large language model as ChatGPT behind the scenes.

Can an AI chatbot replace customer service staff?

It can answer common questions instantly and at any hour, which shortens queues. Complex, sensitive or unusual cases still need people, and good chatbots hand over to a human with the conversation history attached. Treat a chatbot as the first line of support, then check where it fails.

Is it safe to share personal information with an AI chatbot?

It depends on the provider. Read the privacy policy and data-use terms to see whether conversations are stored, used for training or reviewed by staff. Avoid entering passwords, ID numbers or confidential business data into tools your organisation has not approved.