A person typing on a laptop keyboard at a desk

What Is an AI Chatbot? How It Works, Types and Risks

Updated November 21, 2026.

An AI chatbot is a computer program you talk to in everyday language, by typing or speaking, that uses artificial intelligence to understand your message and write a reply. Older chatbots could only pick answers from a fixed script, while modern AI chatbots such as ChatGPT, Gemini and Copilot generate new replies with large language models. That makes them far more flexible, but it also means they can sound confident and still be wrong.

This guide explains how AI chatbots work, how they differ from rule-based bots and AI agents, what people and businesses use them for, and a checklist for using any chatbot safely. If you only want to pick one, see our best free AI chatbots compared side by side.

What is an AI chatbot, in simple words?

"Chatbot" is short for chat robot: software that holds a conversation with you. You will see the same idea called AI chat, a chatbot online, an AI bot, a virtual agent or conversational AI.

Those words overlap but are not identical. IBM describes conversational AI as the broad family of technologies that let computers understand and respond to human language, and a chatbot as one application of it. In the same way, every AI chatbot is a chatbot, but not every chatbot uses AI. The "Is this order delivered? Press 1 for yes" bot on an old help page is a chatbot too, just a very simple one.

Today, when people say "AI chatbot", they usually mean a general assistant such as ChatGPT, Google Gemini, Microsoft Copilot, Claude or Meta AI, or a customer-service bot on a company website that is powered by the same kind of technology. For the bigger picture of what counts as artificial intelligence, start with our guide what is AI?

How does an AI chatbot work?

Whether a bot follows rules or uses AI, every conversation goes through the same basic loop. Here is what happens inside a modern AI chatbot when you send a message:

  1. You send a message. You type, speak or upload something (a photo, a PDF). Voice is turned into text first.

  2. The text is broken into small pieces called tokens. These are words or parts of words the model can work with.

  3. A large language model predicts a reply. The model was trained on huge amounts of text and learned patterns in how language works. It writes its answer one token at a time, choosing what is most likely to come next given your message and the earlier conversation.

  4. Extra tools may be added. Many chatbots can now search the web, read your files, pull answers from a company's help articles or remember facts you shared before. This is how a support bot can quote a store's actual returns policy instead of guessing.

  5. Rules and safety filters shape the output. The company behind the bot sets instructions about tone, topics it should refuse and when to hand you to a human.

The key point: an AI chatbot predicts text that sounds right. Unless it is connected to a trusted source, it does not look facts up. Even OpenAI's own launch announcement for ChatGPT on November 30, 2022 listed "plausible-sounding but incorrect or nonsensical answers" as a known limitation. That is why checking matters (see the checklist below).

A person's hands typing a message on a smartphone

Rule-based chatbots vs AI chatbots vs AI agents

These three terms get mixed up constantly, including by the companies selling them. Here is the plain-English difference:

  • Rule-based chatbot: follows a script or decision tree written by people. It matches keywords or menu choices to pre-written answers. Predictable and cheap, but it gets stuck the moment you ask something it was not built for.

  • AI chatbot: understands free-form language and writes its own replies with a language model. It copes with typos, follow-up questions and unusual requests, but it can make mistakes and needs good guardrails.

  • AI agent: an AI system that can also take actions toward a goal, such as looking up your account, filling in a form or booking a slot, often over several steps. Many agents have a chat window, but conversation is only part of what they do.

In practice the lines are blurring. IBM notes that many platforms now combine all three, and Salesforce describes AI chatbots as one kind of AI agent built mainly to talk with people.

A worked example: one question, three kinds of bot

To make the difference concrete, imagine you type this into an online shop's help window: "My parcel says delivered but it isn't here, and I'm away next week. What can I do?"

  • A rule-based bot spots the word "parcel" and shows a menu: Track order, Start a return, Contact us. It cannot deal with the "away next week" part, so you click around until you find a phone number.

  • An AI chatbot understands both parts of your message. It might suggest checking with neighbors or a safe place, explain that you can report a missing parcel, and offer to draft the message for you. But if it is not connected to the shop's systems, it can only give general advice, and it might describe a refund policy that the shop does not actually have.

  • An AI agent connected to the order system could look up your order, confirm the delivery scan, open a missing-parcel claim and ask whether to hold any redelivery until you are back, checking with you before it does anything.

The more a bot can do, the more useful it is, and the more it matters that it is accurate and asks before acting.

The three kinds of chatbot compared:

Type

How it works

Example

Rule-based chatbot

Follows scripted menus and keywords

Simple "press 1 for billing" website bots

AI chatbot

Uses a language model to understand and write replies

ChatGPT, Gemini, Claude

AI agent

A chatbot that can also take actions with tools

Booking, searching or filling in forms for you

Where did chatbots come from?

Chatbots are older than most people think:

  • Mid-1960s: ELIZA. Joseph Weizenbaum built ELIZA at MIT and described it in a 1966 paper. It used simple keyword patterns to respond roughly like a psychotherapist. Even so, some users confided in it as if it understood them, which gave us the term "ELIZA effect": projecting human qualities onto a machine.

  • 2010s: voice assistants. Assistants such as Amazon's Alexa (launched in 2014) brought spoken chatbots into homes, mostly for short commands like timers and music.

  • November 30, 2022: ChatGPT. OpenAI released ChatGPT, which could answer follow-up questions, admit mistakes and challenge incorrect premises. It kicked off the current wave of AI chatbots. Read more in what is ChatGPT?

  • Today: the leading chatbots can search the web, see images, talk in real time and, increasingly, act as agents that carry out tasks.

What are AI chatbots used for?

For individuals, the most common uses are:

  • Getting quick explanations of a topic, in simpler words if you ask

  • Drafting and editing emails, letters, cover letters and posts

  • Studying with summaries, flashcards and practice quizzes (see how to use AI to study)

  • Translating text and messages

  • Planning trips, meals, budgets and schedules

  • Brainstorming ideas when you are stuck

For businesses, AI chatbots mostly handle:

  • Customer service around the clock: order status, returns, password resets and frequently asked questions, with a handover to a person for harder cases

  • Sales and booking: answering product questions and taking reservations

  • Internal help desks: answering staff questions about IT, HR policies or benefits

  • Collecting feedback through short conversational surveys

If you want to use one well day to day, our step-by-step guide to using an AI assistant covers the basics.

A customer support worker typing on a laptop at a desk

What are the risks of AI chatbots?

AI chatbots are useful, but they come with real limits:

  • Wrong answers. Language models can invent facts, sources, prices or policies. This is often called hallucination. It is most dangerous when the answer sounds confident.

  • The company is still responsible. In February 2024, a Canadian tribunal (the British Columbia Civil Resolution Tribunal) ruled in Moffatt v. Air Canada that the airline had to compensate a customer whose chatbot gave him wrong information about bereavement fares. Air Canada argued the chatbot was responsible for its own actions; the tribunal rejected that and awarded about CA$650 plus interest and fees. The lesson for businesses: you own what your bot says.

  • Privacy. What you type may be stored, reviewed by people or used to improve the model, depending on the service and your settings. Read our guides on turning off ChatGPT memory and deleting your ChatGPT history for one example of the controls available.

  • Scams. Fake "support" chats and bots on look-alike websites may ask for passwords or card details. A real company's chatbot will not need your full password.

  • Over-reliance. The ELIZA effect is stronger than ever with fluent AI. A chatbot is not a friend, therapist or doctor, even when it sounds caring.

  • Children. General chatbots set a minimum age in their terms, and the rules differ between services; some add extra protections or parental controls for teens. Parents should check the age rules and settings of each app their child uses.

Checklist: how to use any AI chatbot safely

Run through this before you rely on a chatbot's answer:

  • Never share secrets. No passwords, full card numbers, passport or Social Security/National Insurance numbers, or private medical records in a general chatbot.

  • Check the source for anything important. Prices, deadlines, laws, medical or financial facts: confirm them on the official website.

  • Ask for sources, then open them. If the bot cannot point to a real page that says the same thing, treat the claim as unverified.

  • Screenshot promises from support bots. If a company's bot tells you about a refund, fee or policy, save a copy with the date.

  • Ask for a human when you need one. Typing "speak to a person" or "human agent" works on many support bots.

  • Check the data settings. Look for options to turn off chat history, memory or model training.

  • Treat health, money and legal answers as a starting point only. Talk to a qualified professional before making decisions. If you are struggling, contact a doctor or a helpline, such as 988 in the US or Samaritans on 116 123 in the UK.

How do I start using an AI chatbot?

You can try one in a couple of minutes:

  1. Pick a well-known free chatbot. Our comparison of the best free AI chatbots shows the main options and their limits.

  2. Open it on the official website or app. Avoid look-alike apps and sites.

  3. Ask a clear question with context. For example: "I'm a beginner. Explain what a credit score is in 5 short bullet points, with one UK and one US example." Our guide on how to write AI prompts has more examples.

  4. Follow up. Ask it to simplify, shorten, give examples or explain one point in more detail.

  5. Check before you act. Use the checklist above, and try the same question in a second chatbot if the answer matters. Our 6 tests for comparing AI chatbots show how.

Chatbots have gone mainstream: about half of US adults now use AI chatbots, according to a Pew Research Center survey from February 2026, and Ofcom reported that 54% of UK adults use AI tools such as ChatGPT, Copilot or Gemini.

Frequently asked questions

Is ChatGPT an AI chatbot?

Yes. ChatGPT is a general-purpose AI chatbot made by OpenAI. Gemini (Google), Copilot (Microsoft), Claude (Anthropic) and Meta AI are other well-known examples.

Are AI chatbots free?

Many have a free version with usage limits. Paid plans usually add more usage, newer models and extra features. Prices change often, so check the provider's pricing page.

Can an AI chatbot replace a human customer service agent?

For simple, repetitive questions, often yes. For complex, emotional or unusual problems, a good service still hands you to a person, and the business remains responsible for what its bot tells you.

What is the difference between a chatbot and conversational AI?

Conversational AI is the underlying technology that lets computers understand and respond to language. A chatbot is one product built with it; voice assistants are another.

Is it safe to talk to an AI chatbot?

It is generally safe for everyday questions if you use an official app, avoid sharing sensitive personal data and double-check important answers.

3 tips before you go

  1. When a customer service bot can't help, type "speak to a person" or "agent"; most systems will hand you over.

  2. Never share card numbers, passwords or one-time codes in a chatbot, even one that looks official.

  3. Test any chatbot with a question you already know the answer to before trusting it on something new.

Related reading

Sources

How we write: see how we test and our editorial policy.

AICYBO is an independent website. We are not affiliated with, endorsed by or sponsored by OpenAI, Google, Microsoft, Anthropic, Meta, Amazon, IBM, Salesforce, Coursera or Air Canada. Product names belong to their owners.

Written and checked under our editorial policy. Spotted a mistake? Tell us and we will fix it.