What Is a Chatbot? How They Work, Types & Examples (2026)

What Is a Chatbot? How They Work, Types and Examples (2026)

A chatbot is a software application that simulates human conversation, letting people interact with a computer through text or voice as if they were messaging a person. You have almost certainly used one: the support window on a website, an automated assistant in a messaging app, or a voice assistant on your phone. At its core, a chatbot takes what you say, interprets it, and responds, automating conversations that would otherwise need a human.

This guide explains what a chatbot is, how rule-based and AI chatbots differ, the main types, real examples, how a chatbot differs from an AI agent, what they cost, where they fail, and how businesses use them in 2026.

TL;DR

• A chatbot is software that simulates conversation via text or voice.
• Two broad kinds: rule-based (follows scripted paths) and AI chatbots (understand natural language using machine learning / LLMs).
• The four types usually named are menu/button, keyword-recognition, contextual/AI, and voice.
• Examples: website support bots, messaging-app assistants, and voice assistants like Siri and Alexa.
• A chatbot responds; an AI agent goes further and takes autonomous actions to complete tasks.
• Businesses use chatbots for 24/7 support, lead qualification, and instant responses that lift conversions.

A very short history of chatbots

Chatbots are much older than the current AI wave, and knowing the lineage makes the current generation easier to understand.

• 1966 - ELIZA. Joseph Weizenbaum at MIT built a program that imitated a psychotherapist by reflecting users' own words back as questions. It understood nothing, but people confided in it anyway. Weizenbaum was unsettled enough by that reaction to spend much of his later career warning about it.
• 1995 - A.L.I.C.E. Richard Wallace's bot used pattern-matching rules written in a markup language called AIML. Still no understanding, but a far larger rulebook.
• 2011 onward - voice assistants. Siri, then Google Assistant and Alexa, brought speech recognition and intent detection to consumers at scale.
• 2016 - the messaging-bot rush. Facebook opened Messenger to bots and thousands of businesses shipped scripted flows. Most were poor, and the backlash is where the "press 1 for..." reputation comes from.
• 2022 onward - large language models. ChatGPT and its successors changed the middle step. Instead of matching patterns, the software generates responses from a statistical model of language.

The important shift is that for fifty years chatbots faked comprehension with rules. Only recently did the interpretation step get good enough that a bot could handle a sentence nobody anticipated.

How chatbots work

Every chatbot does three things: it receives your input, interprets what you mean, and returns a response. How well it does the middle step, interpreting, is what separates a frustrating bot from a genuinely useful one.

Broken down slightly further, a modern chatbot runs roughly this pipeline:

1. Input. Text you typed, or speech converted to text by a speech-recognition model.
2. Intent recognition. Working out what you actually want. "Where's my order", "order status?" and "has my thing shipped yet" are three phrasings of one intent.
3. Entity extraction. Pulling the specifics out of the sentence: an order number, a date, a product name.
4. Dialogue management. Deciding what happens next, and remembering what was already said. This is where most bots fall down, because a question only makes sense in the context of the last three messages.
5. Response generation. Either selecting a written answer or generating one.
6. Action, sometimes. Looking up an order, booking a slot, writing to a CRM. Whether a bot can do this is the line between a chatbot and an agent.

There are two broad approaches to steps 2 and 5:

• Rule-based chatbots follow pre-defined scripts and decision trees. They recognize specific keywords or menu choices and respond with set answers. They are reliable for simple, predictable tasks (store hours, order status) but break down the moment a user phrases something unexpected, because they do not truly understand language, they match patterns.
• AI chatbots use natural language processing and machine learning, today usually powered by large language models, to actually understand the intent behind what someone types, even with typos, slang, or unusual phrasing. They can hold more natural, flexible conversations, handle questions they were not explicitly scripted for, and improve over time.

Modern business chatbots are increasingly AI-powered, because rule-based bots create the "press 1 for..." frustration that drives customers away. For businesses choosing an AI-powered solution, best chatbot software can help narrow down the right options.

The trade-off nobody mentions. Rule-based bots are predictable and cannot say anything you did not write. AI bots are flexible and can say things you did not anticipate, including wrong things. Regulated industries often keep scripted flows for exactly this reason. The practical answer for most businesses is a hybrid: scripted certainty on the handful of paths that carry legal or financial weight, AI everywhere else.

The four types of chatbots

Most classifications land on four categories, ordered roughly by how much interpretation they do.

1. Menu or button-based. You click options rather than type. A decision tree with a friendly face. Reliable, easy to build, and incapable of handling anything outside the menu. Still the right answer for a narrow task like booking a table.

2. Keyword recognition. Scans your message for trigger words and responds accordingly. Better than a menu because you can type freely, worse than it looks because "I do not want to cancel" and "I want to cancel" both contain the keyword.

3. Contextual or AI chatbots. Use natural language understanding and, increasingly, large language models. They track context across a conversation, handle rephrasing, and answer questions nobody scripted. This is what most people now mean by "chatbot".

4. Voice chatbots. Everything above, plus speech recognition on the way in and speech synthesis on the way out. Siri, Alexa and Google Assistant are the consumer examples; phone-based support bots are the business ones.

Two more worth naming, though they cut across the four:

• Generative chatbots, built directly on LLMs, which generate rather than select responses.
• Hybrid chatbots, which combine scripted reliability for known flows with AI for everything else. This is the most common serious business deployment.

Chatbot examples

Customer support. The chat window on a website that answers FAQs, checks order status and escalates what it cannot resolve. The most common deployment by volume.

Messaging-app assistants. Bots on Instagram, Facebook Messenger and WhatsApp that reply to DMs, answer product questions and qualify leads. See best Instagram chatbot for the specifics of that channel, and how to set up an Instagram AI chatbot for the practical steps.

Voice assistants. Siri, Alexa and Google Assistant, which are voice-driven chatbots.

Generative chatbots. ChatGPT and similar, holding open-ended conversation on almost any topic.

Sales and lead qualification. Greets a visitor or a DM, asks a few screening questions, books a call for the ones that qualify and routes the rest to a nurture sequence.

Internal and HR bots. Answering staff questions about policies, booking leave, raising IT tickets. Unglamorous and often the highest return, because the questions are repetitive and the audience is captive.

Outbound and prospecting bots. Qualifying inbound replies to outreach so a human only picks up the interested ones - closer to the job of an SDR than a support agent.

Booking and scheduling. Restaurants, clinics and salons taking reservations conversationally instead of through a form.

E-commerce assistants. Product finders that ask what you are shopping for and narrow the catalogue, plus post-purchase order tracking.

Chatbot vs AI agent: what is the difference?

This is the most important distinction in 2026. A chatbot's job is to converse, it responds within a conversation. An AI agent goes further: it pursues a goal autonomously and takes actions across other systems to complete a task end to end, not just talk about it. A chatbot might tell a lead your pricing; an AI agent qualifies the lead, books the meeting, updates the CRM, and triggers the follow-up. Put simply: a chatbot answers, an agent acts. Many modern business tools pair a conversational chatbot front end with an agent that takes action behind it.

It is also worth separating two neighbouring terms:

• Live chat is a human on the other end, using chat software. No automation involved.
• Virtual assistant is a loose marketing term, sometimes meaning a voice chatbot, sometimes a general-purpose consumer assistant. Ask what it actually does rather than what it is called.

What chatbots are bad at

Any honest guide has to include this section, because most of the disappointment with chatbots comes from deploying them against the wrong problem.

• Anything genuinely novel. If a question has never been asked before and the answer is not in the source material, an AI chatbot will still produce a fluent answer. Fluent and correct are different things.
• Emotionally loaded conversations. Complaints, cancellations, bereavement, anything where a person needs to feel heard. Speed is not the value there.
• High-stakes accuracy. Medical, legal and financial specifics. Use retrieval from a controlled source and cite it, or route to a human.
• Being a substitute for a broken process. A bot in front of a slow fulfilment operation just makes people angry faster. Fix the process first, then automate the conversation about it - see AI automation for where that line sits.
• Pretending to be human. It rarely survives contact, and it costs more trust than it saves effort.

The corollary: chatbots are excellent at high-volume, repetitive, low-emotion, well-documented questions. That is a large share of most inboxes, but it is not all of it.

How much does a chatbot cost?

Pricing in 2026 falls into three bands.

• Free tiers and low-cost tools, roughly $0 to $50 a month. Usually limited by conversations, contacts or "active users". Fine for a single business with modest volume.
• Mid-market platforms, roughly $50 to $500 a month, adding multi-channel support, integrations, team seats and analytics.
• Enterprise and custom builds, from several thousand a month upward, typically involving bespoke integration work and dedicated support.

Two costs regularly surprise people. The first is usage-based billing: many vendors charge per conversation, per resolution or per "monthly active contact", which means a good month for marketing is an expensive month for the tool. The second is maintenance: a chatbot is not a one-off build. Its answers go stale, its intents drift, and somebody has to own it.

Do chatbots actually work? What the numbers say

Be sceptical of vendor statistics in this category, including ours. Most published figures come from companies selling chatbots, are not independently audited, and rarely disclose how they were measured.

What we can speak to is our own data. We measured 9,615 first-contact conversations on business accounts and found that roughly a quarter were never answered at all, and the median business took over two hours to reply. That is the gap automation actually addresses: not replacing good conversations, but ensuring a conversation happens at all. Our cold DM benchmarks cover what reply rates look like in practice.

The honest framing is that a chatbot's value is mostly about coverage and speed rather than persuasion. It answers at 2am, it answers the fortieth identical question without irritation, and it answers within seconds while someone still has their phone in their hand.

How to choose a chatbot

A short checklist that cuts through most vendor comparisons:

• Where does the conversation happen? Website chat, Instagram DM, WhatsApp and voice are genuinely different products. Pick for your actual channel, not the one with the best demo.
• What can it read? A bot is only as good as the material it can draw on. Ask how it ingests your help centre, product catalogue or past tickets, and how it stays current.
• What can it do? Answering is table stakes. Booking, looking up an order, updating a record, taking payment: each one is a separate question. Tools differ on whether actions are built in or need a separate workflow builder.
• How does handoff work? The most important feature. When it cannot help, does a human get the full conversation with context, or does the customer start again? This is much easier when the bot and the team inbox are the same system.
• How is it billed? Per seat, per conversation, per contact or per workspace. Model your busiest month, not your average one. Our own pricing is per workspace for this reason.
• Who owns it after launch? If the answer is nobody, expect quality to decay within a quarter.

Metrics that tell you whether it is working

• Containment or deflection rate - the share of conversations resolved without a human. The headline number, and the easiest to flatter by counting abandonments as successes.
• Handoff rate and handoff quality - how often it escalates, and whether context survives the transfer.
• First response time - usually the clearest win, since a bot answers instantly.
• Resolution rate - did the person's problem actually get solved. Harder to measure and more honest than containment.
• Customer satisfaction on bot conversations specifically - segment it, or a good human team will mask a poor bot.
• Fallback rate - how often it says it did not understand. A rising fallback rate is the earliest signal that your content has gone stale.

How businesses use chatbots in 2026

The biggest business value is speed and availability. A chatbot answers instantly, 24/7, which matters enormously because response speed is one of the strongest predictors of whether a lead converts, a reply in minutes dramatically outperforms one in hours. Common business uses: first-line customer support (deflecting and resolving common questions), lead qualification (greeting and screening inbound interest), instant DM responses on social channels, and booking. For Instagram and DM-first businesses especially, an AI chatbot that replies to every DM the moment it arrives, qualifies the person, and routes hot leads to a human captures pipeline that slow manual replies would lose. That is what Inflowave's AI chatbot does across Instagram DM, comments, and more, feeding qualified conversations into a pipeline rather than leaving them in an inbox.

If you are weighing platforms rather than concepts, our best CRM for marketing agencies guide covers the tools that combine conversation handling with a place to put the resulting leads.

FAQ

What is an example of a chatbot?

Common examples include the customer-support chat window on a website, automated assistants inside messaging apps like Instagram, Messenger, and WhatsApp, voice assistants such as Siri and Alexa, and generative AI chatbots like ChatGPT. In a business context, a lead-qualification chatbot that greets website or DM visitors, asks a few screening questions, and books a call is a typical example.

What are the four types of chatbots?

The four types usually named are menu or button-based bots, keyword-recognition bots, contextual or AI bots, and voice bots. They differ mainly in how much interpretation they do: a menu bot needs you to click a predefined option, while a contextual bot works out your intent from free text and remembers what was said earlier in the conversation. Generative and hybrid bots are often listed as additional categories.

Is Siri a chatbot? Is Alexa a chatbot?

Yes, both Siri and Alexa are voice chatbots (also called voice assistants). They simulate conversation through speech rather than text: you speak, they interpret your intent, and they respond or take an action. They are more sophisticated than simple rule-based text bots because they use natural language understanding, but they fall under the broad chatbot umbrella, software that converses with humans.

Is ChatGPT a chatbot?

Yes, ChatGPT is a generative AI chatbot, it converses with users by generating human-like text responses using a large language model. It is a more advanced, open-ended chatbot than a scripted support bot. Note that ChatGPT on its own primarily converses; it becomes an AI agent when given tools, memory, and the ability to take autonomous actions toward a goal.

What is the difference between a chatbot and an AI agent?

A chatbot converses, it responds to messages within a conversation but generally does not take actions in other systems. An AI agent pursues a goal autonomously and takes actions across tools to complete a task end to end, such as qualifying a lead and then actually booking the meeting and updating the CRM. The shorthand: a chatbot answers, an agent acts. Many products combine a chatbot interface with an agent that does the work behind it.

How can you tell if you are talking to a chatbot?

Signs include instant responses at any hour, slightly generic or repetitive phrasing, difficulty handling unexpected or nuanced questions, and an inability to deviate from certain topics. That said, modern AI chatbots powered by large language models have become much harder to distinguish from humans, so the old tells are less reliable than they used to be. Reputable businesses typically disclose when you are chatting with a bot and offer a path to a human.

How much does a chatbot cost?

Free and entry-level tools run from $0 to about $50 a month, mid-market platforms from roughly $50 to $500, and enterprise or custom builds from several thousand upward. Watch for usage-based billing, where the charge scales with conversations, resolutions or active contacts, and budget for ongoing maintenance rather than treating it as a one-off build.

Do I need a chatbot for my small business?

It depends on volume and repetition rather than size. If you are answering the same handful of questions many times a week, or losing enquiries that arrive outside working hours, a chatbot pays for itself quickly. If your enquiries are few, high-value and highly varied, your time is better spent answering them properly than automating them.

Can a chatbot replace customer service staff?

Not in any business we would want to be a customer of. A chatbot handles the repetitive, well-documented share of an inbox, which is often substantial, and frees people for the conversations that need judgement, authority or empathy. Teams that deploy chatbots to remove headcount rather than to redirect it usually end up rehiring.

What is the difference between a chatbot and live chat?

Live chat is a human replying through chat software. A chatbot is software replying automatically. Many businesses run both: the bot handles first contact and out-of-hours coverage, and hands over to a person when the question needs one. The handoff quality is what determines whether customers experience that as helpful or as an obstacle.