What does an AI chatbot for your website cost?

"AI chatbot cost" has no single answer, because three quite different things are sold under the name. What an AI chatbot for your website costs depends first on which of the three you need, and only then on the vendor.

• A SaaS widget: a chat tool you subscribe to, point at your website or FAQ, and paste onto your pages. You set it up yourself.
• A configured assistant on your own documents: an AI assistant that searches your manuals, price lists, terms and product data, and answers from them. Someone prepares the content, tests the answers and keeps it running.
• A custom assistant with integrations: the same, plus live connections to your systems, so it can look up an order, book an appointment or add a lead to your CRM.

For most small businesses the middle column is the sensible start, which is why it is highlighted. It answers from your own content instead of the open internet, it does not carry the cost of integrations, and you can connect it to your systems later, once you know which questions it gets.

SaaS chatbots charge in one of three ways: a monthly plan with a quota, a price per AI conversation, or a price per resolved question. Most show their prices in US dollars, excluding VAT.

Under every chatbot sits a language model, billed per token (a token is roughly three quarters of a word). With stated assumptions you can work out what a conversation costs. We assume four questions per conversation, each sending about 4,000 tokens (the instructions, the relevant passages from your documents and the conversation so far) and receiving about 250 tokens of answer: 16,000 tokens in and 1,000 out.

A thousand conversations a month on a mainstream model cost about $42 in model usage. What you really pay for is everything around the model: preparing your content, testing, hosting, monitoring and someone who fixes wrong answers. Over three years, that looks like this:

Two assistants that look the same in the chat window can differ by thousands of euros. The difference is almost always in one of these six things.

1. 1

   Messy source documents

   Outdated PDFs, price lists that contradict each other and knowledge that lives only in people's heads. Sorting this out is often the largest part of the setup, and it is what makes the answers right.

2. 2

   Integrations

   Every system the assistant talks to, such as your webshop, agenda or CRM, needs a connection, permissions, error handling and tests. Each one is development work.

3. 3

   Languages

   Every language needs its own test questions and checked answers. An assistant that answers in German from Dutch documents must be tested in German.

4. 4

   Handover to a person

   When should the bot step aside, how does a colleague pick up the conversation, and what happens outside office hours? A good handover passes on a summary, so the customer does not have to repeat everything.

5. 5

   Logging and review

   Someone should read a sample of conversations every week, spot wrong or missing answers and fix the source. Without that, quality slowly drifts.

6. 6

   Volume and model choice

   More conversations, longer documents in every request and larger models all raise usage. Per conversation it is small, but at thousands a month it shows on the bill.

The one place not to save is testing. Before launch, collect fifty real questions from your inbox and check every answer. It costs a day and prevents the embarrassing mistakes that make customers stop trusting the bot.

Visitors type names, email addresses, order numbers and sometimes far more personal details into a chat window. That makes the chat log personal data under the GDPR, and you are responsible for it, even when a vendor runs the chatbot.

• Sign a data processing agreement with every provider that handles the conversations, as article 28 of the GDPR requires. When we build an assistant for you, we sign one with you, as our terms and conditions describe.
• Mention the chatbot in your privacy statement: what it collects, why, who processes it and how long you keep it.
• Set a retention period for chat logs, and delete them on schedule.
• Ask where the provider processes the conversations and, if that is outside the European Economic Area, on what legal basis the data is transferred.
• Check that conversations are not used to train the provider's models, and that the setting for this is correct in your account.
• Keep sensitive data out: ask visitors not to share health details or passwords, and connect only the systems the assistant really needs.

The EU AI Act adds one duty that every website chatbot meets. Article 50 requires that an AI system talking directly to people is built so they are told they are interacting with AI, at the latest at the first interaction, unless that is obvious. This duty has applied since 2 August 2026. The Digital Omnibus on AI, published on 24 July 2026, moved the deadlines for high-risk systems but left this duty unchanged.

A chatbot earns its money in three ways: questions it answers that a person would otherwise have answered, time your team gets back, and visitors who buy or book because they got an answer at 10 at night. The first two are easy to measure; conversion is harder, so leave it out of the first calculation.

The weakest assumption is the 40%. Some chat questions would never have reached your inbox, and some answers send the customer back to you anyway. Count only questions a person would otherwise have handled, and measure the share during a pilot instead of taking a vendor's average.

We build the middle and right-hand columns of the comparison: assistants that answer from your own documents, and, where it pays, connect to your systems. Every project follows the same steps as the rest of our AI work, and nothing goes live before you have seen it answer your own questions.

1. 1

   Discover

   We look at the questions you get, the documents that answer them and where a chatbot has the highest return.

2. 2

   Pilot

   A working assistant on your real documents, tested on real questions, so you can judge the answers. Then it is your call: scale up or stop.

3. 3

   Deploy

   We put it on your website, set up the handover to your team, secure it and show your team how to review conversations.

4. 4

   Improve

   We measure the results, fix gaps in the answers and adjust the assistant as your products and questions change.

We build on SereChat, the AI platform we develop and operate ourselves, which includes a knowledge library for your documents. We choose the model and provider per use case, name them in the quote, and choose providers and settings that do not use your data to train public models.

An AI assistant on your own documents starts at €2,500, and automation that connects to your systems at €3,500. Running costs start at €150 a month and grow with volume; you know the expected amount before you start. See AI prices, or ask us about your chatbot.