What Is a Chatbot?

The word combines chat, a written conversation, and bot, a program that acts on its own. What makes something a chatbot is how it is used: someone types or speaks, the program replies, and the conversation goes on in turns. Two very different techniques can sit behind it.

An early, well-documented example is ELIZA. Joseph Weizenbaum described it in Communications of the ACM in January 1966. The program looked for keywords in the user's sentence and transformed it with a rule attached to each keyword. If it found none, it replied with a content-free remark. Weizenbaum noted that some people were very hard to convince that ELIZA was not human.

A chatbot built on an LLM works differently. As the LLM entry explains, the model calculates, piece by piece, the most probable continuation of a text. The chatbot is the product around it: the interface, fixed instructions, the conversation history and, sometimes, tools that look up documents before it answers.

How it differs from an AI agent is covered in the AI agent entry. The short criterion: the chatbot answers inside the conversation, while the agent leaves it to complete a task, for example by visiting web pages at that moment.

There are three kinds of conversational system. They differ in where the answer comes from, whether the system acts, and what its typical failure is.

A rule-based chatbot answers with rules and texts written by people: keywords, decision trees or buttons with options. It does not act beyond what its script provides for. Its typical failure is the question the script does not cover: the user gets a generic line or lands back at the start menu. In return, every answer is predictable and can be reviewed before it goes live.

An LLM chatbot generates the answer with a language model, sometimes backed by documents the system retrieves before replying. It understands questions phrased in many ways, but it does not act. Its typical failure is a false statement delivered with confidence. Google explains why in its guidance on generative AI content, last updated 1 October 2026: generative models don't retrieve facts, they predict a likely sequence of words, so their output may contain inaccuracies, known as hallucinations.

An AI agent is started by a user to complete a specific task, and during that task it visits web pages in real time: it compares, looks things up or fills in forms. Its answer is written by a language model using data from the pages it visits during the task. It is the only one of the three that acts outside the conversation. Its typical failure depends on the website: if the site blocks it, or a form needs JavaScript the agent doesn't run the way a browser does, the task is left half done.

The first reason is legal. Article 50 of the EU Artificial Intelligence Act, Regulation (EU) 2024/1689, has applied since 2 August 2026. Under paragraph 1, providers must design AI systems intended to interact directly with people so that those people are informed they are interacting with an AI system, unless that is obvious from the circumstances. Under paragraph 5, the information must be clear and given at the latest at the first interaction. A provider is whoever develops the system, or has it developed, and places it on the market or puts it into service under its own name or trademark.

A purely rule-based chatbot may fall outside this. According to recital 12, the notion of an AI system should not cover systems based on rules defined solely by people. It depends on how the chatbot is built.

The second reason is website content. Google's guidance warns that generating many pages without adding value for users may violate its scaled content abuse policy, and asks site owners to fact-check everything generated by hand before publishing it.

The third is search itself. AI Overviews and AI Mode answer with AI-generated text inside Google. According to Google's documentation, last updated 10 December 2025, there are no additional requirements to appear in them: the page must be indexed and eligible to be shown with a snippet, and the traffic is counted in Search Console under the "Web" search type.

• Tell people at the start of the conversation that they are talking to an automated system.
• Use rules when a small set of answers has to be exact, and an LLM when questions vary widely.
• Restrict an LLM chatbot to your own reviewed documents and show where each answer comes from.
• Always offer a way to reach a person.
• Check by hand every text written with a chatbot before publishing it, including the title, meta description, structured data and alt text.

• Using chatbot and AI agent as synonyms, when only the agent acts outside the conversation.
• Taking a figure, date or quote from a chatbot at face value without checking the source.
• Publishing pages written by a chatbot in bulk without adding anything of your own.
• Saying it is an automated system only when the user asks.
• Believing you need a special file or special markup to appear in AI Overviews or AI Mode.