All articles

How much does a chatbot cost: the full first-year budget

In short. How much does a chatbot cost? Count four lines, not one. Setup for our website assistant starts at 790 €. The model cost per answer is a fraction of a cent: about 0.0036 USD with a small current model, so 1,000 answers a month cost around 3.60 USD. Hosting is agreed separately. The line most quotes forget is the owner’s time, about an hour a month keeping the sources current. In our example the assistant gives back roughly ten hours a month.

A reception desk in Nice with a closed laptop, a small brass bell, a neat stack of blank folders and a ceramic bowl of lemons, soft morning light through tall shutters
The chat window is the small part; the sources behind it are the real work.

Quotes for a website chatbot vary from a free widget to several thousand euros, and they are hard to compare because each one counts different things. This guide breaks the price into the lines that actually appear over a year: setup, the model cost per answer, hosting, the owner’s time and the legal minimum. It ends with a worked first-year budget for an invented business, so you can replace our numbers with yours. For what a chatbot should and should not answer, read our companion guide on an AI chatbot for your website.

What does the setup price pay for?

Not mainly the chat window. Most of the work is choosing and cleaning the documents the assistant may answer from, writing the rules for what it must refuse, deciding when it hands over to a person, and testing all of that before launch.

Our assistant that knows your business starts at 790 €. That covers one website and one language, the approved source set, one contact route to your team, 20 agreed test questions, two revision rounds, source checks and handover instructions. The formats and volume of your sources are fixed after we look at samples, because a tidy two-page price list and two hundred scanned PDFs are not the same job. More languages, a CRM connection, bookings, private document search and later updates are separate options.

A cheap widget that skips these steps is not cheaper. It moves the cost to the day a customer is told a price or a rule that is not true.

How much does each answer cost?

Less than most owners expect, and it is easy to calculate. AI providers charge per token, roughly a piece of a word, separately for the text that goes in and the text that comes out. OpenAI’s pricing page, read on 16 November 2026, lists a small current model, gpt-5.4-mini, at 0.75 USD per million input tokens and 4.50 USD per million output tokens. Prices change often, so check the page on the day you budget.

A typical answer from a website assistant sends about 3,000 tokens in: the visitor’s question, the assistant’s instructions and the relevant excerpts from your documents. It sends about 300 tokens back. At the prices above:

Line Tokens Calculation Cost
Input 3,000 3,000 × 0.75 ÷ 1,000,000 0.00225 USD
Output 300 300 × 4.50 ÷ 1,000,000 0.00135 USD
One answer 0.00225 + 0.00135 0.0036 USD
1,000 answers a month 1,000 × 0.0036 3.60 USD

Check the arithmetic: 3,000 × 0.75 is 2,250, divided by a million is 0.00225; 300 × 4.50 is 1,350, divided by a million is 0.00135; together 0.0036, and a thousand of those is 3.60. Larger models cost more per token, and long documents pasted into every answer raise the input side, which is one reason we keep the excerpts short. The account and key can be in your name, so you see every cost directly.

What else runs every month?

Hosting, which depends on where the assistant runs and is agreed separately before work. Some businesses prefer a model running on their own machine for data reasons; our guide on a local LLM for a small business explains that route and its costs.

Then the owner’s time. Prices change, a new service appears, the summer hours start. If the documents are not updated, the assistant keeps giving the old answer with full confidence. Budget about an hour a month for reading a sample of conversations and updating the sources, more in a busy season.

Pay per answer or a fixed monthly fee?

Many chatbot platforms sell a fixed monthly subscription, often with a cap on conversations and the platform’s own model and storage behind it. That is simple to budget, but you pay the same in a quiet month as in a busy one, and your documents and conversation history live on the platform. With your own key you pay only for the answers actually given, and the sources stay yours. For a small business with seasonal traffic, usage-based pricing is usually cheaper; for a team that wants no technical decisions at all, a subscription can be worth the premium. Ask any provider to show both numbers side by side.

What does the law require?

Transparency. Article 50 of the EU AI Act requires providers to design AI systems that interact directly with people so that those people are informed they are dealing with an AI system, unless it is obvious from the context. In practice we state it in the chat window from the first message. The assistant also collects whatever visitors type, so your privacy information must cover it. Neither costs much if it is planned from the start; both are expensive to add after a complaint.

What does the first year cost? A worked example

Here is an invented holiday-rental agency in Nice, to show the method rather than a promised result. Visitors ask the same things all year: check-in times, deposits, parking, pets, cleaning fees. The agency gets about 1,000 questions a month on its site, and the owner used to answer around 150 of them by email, at about 4 minutes each.

Line Amount Calculation
Setup, one site and one language 790 € one-time
Model cost, 1,000 answers a month 43.20 USD a year 3.60 USD × 12
Hosting agreed separately depends on where it runs
Owner time keeping sources current 12 hours a year 1 hour × 12
Owner time saved on email 120 hours a year 150 × 4 min = 600 min a month, 10 hours × 12

Check the arithmetic: 3.60 × 12 is 43.20, 150 × 4 is 600 minutes or 10 hours a month, and 10 × 12 is 120 hours a year. The model cost is almost invisible next to the setup. What matters is the time: 120 hours given back against 12 hours of upkeep, as long as the assistant answers only what the sources cover and hands everything else to the owner.

A wooden key cabinet in a small rental office in Nice with rows of brass keys on hooks, a blank notice card and a pot of geraniums on the windowsill, afternoon light
Check-in times, deposits and parking: the questions an assistant can answer every day of the year.

When is a chatbot not worth the money?

When there are not enough repeated questions. If your site gets a few enquiries a week and each one is different, a clear FAQ page and a good contact form cost less and serve better. A chatbot is also the wrong tool when every answer depends on private data or a judgement the business must make, such as a medical, legal or financial decision. In those cases it should only route the visitor to a person.

If you are weighing other AI projects at the same time, our overview of AI for small business helps put them in order. And if what customers really need is a tool they keep on their phone, read about the cost of a PWA versus a store app.

What should you prepare?

Bring a short list:

  • The 20 questions customers ask most often, in their own words.
  • The documents that answer them: prices, conditions, hours, policies.
  • The questions the assistant must never answer itself.
  • Who receives the hand-overs, and how fast they reply.
  • How many questions your site gets in a month, even a rough figure.

How do we build it?

We start from your questions and documents, not from a chat widget. We agree the sources and refusal rules in writing, build the assistant, run the 20 test questions with you, fix what fails and hand over instructions for keeping the sources current. You approve before launch.

Send us your business name and the five questions customers ask most. We reply in writing within 48 hours.

Questions

How much does a chatbot cost to set up?

Our assistant that knows your business starts at 790 €: one website, one language, approved sources, one contact route, 20 test questions, two revision rounds and a handover. More languages, CRM, bookings, hosting and updates are agreed separately.

How much does each answer cost?

It depends on the model and on how much text goes in and out. With a small model priced at 0.75 USD per million input tokens and 4.50 USD per million output tokens, an answer using 3,000 input and 300 output tokens costs about 0.0036 USD.

Do I have to tell visitors they are talking to an AI?

Article 50 of the EU AI Act requires providers of AI systems that interact directly with people to design them so that people are informed they are interacting with an AI system, unless it is obvious from the context. We show it in the chat window from the first message.

What is the biggest hidden cost?

Keeping the sources current. If prices, hours or rules change and nobody updates the documents, the assistant repeats the old answer. Budget about an hour a month for this, more in busy seasons.

Contact

One person reads this inbox — the same person who builds what you order.

Describe your task

Tell us what you have and what result you need — even a single small task. No brief required: within 48 hours you get a plan and a price in writing, a question, or an honest “not a fit”.

Include the situation, files or links, deadline, language, privacy limits and budget boundary if you have one.

File (optional)
No file chosen

A brief, a photo, a spreadsheet — whatever explains the task faster than words. Up to 8 MB, stored with our hosting provider. Please don't send sensitive personal or medical documents through the site.

Sending creates no order and no payment. We reply within 48 hours. Prefer email? contact@castel.studio

Talk to us

Prefer a call or a messenger? Write or call — the studio's founder answers.

Connect on LinkedIn

Prefer a freelance platform?

Order the same work there, with the platform's secure payment.

A one-person AI studio in Nice. Whoever answers your message is the one who does the work — no account managers in between. Files are processed on our own machine, not in someone else's cloud.

Nice 06100, France · Open in maps →

SIREN 103 115 705

Describe your task