What is an AI chatbot?

"Chatbot" is an old word that now means two very different things, and the confusion causes a lot of people to dismiss AI before they have tried it.

The old chatbots

For years, "chatbot" meant the annoying box on a company website. You typed "I want to cancel my order," it replied "I didn't understand. Did you mean: Track my order?" and you started hunting for a phone number.

Those bots worked from a fixed script. Someone wrote a list of expected questions and matching answers. Say anything outside the list and the bot was lost. They were not intelligent in any sense. They were menus pretending to be conversations.

The new chatbots

An AI chatbot is something else entirely. It is a program built on a large language model (see What is AI?) that has read an enormous amount of text and learned the patterns of language well enough to hold a genuine, open-ended conversation about almost anything.

There is no script. You can ask it about medieval history, then ask it to help write a birthday poem, then paste in a spreadsheet formula that is not working, then ask it to explain the poem in French. It follows along.

The three you will hear about most are ChatGPT (from OpenAI), Claude (from Anthropic) and Gemini (from Google). Each has its own article on this site. They are more alike than different, and everything on this page applies to all of them.

How it works, in one paragraph

When you type a message, the model reads everything in the conversation so far and predicts what a helpful reply would look like, one word at a time, based on the patterns it learned in training. It is not searching a database of answers. It is generating a new answer on the spot. That is why it can respond to questions nobody has asked before, and also why it can be wrong: it is producing plausible text, and plausible is not the same as true.

What it is good at

• Explaining. Any topic, any level, as many times as you need, with no impatience.
• Writing and rewriting. Emails, letters, summaries, outlines, cover letters, speeches, captions.
• Summarising. Paste in a long text and ask for the key points.
• Brainstorming. Lists of ideas, names, angles, counter-arguments.
• Structured thinking. "Help me decide between these two apartments. Ask me questions."
• Language. Translation, grammar, tone, "does this sound rude?"
• Code. If you ever need a spreadsheet formula or a little script, it is remarkably good.
• Patience. It will happily answer the same question phrased five different ways.

What it is bad at

• Facts it did not learn. Anything that happened after its training, anything obscure, anything about your specific situation that you have not told it. Many chatbots can now search the web to fill this gap, but not always, and not always well.
• Precise numbers and citations. It may invent a statistic or a source that looks real. Always verify these.
• Knowing when it is wrong. It sounds equally confident whether it is right or wrong. The confidence tells you nothing.
• Remembering you. By default each new conversation starts fresh. Some tools now offer memory features, but do not assume it knows what you said last week.
• Judgement calls that are really yours. It can lay out the options for a medical, legal, or financial decision. It should not make the decision.

How to have your first conversation

Pick one of the three tools, open it, and try these, in order:

1. Ask it something you actually want to know. Not a test question. A real one. "Why does my sourdough come out flat?" "What does 'escrow' mean?" Notice how it answers.
2. Push back. "That's too technical, simplify it." Or "Give me an example." Or "I don't think that's right, are you sure?" A conversation, not a search.
3. Give it something of yours. Paste in an email you need to reply to and say "Help me write a polite but firm reply." Watch what it does with context.
4. Ask it to ask you questions. "I want to plan a small vegetable garden. Ask me what you need to know first." This flips it into an interviewer, which is often more useful.

Ten minutes of that will teach you more than an hour of reading.

A few habits worth building early

Give context. "Write a thank-you note" gets a generic result. "Write a short, warm thank-you note to my neighbour who fed my cat for a week; we're friendly but not close" gets a good one. The model can only use what you tell it.

Say what you want it to be. "Act as a strict editor." "Explain like a patient teacher." "Be brief." It adjusts.

Treat the answer as a draft. Read it. Edit it. Make it yours. Sending AI text unread is how people end up with embarrassing mistakes in important emails.

Keep private things private. Do not paste passwords, bank details, or other people's personal information. Assume the company might store what you type.

Don't argue with it about facts. If you suspect it is wrong, check somewhere else. It may "agree" with you just to be agreeable, which proves nothing.

The short version

An AI chatbot is a conversation partner that has read more than any human could, thinks faster than any human can, and has no idea when it is wrong. Used with that in mind, it is one of the most useful tools you will ever have access to. Next, let's meet the three big ones, starting with the one that started it all.