AI Chatbot for Website: Costs, Best Tools and How to Add One the Right Way (2026)

An AI chatbot for website visitors can answer questions at 3 AM, book calls and take pressure off your inbox. However, a careless setup can also give customers wrong prices or invent policies. So you want a bot that answers from your own content, knows its limits and hands hard cases to a human.

Therefore, this guide covers real 2026 prices, the main tool types and a simple plan to launch safely.

How a modern AI chatbot for website support works

Old chatbots followed fixed scripts. In contrast, modern ones use a large language model, which understands normal questions and writes natural replies. The best setups add a step called retrieval augmented generation, or RAG.

With RAG, the bot first searches your own pages, help articles or PDFs for the right passages. Then it writes an answer based on those passages and can link to the source. As a result, answers stay closer to your real policies. AWS describes RAG as one approach to reducing inaccurate answers. It does not remove every error, so testing still matters. Our guide to retrieval augmented generation explains the idea in plain words.

AI chatbot for website: tool options and prices

In general, you have three broad choices: a no code builder, a chatbot inside your help desk, or a custom build on an AI API.

Prices come from each official pricing page in October 2026. For example, Intercom’s pricing page charges per outcome, while Botpress charges per conversation bundle.

What it really costs

There is no reliable “average” price, because usage drives the bill. Instead, work out your own number.

For example, say your site gets 1,000 support chats a month and the bot solves 600 of them. With a per outcome tool like Intercom Fin, that is about 600 × $0.99, or roughly $594 a month. You also pay for seats if you use Intercom’s own help desk, but Fin can run on your existing help desk with no seats. In contrast, a custom build on an API often costs far less in model fees. For instance, mid tier models from OpenAI and Anthropic cost about $2 per million input tokens and $10 per million output tokens. A typical support chat uses a few thousand tokens, so model costs can stay in the tens of dollars. However, someone has to build and maintain the system, and that is the larger cost.

In short, no code tools cost more per chat but less in setup time. Custom builds cost less per chat but need skilled work.

When to build a custom chatbot

A custom AI chatbot for website use makes sense when you need things a template cannot do. Examples include answering from private product data, taking bookings in your own system, or matching your brand exactly. It also helps when you want full control over where customer data goes.

In addition, a good build includes:

1. A clean knowledge base built only from checked, current pages.
2. A retrieval layer that finds the right passages and cites them.
3. Clear rules about topics the bot must refuse, such as legal or medical advice.
4. A handoff button that sends the chat to a human with the full history.
5. Logs and a weekly review of wrong or weak answers.

If you would rather not build it yourself, AllInfo AI offers chatbot and RAG building services for small businesses worldwide.

Rules you must follow

Tell people they are talking to AI. Under Article 50 of the EU AI Act, people must be told they are dealing with an AI system unless it is obvious. This duty applies from 2 August 2026. A short line in the first message, such as “Hi, I’m the AI assistant for this site,” covers it.

You are responsible for what your bot says. In Moffatt v. Air Canada (2024), a Canadian tribunal rejected the airline’s claim that its chatbot was a separate party. The airline had to pay back the fare difference that its bot had wrongly promised. Therefore, never let a bot invent discounts, prices or policies.

Protect personal data. After all, chat logs often contain names, emails and order details. Under GDPR, you need a lawful basis to process them, and your privacy policy should mention the chatbot. In addition, avoid free API tiers for customer data, since some providers use free tier data to improve their products.

Make it accessible. The chat widget should work with a keyboard and screen readers, in line with WCAG 2.2.

How to launch an AI chatbot for website visitors

1. Plan. List the 20 questions customers ask most, and decide what the bot must never answer.
2. Train. Load only pages you have checked. Remove outdated offers first.
3. Test. Ask at least 50 real questions, including tricky and off topic ones. Check every cited source.
4. Launch. Add the AI notice, link your privacy policy and switch on human handoff.
5. Improve. Review conversations every week and fix the content behind wrong answers.

Watch a few numbers after launch: resolution rate, handoff rate, customer satisfaction and cost per resolved chat. If the bot also needs to take actions, such as booking a slot, read our guide to AI agents. For wider workflows, see AI automation for small business.

FAQ

How much does an AI chatbot for a website cost?

No code tools range from free plans with tight limits to about $30 to $150 a month, while support platforms charge around $1 to $2 per resolved chat. Custom builds have low model costs but higher setup and maintenance costs.

Can I train a chatbot on my own website content?

Yes. Most tools can read your pages, help articles or PDFs and answer from them using retrieval. Keep that content accurate and re index it after every update.

Do I have to tell visitors they are chatting with AI?

In the EU, yes, since 2 August 2026 under the AI Act, unless it is already obvious. Elsewhere it is still good practice and builds trust.

Is a company liable if its chatbot gives wrong information?

It can be. In the Air Canada case, the tribunal held the company responsible for its chatbot’s answer and ordered it to pay. Treat the bot’s replies as your own statements.