Adding a chatbot to your website: cost and when to skip it

en 2026-09-28 6 min read

Adding a chatbot to the website you already have is an add-on at EUR 750-2.000, with the quote confirming the scope. Built into a new site, it puts that site at step 04: EUR 3.500-7.500 and 6-12 weeks. If your visitors ask the same handful of questions, a well-written FAQ section does the job for far less, as an extra simple page at EUR 150-250 on your current site or on a single page at step 01, EUR 600-900.

A chatbot is worth its price when the answers depend on the person asking. When they do not, it is an expensive way to show text you could have written on the page.

What a chatbot actually does for a small business

Strip away the demo and a website chatbot does four jobs:

• It answers questions phrased any way, from content you have approved: your services, opening hours, delivery terms, what a first appointment involves.
• It answers when nobody is at the desk, in the evening and at weekends.
• It looks things up, such as an order status or open appointment slots, if it can ask another system through an API.
• It hands over to a person, by collecting a name, an email address and the question, and passing them to your inbox or CRM.

What it cannot do is know things you have not given it. A chatbot built on an AI language model can also state a wrong answer with confidence. So a useful chatbot is built around limits: what it may answer, what it must refuse, and when it passes the conversation to you.

Why a chatbot is more than a chat window

The chat window is what visitors see. Behind it sits a small application, which is why a chatbot in a new build belongs with the other step 04 work: logins, dashboards, API and CRM connections, and payments. That application has six parts:

• A source of answers. The pages, documents or product data it is allowed to draw on, kept current as your offer changes.
• Rules. Topics it answers, topics it refuses, and what it says when it does not know.
• Lookups. Live information, such as bookings or orders, comes from another system through an API.
• A handover. Where a conversation goes when a person needs to take it: an email, or your CRM through its API, which works the way connecting your website to a CRM describes.
• Logs. A record of conversations, so you can see what people asked and where the answers fell short.
• Testing. Your real questions, asked in different ways, including the ones it should refuse.

What moves the price

The range is wide because chatbots differ in what they are allowed to touch. Scoping answers these questions:

• Does it answer from a few pages, or from a large catalogue or set of documents?
• Does it only answer, or does it also look things up and take actions, such as creating a lead or a booking?
• How many systems does it connect to?
• How many languages does it need to answer in?
• Where do handovers go, and who answers them?

There is also a running cost. The AI model behind a chatbot has usage charges from the model provider, and how those are paid is agreed at scoping. Ask about it before you commit, whoever builds it.

Chatbot, FAQ section or contact form

An FAQ written as plain text does a second job: search engines can read it, and so can AI assistants that crawl your pages. A chatbot's answers are invisible to both. Getting cited by AI assistants explains why written answers on the page matter there.

A test before you decide

Collect the questions you have answered by email and phone recently and sort them into two piles. General questions, where the answer is the same for everyone, belong in an FAQ section. Personal questions, where the answer depends on an order, a booking or an account, are the case for a chatbot with an API behind it. If the second pile is small, you have your answer.

Process and timeline

1. Scoping. Your real questions, the sources the chatbot may use, what it must refuse, the systems it looks things up in, and where handovers go.
2. A fixed quote in writing, for the add-on or for a new step 04 build.
3. Source content. The material the chatbot draws on has to be approved by you. This is the part that waits on you, and the chatbot can only be as accurate as that material.
4. Build. The chat interface, the connection to the model and your content, any API lookups and the handover.
5. Test. Your questions asked in different ways, plus questions it should refuse.
6. Launch and review. Read the conversation logs and adjust the content where answers fall short.

A new step 04 build runs 6 to 12 weeks. You own the code, the repository and the database, and hosting is agreed per project. Afterwards, Care at EUR 100-300 a month covers updates, monitoring, backups and small fixes. Teaching the chatbot a new lookup is a new feature, quoted as one. How we design and build is set out on design and engineering.

When to spend less

• The questions repeat. Write an FAQ section. On a single page it fits into step 01 at EUR 600-900, and on a larger site it sits among the pages you already have.
• The answers are not on your site yet. A chatbot has nothing to say about services you have not described. Fix the service pages first: an extra simple page is an add-on at EUR 150-250, and a business website with proper service pages is step 03, EUR 1.500-3.500.
• You want people to reach you quickly. A short contact form, a phone number, or a link to the messaging app you already answer does that without a bot.
• Nobody visits the site. A chatbot does not bring visitors. The SEO foundation, a one-off at EUR 450-750, deals with whether search engines can find and index the site at all.

In each of these cases, do not buy a bigger build yet.

Common questions

Will a chatbot take enquiries off my desk?

It can take the general questions, as long as the answers exist in content you have approved. Anything that needs judgement, such as a price for a specific job or a complaint, should reach a person, and that handover is part of the build. You still read what it passes on.

Can it give wrong answers?

Yes. A chatbot built on an AI language model can state something wrong with confidence. Limits reduce the risk: it answers only from approved content, says when it does not know, and hands over. Testing against your own questions before launch is part of the process.

Does a chatbot help my search rankings?

Not directly. Search engines read the text on your pages, not conversations inside a chat window. If you want answers that can be found, write them on the page as well.

What about privacy?

Conversations can contain names, email addresses and details about someone's situation, which makes them personal data under the GDPR. Decide how long transcripts are kept and say so in your privacy notice. GDPR and your website covers the wider picture.

Send us the questions your customers ask, and we will tell you whether they need a chatbot or a better FAQ. Get in touch here.