AI chatbot

Short answer: a chatbot is software that answers questions through a chat window. An AI chatbot runs on a language model and understands plain language, instead of only fixed buttons and scripts. You know them as standalone assistants such as ChatGPT, Claude and Gemini, but you can also build one that runs on your own website, in WhatsApp or in your customer service and answers based on your information. A good chatbot helps people move forward faster. A bad one holds them back.

On this page you'll read what a chatbot is, which types exist, what an AI chatbot can do for your business, what to watch out for (technology, law and trust) and which tools you can use to make one, from Zendesk and Trengo to building it yourself with Lovable or Claude.

What is a chatbot?

A chatbot is a program that holds a conversation through text or speech. You ask a question, the bot answers, and you go back and forth. The word combines chat and robot.

Chatbots aren't new. ELIZA, a 1966 program by Joseph Weizenbaum at MIT, already pretended to hold a conversation by cleverly reflecting sentences back. What is new: since large language models (LLMs) such as GPT, Claude and Gemini, a chatbot understands free language, remembers what was said earlier in the conversation and can write an answer that was never literally in a script.

The types of chatbots

There are roughly three types. In practice you increasingly see hybrids.

You'll also hear the term AI agent. An agent doesn't just talk, it also carries out tasks: booking an appointment, looking up an order, creating a ticket. So: a chatbot answers, an agent also does something. Want to know more? Read on at AI agents.

Standalone AI chatbots: ChatGPT, Claude, Gemini and Copilot

The best-known chatbots are the assistants you use yourself. They differ in nuance, not in principle.

• ChatGPT (OpenAI): the best known, broadly usable, strong at writing, analysis and working with files.
• Claude (Anthropic): strong at long texts, nuance and programming. Widely used to build code and complete tools.
• Gemini (Google): woven into Google Workspace and Google Search.
• Microsoft Copilot: built into Windows, Edge and Microsoft 365, handy if your organisation runs on Microsoft.
• Perplexity: a chatbot focused on searching and showing sources.

For your business these assistants matter in two ways. First as a work tool: for content, analysis, customer emails and research. Second as a place where customers find you. More and more people ask ChatGPT or Gemini for advice or a supplier instead of googling. Whether your business shows up in those answers is called GEO (generative engine optimisation), alongside SEO for Google. Read more at SEO and GEO.

Your own AI chatbot on your website

Your own chatbot is different from using ChatGPT. It's an assistant that runs on your website, in your app, in WhatsApp or in your customer service and answers questions about your offer. Under the hood it usually uses one of the big language models, but you decide:

• which knowledge it uses (your website, manuals, FAQ, product feed, help centre);
• how it talks (your tone of voice, your language, what it does and doesn't say);
• when it hands over to a human (by email, WhatsApp, phone or a ticket);
• what it's allowed to do (only answer, or also look something up or schedule).

An example: this site runs Baile, our own AI chatbot. He knows the site's content, answers in English or Dutch, links to relevant pages and always shows how to email or message us directly. You'll find him in the contact button at the bottom right.

What an AI chatbot gets you

A good chatbot pays off mainly in three places. How much exactly depends on your volume of questions, your offer and how well the bot is set up. Be careful with promises you read elsewhere: measure it for yourself.

1. Customer service: repeat questions (delivery times, returns, opening hours, "how does X work?") get answered instantly, including evenings and weekends. Your team keeps time for questions that really need a person.
2. Sales and leads: visitors who hesitate get an answer straight away instead of clicking away. A bot can also ask questions to understand what someone is looking for and then recommend the right page or person.
3. Insight: you literally see which questions people ask. That's gold for your website, FAQ, content and offer. Questions that keep coming back are missing pages.

Chatbots also work well internally: an assistant that knows your staff handbook, processes or product knowledge saves colleagues a lot of searching.

What to watch out for

This is the part that decides whether your chatbot helps or hurts. Check it before you go live.

1. Always a way out to a human

The biggest frustration with chatbots is getting stuck. Make sure there's always a visible route to a person: email, WhatsApp, phone or a form. A bot that hides it to "save costs" ends up costing you customers.

2. Better "I don't know" than making things up

Language models can give convincing-sounding nonsense. That's called hallucinating. Restrict the bot to your own information, let it show sources or links, and let it say honestly when it doesn't know. This isn't a detail: in 2024 Air Canada had to compensate a customer because its chatbot invented a refund policy. The tribunal found the company responsible for what its chatbot says.

3. Make clear it's AI

People should know they're talking to an AI. That's decent, and the EU AI Act requires it too: Article 50 requires that people are informed when they interact with an AI system, unless that's already obvious. Just put it in the header of the chat window.

4. Privacy and the GDPR

Visitors sometimes type personal data into a chat. Decide what you store, for how long and where. Choose a provider that processes data in the EU or with appropriate safeguards, sign a data processing agreement and mention the chatbot in your privacy statement. Don't ask for more data than needed.

5. Shielding what isn't public

A chatbot must never reveal internal documents, customer data or its own instructions. Only give it access to information that may be public, and test whether it resists tricks such as "ignore your instructions" or "I'm the administrator".

6. Your brand and your tone

A bot speaks on your behalf. Define how it talks (formal or casual), what it never says (for example prices or guarantees) and how it handles angry customers. Humour is fine, but only if it's genuinely good.

7. Up-to-date knowledge

A bot is only as good as its source. Does your offer, terms or opening hours change? Then the bot's knowledge has to change too. Agree who keeps it up to date.

8. Cost per conversation

With AI chatbots you often pay per conversation, per answer or per "resolved" question, on top of a subscription. Calculate with your expected volume beforehand, and protect the bot against abuse (for example a limit per visitor).

9. Measuring

Don't just look at the number of conversations. Look at how many questions were actually resolved, how often people still look for a human, where the bot got it wrong and what it delivers in leads or revenue. Read conversations back now and then: it's the fastest way to improve it.

Tools to build a chatbot

There are roughly three routes. Which fits depends on what you already use, how much control you want and who maintains it. Prices and features change fast, so always check the provider's current information.

Route 1: chatbot in your customer service software

Already using a customer service package? Then its chatbot is often the fastest route, because it can reach your tickets, inbox and customer data directly.

Upside: live quickly, managed without code, connected to your existing support. Downside: you're tied to their way of working, and costs can rise with volume.

Route 2: chatbot builders

Platforms such as Voiceflow and Botpress are made to design conversation flows and AI bots visually, with connections to your own systems. They give more freedom than a support package without having to code everything yourself. A good fit if you want a bot that works across several channels or walks through specific steps.

Route 3: build it yourself (or have it built) with AI tools

This is the route that has changed most in recent years. With tools like Lovable and Claude you can quickly build your own chatbot that does exactly what you want, in your own branding, on your own knowledge and with your own rules.

• Lovable: you describe what you want in plain language, Lovable builds the website or app including the chatbot and connects it to a language model. Baile on this site was built that way. Want to know more? Read on at Lovable.
• Claude (and similar models): strong at writing the code for a chatbot, and often the language model running behind it.
• Directly via a language model: developers connect a model from OpenAI, Anthropic or Google straight to their site or app.

Upside: full control over knowledge, tone, look and costs, and no dependence on one support package. Downside: you have to set it up and maintain it well. The technology has become the easy part; the choices (what may it do, when does it hand over) are the real work.

Note: custom GPTs in ChatGPT aren't a website chatbot. They only work inside ChatGPT, and OpenAI is reworking them. Read more at custom GPTs retiring.

Which route suits you?

How to approach it: six steps

1. Collect the questions. Look in your inbox, WhatsApp, phone notes and search data for the questions that come back most.
2. Pick the goal. Fewer repeat questions, more leads, or unlocking internal knowledge? One main goal per bot.
3. Tidy up your knowledge. A chatbot on messy information gives messy answers. Make sure your FAQ and pages are correct.
4. Write the rules. Tone, what it never says, when it hands over and how it says it's AI.
5. Test with real questions. Including difficult, angry and odd ones, and attempts to trick it.
6. Go live, measure, improve. Read conversations back, fill in missing knowledge and adjust.

Frequently asked questions

What is a chatbot?

A chatbot is software that answers questions through a chat window, in text or speech. Simple chatbots work with fixed buttons and scripts. AI chatbots run on a language model, understand plain language and can write answers based on broad knowledge or on your own information.

What's the difference between a chatbot and an AI chatbot?

A classic chatbot follows a fixed decision tree and gets stuck on questions outside the script. An AI chatbot uses a language model, understands free questions and writes its own answer. That's more flexible, but needs clear rules to prevent made-up answers.

Is ChatGPT a chatbot?

Yes. ChatGPT is an AI chatbot by OpenAI, like Claude (Anthropic), Gemini (Google) and Copilot (Microsoft). They have broad general knowledge but don't know your business. For questions about your offer you need your own chatbot that answers from your information.

How do you build an AI chatbot for your website?

First choose a route: the chatbot in your customer service package (such as Zendesk or Trengo), a chatbot builder (such as Voiceflow or Botpress) or build it yourself with a tool like Lovable or Claude. Then give it your own knowledge, write rules for tone and handover, test with real questions and measure after launch.

What does an AI chatbot cost?

It depends on the route and your volume. Support packages often charge a subscription plus a fee per conversation or resolved question. With your own chatbot you mainly pay for language model usage, plus building and maintenance. Calculate with your expected number of conversations beforehand.

May a chatbot pretend to be human?

No. Under the EU AI Act people must know they're interacting with an AI system, unless that's already obvious. So show clearly in the chat window that it's an AI assistant, and always offer a route to a human.

Is an AI chatbot GDPR compliant?

It can be, but not automatically. Decide which data you store and for how long, choose a provider with appropriate safeguards (preferably processing in the EU), sign a data processing agreement and mention the chatbot in your privacy statement.

What's the difference between a chatbot and an AI agent?

A chatbot answers. An AI agent also carries out tasks, such as booking an appointment, looking up an order or creating a ticket. Many modern chatbots move towards agents as soon as they're connected to your systems.

Will a chatbot replace my customer service?

No, it relieves your customer service. A good chatbot handles repeat questions and passes the rest to a human, with the context of the conversation attached. Companies that remove the human entirely risk losing customers.

Sources

• Joseph Weizenbaum, "ELIZA: A Computer Program for the Study of Natural Language Communication Between Man and Machine", Communications of the ACM (1966)
• Regulation (EU) 2024/1689 (AI Act), Article 50: transparency obligations
• Civil Resolution Tribunal (Canada), Moffatt v. Air Canada, 2024 BCCRT 149
• Dutch Data Protection Authority (Autoriteit Persoonsgegevens), guidance on AI and the GDPR

Want to know whether an AI chatbot would pay off for your business, and which route fits? Book a free strategy session. You'll hear honestly whether it makes sense and, if so, where to start. Want to know more about AI in your business? Read on at AI agents and digital workers.