How to train an AI chatbot on your website (without writing a single prompt)
A practical walkthrough of website crawling, documents, and FAQs, and how to tell when your AI knows enough to go live.
HHiberChat team · Nov 2, 2026 · 6 min read
The fastest way to get a useful AI agent is to stop thinking about prompts and start thinking about sources. Your AI can only be as good as the content you give it, so most of the work is choosing and tidying that content.
Start with the pages customers already read
Your shipping page, returns policy, pricing page, and FAQ answer most questions before anyone types them. In HiberChat, open your chatbot, choose Website in the knowledge menu, paste your address, and click Learn site. HiberChat reads up to 20 pages per scan, follows links two levels deep, uses your sitemap, and skips anything blocked by robots.txt.
- Crawl your main site first, then your help center as a separate source.
- Text-heavy pages work best. Pages that load all their text with JavaScript may come back empty.
- Re-sync a source after you update the page so the AI sees the new version.
Upload what isn't on your website
Price lists, internal policies, menus, and product manuals often live in PDFs or Word files. Upload them as documents. They're split into passages and searched the same way as web pages.
Write FAQs for answers that must be exact
If an answer has a legal or financial edge — refund windows, cancellation fees, warranty terms — write it as a question-and-answer pair. FAQ pairs are short, precise, and match strongly.
Test before you publish
Use the preview panel to ask the ten questions you get most. Check the answer and the sources it used. If the AI says it doesn't know, that's good: it means it isn't guessing. Add the missing content and ask again.
Publish, then keep an eye on handoffs
After you publish, chats the AI hands to your team are your to-do list. Each one is a question your content doesn't answer yet.