How to train a chatbot on your own website

"Train the chatbot on my website" sounds like the AI studies your site and learns your business. In practice, for a business assistant, it means something simpler and safer: your website's facts are written down as approved knowledge, and the assistant answers from that text.

What "training" really means for a business chatbot

The AI model behind an assistant already knows how to read and write. It does not need to be retrained on your website, and it shouldn't be: a model that has "learned" your prices can't easily forget them when they change. Instead, the assistant is given a block of text, your knowledge, and told to answer only from it. Change the text and the answers change with it.

So the real work of "training" is producing good knowledge: accurate, complete and current. Your website is a natural starting point, because it already holds most of what customers ask about.

The website scan: up to 10 pages, your domain only

In step 2 of the Allinaa setup wizard you can ask it to read your website. The scan works within tight limits:

• It reads at most 10 pages: your homepage and pages linked from it.
• It stays on your own domain. A link to another site is never followed, and a redirect to another domain is not followed either.
• It honors each host's robots.txt. If your site tells crawlers to stay out of a page, the scan skips it.
• It reads only the public text of the pages. It does not sign in, fill in forms or read anything behind a login.
• It has a time and size budget per scan, so a very large site is read only in part.

The text of those pages is sent to the AI provider to write the draft. The wizard tells you this before you start.

Facts only: no invented prices

The draft is written with a strict instruction: use only facts that appear on your pages. If your site doesn't state a price, the draft doesn't contain one. If your hours aren't on the site, they won't be in the draft. That is deliberate. A gap you can see and fill is far better than a plausible number the AI made up.

The AI can still get a fact wrong, for example by mixing up two services or misreading a table. That is why the next step matters.

Reviewing the draft before you press "Use this"

The draft appears in an editor. Nothing is applied until you press Use this. Read it with these questions in mind:

1. Is every service correct and current? Remove anything you no longer offer.
2. Are prices stated the way you want them quoted? Exact, as a range, or "quote required".
3. Are hours, address and service area right? Old pages often hold old hours.
4. Is anything there that you don't want repeated? An old promotion or a staff name, for example.
5. What is missing? See the next section.

After you press Use this, test the assistant in the wizard's test chat with questions your customers really ask.

Filling gaps your website doesn't cover

Websites are written to sell, not to answer every question. Common gaps the scan can't fill:

• What happens if a customer is late, cancels or needs to reschedule.
• Payment methods and whether a deposit is needed.
• Parking, accessibility and how to find the entrance.
• What you don't do, so the assistant can say no politely instead of guessing.
• When the assistant should hand the visitor to a person.

Add these directly in the editor. The guide on writing chatbot knowledge has a template you can copy.

Rescanning after site changes

When you redesign the site or add a service, you can run the scan again, up to 3 times a day. Each scan produces a new draft; it never overwrites your current knowledge by itself. Compare the draft with what you have, copy over what's new, and keep your manual additions.

If the Allinaa team set up your assistant, knowledge changes go through them: Standard includes up to 2 hours a month of knowledge updates and response improvements.

Once the draft is in place, the knowledge base guide shows how to fill the gaps, and the launch checklist lists what to test before going live.

Frequently asked questions

Does the scan read pages behind a login?

No. It reads only public pages on your own domain, up to 10 of them, and it respects robots.txt.

Will the assistant answer from the whole internet?

No. It answers from the knowledge you approved. The scan only helps you write that knowledge faster.

Can I skip the scan and write the knowledge myself?

Yes. The scan is optional; you can type or paste your knowledge into the editor instead.

How often can I rescan?

Up to 3 times a day. Each scan makes a new draft and changes nothing until you press Use this.