AI agents · Customer service AI chatbot for your website: what a good chatbot must be able to do and what to watch out for

What a good AI chatbot must be able to do

• Answer from your content: the basis is your website, product documentation, prices and rules, not the model’s general knowledge.
• Name sources: with every answer, a link to the page or document it comes from.
• Know its limits: if something is not in the documents, the chatbot says so instead of guessing.
• Hand over to people: for complaints, sensitive questions or uncertainty, a ticket or call-back request is created with a summary of the conversation.
• Look up data: through interfaces it can query delivery status, free appointments or availability, with minimal permissions.
• Answer in several languages: in the customer’s language, with the same checked content.
• Disclose that it is an AI: clearly, right at the start and accessibly.
• Be usable by everyone: with a keyboard, a screen reader and on a mobile phone.

Rule-based bot, AI chatbot or AI agent?

For most businesses, an AI chatbot with a knowledge base is the right way in. How the knowledge search behind it works is explained in our article AI with your own company data. When an agent makes sense is described on the page AI customer service.

What businesses are liable for

In Moffatt v. Air Canada, the airline’s chatbot had promised a customer a discount that did not exist in that form. The airline argued that the chatbot was responsible for its own statements. The Civil Resolution Tribunal in British Columbia disagreed: the company was responsible for all the information on its website, including that given by a chatbot, and had not taken enough care to ensure it was accurate. The decision comes from Canada and binds no one in Europe. The lesson still applies: a note such as “information without guarantee” does not replace correct answers. Limit the chatbot to checked content and leave binding commitments to people.

Law and obligations in the EU

• AI notice: under Art. 50 of the EU AI Act, people must have been informed since 2 August 2026 that they are interacting with an AI, unless this is obvious. More on the transparency obligations.
• Data protection: chat histories are personal data. You need information in your privacy policy, deletion periods, a data processing agreement and, where possible, servers in the EU.
• Accessibility: since 28 June 2025 the European Accessibility Act has applied to many consumer services, including e-commerce. A chatbot on such a website must be accessible too. More under accessibility.
• Binding commitments: the chatbot should only state prices, deadlines and goodwill gestures from checked sources or hand over to people.

What the costs depend on

The cost of a chatbot depends less on the technology than on the scope: how much content is prepared, which systems are connected, how many languages are needed and how many conversations take place each month. On top of that come operation and content maintenance. Off-the-shelf builders are quick to set up, but reach their limits as soon as data from your own systems is needed. A custom solution pays off if the chatbot should do more than answer frequent questions.

How to introduce a chatbot

1. Collect the most frequent questions: from emails, phone notes and the contact form.
2. Prepare content: answers to these questions belong on the website, clear and current. Your visibility in search engines benefits too.
3. Define the handover: when does a conversation go to a person, by which route, with which information?
4. Test with real questions: at least 50 questions from everyday work, including difficult and unfriendly ones.
5. Start small: first on a few pages or at certain times, then expand.
6. Evaluate: resolution rate, handovers and above all unanswered questions. Every unanswered question points to missing content.