AI chatbot development that answers the way you do

Most small-business websites already hold the answer to what visitors ask. It just sits where nobody finds it, or it arrives the next working day. This is what we see on almost every site we review:

• The question comes after hoursA visitor looking for a price at 22:40 fills in a form or leaves. Tomorrow they call the competitor who replied at once.
• Nobody reads the FAQ pageTwenty questions in a list, and the visitor needs the twenty-first. A chatbot searches those twenty answers plus your service pages and price list.
• Requests arrive half-complete"Please quote for my bathroom" with no address, size or timing. Three emails follow before you can price anything.
• An off-the-shelf bot guessesA generic bot that does not know what you do and do not offer promises a service you never sold, or invents a price.
• Nobody says it is a botSince 2 August 2026, Article 50 of the EU AI Act requires that people know they are talking to AI. A bot with a first name and a stock photo does not meet that bar. What the rule asks for is on EU AI Act transparency obligations.

A good bot fixes all five at once. It answers at any hour, only from what is true, and hands over a request you can answer straight away. How such a site looks with the bot on it, see website design.

A website chatbot does not do one thing but a handful. For each type of question: what happens today, what the bot does and what you gain.

The bot belongs to your website, not your inbox. What happens next with your team's emails and tickets is the job of AI customer service. If the bot should also act, such as booking an appointment or drafting a quote, it becomes an agent: see AI agent development and the AI quote generator. If your team would rather have an assistant on internal documents, see custom ChatGPT for your business. The wider offer is on AI for business, and automation around it on AI automation.

Take a renovation company with a busy site and 400 chat conversations a month. The figures are assumptions; run them again with your own site and your own hourly cost.

The sum: 34 hours a month times twelve is 408 hours, times 45 euros is 18,360 euros. Model cost and follow-up come off that. The other 100 conversations a month go to a person and save no time in this example.

The model cost: a ten-message conversation with 3,000 input tokens per question and 250 per answer costs 0.04 dollars on Claude Sonnet 5.5 (2 dollars per million tokens in, 10 out) and 0.002 dollars on Claude Haiku 5.5 (0.10 and 0.50). On Haiku the model bill in this example drops to 85 cents a month. Run your own volume in the AI cost calculator; which model suits which work is in ChatGPT vs Claude.

What does that say about the build? In this example a custom AI chatbot may cost about 17,000 euros to pay back in one year, and twice that to pay back in two. A request that reaches you at 22:40 instead of your competitor is not even counted: what an extra client is worth, you know better than we do.

There are good chatbot tools with a monthly price. Three well-known ones, in dollars, before VAT:

In the worked example above, Intercom Fin would cost 237.60 dollars a month for 240 resolved questions, or 2,851 dollars a year before seats. Chatbase Standard costs 1,440 dollars a year on annual billing. The model in a custom build costs 204 dollars a year.

Choose a tool if you mainly want to catch FAQ questions, need few integrations and do not mind the vendor's brand in your chat window.

Choose a custom build if the bot has to talk to your calendar, quoting tool or CRM, has to work in your own branding and languages, handles enough volume that a price per resolution starts to weigh, or if you want to decide where conversations are stored. The broader trade-off is on build vs buy software.

A chatbot on a website that serves people in the EU falls under two European rules every buyer should know. They apply wherever your company sits, as long as your visitors are in the EU.

The AI Act, Article 50. Since 2 August 2026, an AI system that talks directly to people must be built so they know they are dealing with AI, unless that is already obvious to an ordinary visitor. The notice must be clear and distinguishable, given at the latest at the first interaction, and accessible to people with disabilities. In practice: one line at the top of the chat window, in the first message. Article 4 has also required since 2 February 2025 that staff who use AI understand its limits. Which duties your use case carries, see the EU AI Act checker; whether you need a policy, see do I need an AI policy?

The GDPR. When the bot asks for a name, address or phone number, you process personal data. A vendor doing that on your behalf is a processor, and Article 28 requires a contract that sets out what is processed, for how long and why. Mention the chatbot in your privacy policy too. We keep conversations in an EU environment and let the bot ask only what the request needs. Anthropic's models can run through Amazon Bedrock with processing kept inside the EU; there, the model's maker does not see your prompts and answers.

One practical note: a chatbot that also answers on WhatsApp has paid, since 1 October 2026, for service messages above 1,000 per phone number per month. See WhatsApp Business pricing.

A chatbot on a client site is not in our portfolio yet; the parts are. For AMOSI, our own AI software that went live in July 2026, we built the knowledge layer shown above: every fact about a company with the page it came from, and a question back where it is unsure. That layer is what keeps a chatbot to your information. For Business Centrum we put every figure on the site in a register with its official source, plus an llms.txt so AI assistants find the right numbers; how that helps you show up in AI answers is on found in ChatGPT.

What you get from us:

• A bot that knows your own content, reviewed by you before it goes live.
• The AI notice and a privacy text from the first message.
• Requests that arrive complete, in your inbox or CRM.
• A monthly look at the conversations: what it could not answer and what we add.

We are a team in Ghent and handle the whole route in English, from the first call to the upkeep; why a team in Belgium can be the right call is on nearshore software development. Send us the ten questions you get most. Book a call and we will show how your bot would answer them.