What Is a Chatbot?

A chatbot is software that holds a back-and-forth conversation with people by text or voice. Older chatbots follow scripts: they offer menus or match keywords to prewritten replies. AI chatbots, such as ChatGPT, use large language models to interpret open-ended messages and generate new responses. You find both kinds on websites, in messaging apps and in customer service.

What a chatbot is

A chatbot is defined by how you use it: you type or speak, and it replies in a back-and-forth exchange. IBM describes a chatbot simply as "a software application that communicates with people through text or voice."

What happens behind that conversation varies enormously. Some chatbots can only show you a menu of buttons. Others can draft a report, explain a contract or write code.

It helps to separate the chatbot from what powers it. The chatbot is the application: the chat window, the app or the help widget you see. Modern AI chatbots run on an underlying language model, often a chat model, which is a language model adapted for back-and-forth conversation. The same model can power many different chatbots, and one chatbot can switch between models.

The two generations of chatbots

Scripted chatbots

Scripted chatbots, also called rule-based chatbots, follow paths that a person designed in advance. IBM describes traditional chatbots as following "predefined rules, decision trees and scripted conversation flows."

They typically work in one of two ways:

• Menus and buttons. You choose from options such as "Track my order" or "Change my address," and each choice leads to the next set of options.
• Keyword matching. The bot scans your message for certain words, such as "refund" or "password," and responds with the reply linked to that keyword.

The idea is decades old. ELIZA, created at MIT by Joseph Weizenbaum in 1966, is often described as one of the first chatbots. It simulated a psychotherapist by picking out keywords in a person's message and building its replies around them, with no understanding of meaning.

Scripted chatbots are predictable and inexpensive, which makes them useful for narrow, repetitive tasks. Their weakness is anything unexpected. Phrase a question in a way the designer did not anticipate, and they get stuck or repeat the same reply.

AI chatbots

AI chatbots interpret and generate language rather than following a fixed script. According to IBM, they use technologies such as natural language processing, generative AI and large language models "to understand user intent, generate more natural responses and adapt to a wider range of conversations."

OpenAI released ChatGPT on November 30, 2022, describing a model that "interacts in a conversational way" and can answer follow-up questions. Its rapid adoption made AI chatbots mainstream. A Pew Research Center survey conducted in February 2026 found that about half of U.S. adults now use AI chatbots, and 44% use ChatGPT.

AI chatbots can handle questions phrased in countless ways, keep track of the conversation and produce original text. Their main weakness is accuracy. OpenAI's launch announcement acknowledged that ChatGPT "sometimes writes plausible-sounding but incorrect or nonsensical answers," a limitation that still applies to AI chatbots generally.

Scripted vs AI chatbots at a glance

How an AI chatbot works

When you send a message to an AI chatbot, a sequence like this happens in seconds:

1. You send a message. You type or speak a question or request.
2. The app adds context. It combines your message with the earlier conversation and hidden instructions from the company that built the chatbot, known as a system prompt.
3. A language model reads it all. The model can only consider a limited amount of text at once, called its context window.
4. It may use tools. Depending on the product, the chatbot might search the web, check a company knowledge base or look up your account details before answering.
5. The model generates a reply. It produces the response piece by piece, predicting what should come next based on patterns learned during training.
6. The app delivers it. Safety filters may check the reply before you see it.

The chatbot does not look up a stored answer. It generates a new response each time, which is why the same question can produce different wording, and occasionally different facts. What Is a Large Language Model? explains the model side in more detail.

Types of chatbots

There is no single official classification, but chatbots are commonly grouped by how they work:

• Menu-based chatbots guide you through buttons and fixed options.
• Rule-based or keyword chatbots match words in your message to scripted responses.
• AI chatbots use language models to understand free-form messages and generate replies.
• Hybrid chatbots combine the two, using scripts for routine steps such as verifying your identity and AI for open-ended questions.

They are also grouped by purpose: general-purpose assistants, customer service bots on company websites, and internal assistants that answer employees' questions about company policies and documents.

Examples of chatbots

• General AI assistants: ChatGPT, Google Gemini, Anthropic's Claude, Microsoft Copilot and Meta AI.
• Customer support bots: help widgets on bank, airline, retail and utility websites that answer questions or route you to a person.
• Messaging bots: automated accounts in apps like WhatsApp or Facebook Messenger that confirm appointments or answer store questions.
• Workplace assistants: chatbots inside company tools that answer questions about HR policies, IT issues or internal documents.

Chatbot vs conversational AI

Conversational AI is the broader technology that lets computers understand and respond to human language in a natural way. A chatbot is one application of it. IBM defines conversational AI as "technologies, such as chatbots or virtual agents, that users can talk to."

The two terms overlap but are not identical:

• An AI chatbot uses conversational AI.
• A scripted chatbot usually does not. It has a conversation-style interface without any real language understanding.
• Conversational AI also powers things people rarely call chatbots, such as voice assistants and automated phone systems that understand spoken requests.

Are voice assistants like Siri and Alexa chatbots?

Siri and Alexa are usually called voice assistants or virtual assistants rather than chatbots. They use conversational AI, but they were originally designed mainly for short spoken commands, such as setting timers, playing music or controlling smart home devices, rather than extended conversations.

That line is blurring. Apple added an optional ChatGPT integration to Siri in December 2024, letting Siri pass certain requests to ChatGPT. In June 2026, Apple announced Siri AI, a more capable new version of Siri that it plans to release as a beta later in 2026. Amazon introduced Alexa+, a generative AI version of Alexa, in February 2025, and says it is now available to all customers in the U.S. and Canada. As voice assistants adopt large language models, they increasingly work like AI chatbots you talk to.

AI agent vs chatbot

A chatbot mainly talks: it answers questions, explains and writes. An AI agent acts: it takes steps in other apps and systems to complete a task toward a goal, such as booking a meeting or processing a refund.

The difference is what happens after the reply. A chatbot hands the next step back to you. An agent carries the task forward, often checking with you before important actions. Many chatbot products now include agent features, so a single app can do both. For the details, see What Is an AI Agent?. For the broader idea of AI that plans and acts, see What Is Agentic AI?.

How to tell what kind of chatbot you're using

A few quick checks reveal a lot:

• You can only click buttons or choose options. It is almost certainly scripted.
• Rephrasing your question gets the same canned reply. It is likely keyword-based.
• It understands unusual phrasing and follow-up questions. It is probably an AI chatbot.
• Replies are fluent but vague, or confidently wrong about specifics. That is typical of an AI chatbot without access to the right information.
• Look for a disclosure. Many services label automated assistants. In the European Union, the AI Act requires providers to ensure people are informed when they are interacting with an AI system, unless it is obvious, under transparency rules that apply from August 2, 2026.
• Ask for a person. Many customer service bots can hand you to a human agent on request.

Why chatbots matter, and their limits

Chatbots are now one of the main ways people encounter AI, at work and as customers. Knowing what kind you are using helps you decide how much to trust it.

• AI chatbots can be wrong. They can state incorrect information confidently. Check important facts, especially on health, legal and financial questions.
• Companies can be held responsible for their bots. In 2024, a British Columbia tribunal ruled that Air Canada was responsible for incorrect fare information its website chatbot gave a customer, rejecting the argument that the chatbot was a separate entity.
• Be careful what you share. Conversations may be stored and, depending on the product and your settings, used to improve AI models. ChatGPT, for example, lets users turn off an "Improve the model for everyone" setting.
• A chatbot only knows what it can access. Unless it is connected to your account or current information, it cannot know your order status or today's news.