What chatbot means and where the word comes from
The word joins “chat”, a written conversation, and “bot”, short for robot. You will also hear conversational bot, virtual assistant or simply bot. They all name the same thing: a program that receives messages and answers them.
It is not a recent invention. The first programs able to simulate a conversation are more than half a century old; what has changed in recent years is that artificial intelligence lets them understand sentences as anyone types them, with spelling slips and without following a menu.
How a chatbot works
From the outside it looks like a conversation; inside, each message goes through four steps. How good a chatbot is depends almost entirely on the second and third: what it understands and where it gets its answers from.
- It receives the message on the channel where it lives: a website chat, WhatsApp or Instagram.
- It interprets it: it looks for a keyword, recognises what the question is about or, if it uses a language model, understands the whole sentence.
- It decides what to reply: a fixed text, a piece of data it looks up (opening hours, a price, a free slot) or an answer written on the spot.
- It replies and, if it does not know or the matter is not for it, passes the conversation to a person.
An example, message by message
A customer messages a hair salon on a Friday, just after eleven at night: “Have you got anything tomorrow morning for a cut and colour?”. Depending on the type of chatbot, the conversation goes in very different directions.
A menu chatbot replies with five numbered options and asks her to pick one; she has to guess that what she wants is under bookings, then choose a service, a day and a time in three more menus. A keyword chatbot spots “anything tomorrow” and sends her the link to the diary, even though she was asking about two services at once.
An AI agent understands the whole sentence, checks Saturday’s slots for that service and offers her two times. When she picks one, it shows her the summary and waits for her to tap the confirm button; the request then holds its slot in the calendar until the salon confirms it with her. If she also asks the price of colour with highlights and that price has not been published, it does not make one up: it passes the question to the team for first thing on Saturday.
Types of chatbot
The types differ in how much they understand and how much they can do. None is best for everything: a button menu is enough to give opening hours and falls short when a question has nuances.
- Button or menu chatbots: the user chooses between options. Predictable and limited; if the question is not on the menu, it gets stuck.
- Rule and keyword chatbots: they reply when the message contains certain words. They fail with synonyms, typos and long sentences.
- Intent recognition: they learn to group similar questions (“what time do you open”, “what are your hours?”) and answer each group.
- Generative: they use a language model that writes new replies. They understand almost any sentence but need limits so they do not stray from the business’s information.
- AI agent: as well as answering, it carries out tasks with tools, such as checking a diary or preparing a booking, with permissions and confirmations.
What a chatbot does for a business
In a service business, a good share of messages are the same ten questions: opening hours, prices, whether there is space, how to book, what to bring. A chatbot answers them instantly and consistently, at night and at weekends too, and leaves the team what genuinely needs a person.
This is how it looks in different sectors:
- An academy: which groups suit a student’s age and when the next trial class is.
- A gym: the class timetable, whether a session is full and how to cancel a membership, which goes to the team.
- A hair salon: which services there are, how long they take and which slots are left on Saturday.
- A physiotherapy clinic: where it is, what to bring to the first visit and how to move an appointment.
Chatbot vs ChatGPT: the difference
ChatGPT is a general-purpose artificial intelligence assistant: anyone can ask it anything and it answers with what it knows. A business chatbot does something else: it serves customers on behalf of a business, on its channels, with its information and under its rules.
Inside, a generative chatbot may use the same kind of language model, but trimmed down: it only talks about the business, answers with the data the company has published, does not improvise prices or conditions and knows when to hand over to a person. That difference is what separates a useful tool from a risk to your reputation.
Is WhatsApp a chatbot?
No. WhatsApp is a messaging app; a chatbot is a program that can reply inside it. The WhatsApp Business app comes with a business profile and a greeting for whoever opens a conversation, but that greeting does not converse: it is a fixed message.
For a chatbot to answer on a company’s WhatsApp, the number is connected to the official route Meta offers businesses, known as the WhatsApp Business API. The number must belong to the business, receive the verification code by SMS or by call and not be in use in the app. What a well-built WhatsApp chatbot does is covered separately.
A chatbot on your website or on WhatsApp
A website chatbot only serves whoever is on the page at that moment: if the person closes the tab, the conversation is gone and cannot be picked up again. On WhatsApp the conversation stays on the customer’s phone, they can come back to it hours later, and the business does not depend on anyone visiting its site.
For a service business whose customers already write on WhatsApp, that is the chatbot’s natural home. The website is still useful for explaining what you do; day-to-day contact with customers happens in the chat.
Advantages and limits of a chatbot
The clearest advantage is time: the customer gets an answer straight away and the team stops replying to the same thing twenty times a day. Next come consistency, because it always answers with the same information, and order, because the conversations that end up with the team arrive grouped and with context.
The limits are just as clear. A button chatbot frustrates people as soon as the question goes beyond the menu. A generative one without limits can invent a price or a condition. And none should handle a delicate complaint or a case that calls for judgement: that is where a person has to step in.
There are obligations too. Since 2 August 2026, Article 50(1) of the EU AI Act requires people to be told from the first interaction that they are talking to an artificial intelligence, unless it is obvious. WhatsApp’s policy allows automated replies as long as there is a quick, clear way to reach a person, and forbids asking for full card numbers, bank accounts or identity documents.
How to create a chatbot for your business
Building a useful chatbot is less about technology than preparation. This is the order that works:
- Decide what it should handle and what it should not. Opening hours, prices and bookings, yes; complaints, payments and special cases, to a person.
- Write down the information it can use: services, opening hours, prices and rules, checked and current. Anything not written down, it should not answer.
- Choose the channel. If your customers write on WhatsApp, connect the business number through the official platform.
- Define the handover: when it passes to a person and what it notes down so they do not have to ask again.
- Tell the customer from the first message that an AI is serving them and how to ask for a person.
- Test it with real questions from your customers and review its conversations during the first weeks.
Common chatbot mistakes
Nearly every annoying chatbot makes one of these mistakes, and nearly all of them are avoided with preparation rather than technology:
- No way out: the customer asks for a person and the chatbot goes back to the menu.
- Asking for too much: full name, email and phone number before answering something as simple as opening hours.
- Answering everything: a business chatbot that offers opinions on any subject ends up saying something the business would never sign off.
- Out-of-date information: prices or hours that changed months ago and nobody updated.
- Passing itself off as a person: besides breaking the European rule, it destroys trust the moment the customer notices.
- Very long messages: in a chat people read short sentences, not paragraphs.
How much a WhatsApp chatbot costs
There are two parts. The first is what Meta charges: each message delivered, according to its category (marketing, utility, authentication or service) and the country. Since 1 October 2026, replies sent within 24 hours of the customer’s last message are service messages, and the first 1,000 each month per number are not charged; outside that window, only templates can be sent.
The second is the software acting as the chatbot, which each provider prices in its own way. A chatbot that only replies to people who write in sends service messages, so what Meta charges depends on how many go beyond those 1,000 a month.
Chatbot or AI agent
The line lies in what it can do. A chatbot answers. An AI agent also acts: it checks the diary, prepares a booking that is not valid until the customer confirms it, notes in the CRM what they have said and passes the case to a person with a summary. For a business that serves customers on WhatsApp, that difference is what really saves work.
We explain it in more depth in the guide to AI agents, along with the controls it needs before you can trust it.
