Chatbot

What it means

A chatbot is a computer program you talk to by typing or speaking, and it talks back. The help box on an airline's website and the assistant in a banking app are everyday examples. IBM gives the plain version: "A chatbot is a software application that communicates with people through text or voice."¹

What matters most for leaders is that a chatbot answers but does not act. OpenAI makes this distinction in its guide for businesses, where it says that "simple chatbots, single-turn LLMs, or sentiment classifiers—are not agents."² If you ask a chatbot how to change a booking, it explains the steps, and you still change the booking yourself. Systems that take actions are a separate category; see [AI agent](/definitions/ai-agent).

Where the word comes from

The idea came before the word. In January 1966, Joseph Weizenbaum of MIT published a paper about ELIZA, a program that held typed conversations with people.³ Researchers writing in 2016 called it "the first chatbot ever, developed in 1966."⁴ Weizenbaum wrote the paper partly to take away its magic. Once a program like this is explained, he said, it "stands revealed as a mere collection of procedures, each quite comprehensible."³ He also noticed something about the people using it. A person in a conversation keeps up the sense of being understood "by attributing to his conversational partner all sorts of background knowledge, insights and reasoning ability."³

The word itself came later. Dictionary.com dates "chatbot" this way: "First recorded in 1990–95."⁵ In 1994, the researcher Michael L. Mauldin presented a paper at the AAAI Conference on Artificial Intelligence titled "CHATTERBOTS, TINYMUDS, and the Turing Test: Entering the Loebner Prize Competition."⁶ Dictionaries now list the shorter form, "chatbot."⁵

The meaning has changed as the technology changed. Early chatbots followed procedures their makers had written out by hand. Today, as the U.S. Federal Trade Commission put it in 2025, "AI chatbots may use generative artificial intelligence technology to simulate human-like communication."⁷

How others define it

• Dictionary.com: "a computer program designed to respond with conversational or informational replies to verbal or written messages from users."⁵
• Collins English Dictionary (as given on Dictionary.com): "a computer program in the form of a virtual e-mail correspondent that can reply to messages from computer users."⁵
• IBM: "A chatbot is a software application that communicates with people through text or voice."¹
• U.S. Federal Trade Commission, on chatbots built as companions: "AI chatbots can effectively mimic human characteristics, emotions, and intentions," and they "generally are designed to communicate like a friend or confidant."⁷

Where they agree and differ

Every definition agrees on one thing: a chatbot holds a conversation. They differ on two points. The first is whether AI is involved. The dictionaries and IBM say only "computer program" or "software application," while the FTC describes chatbots built on generative AI.¹⁵⁷ The second is how human the chatbot seems. The Collins definition pictures an email correspondent, while the FTC warns about chatbots that imitate emotions.⁵⁷

The strongest objection to our short definition is that "it answers, but it does not do the work" makes chatbots sound harmless. That objection is fair. An answer can be a commitment. The Air Canada case below shows that when a chatbot tells a customer something, the company may have to stand behind it. Our definition still holds, because the chatbot did not act. It does mean that "only answering" is not low risk.

In the news

Air Canada's chatbot, 2024. A small claims tribunal in British Columbia ruled on a case in which the airline's chatbot "gave misleading information about the airline's bereavement fares," according to The Canadian Press.⁸ The report says Air Canada "tried to deny liability." The tribunal member, Christopher Rivers, said "it should be obvious Air Canada is responsible for the information on its website," and found that the airline "did not take reasonable care to ensure its chatbot was accurate."⁸ How sure we are: we are confident about these points. They come from a wire report that quotes the decision. We could not open the decision itself.

Europe's disclosure rule, 2026. The EU AI Act "shall apply from 2 August 2026."⁹ Its Article 50 begins: "Providers shall ensure that AI systems intended to interact directly with natural persons" are designed so that people know they are dealing with an AI, unless that is already obvious.¹⁰

What it means for your organization

Your chatbot speaks for you. Customers do not separate what the chatbot says from what the company says, and the Air Canada tribunal did not separate them either.⁸ Treat a chatbot the way you would treat a new employee at the front desk. Give it accurate policies, set [guardrails](/definitions/guardrails) on what it may promise, and put [a person in the middle](/definitions/a-person-in-the-middle) who owns its answers and checks them. If you work in Europe, check that it tells people they are talking to an AI.¹⁰

One question to ask your team: Who is responsible for what our chatbot tells customers, and when did they last check its answers against our actual policies?