How to Train an AI Chatbot on Your Own Website Content

7 min read • Updated October 2026

QUICK ANSWER

Build the agent from your website, then fill the gaps: write plain sentences in Knowledge, add prices and hours as key facts, upload text PDFs and review unanswered questions. Re-read your site after every change. Up to 160 knowledge blocks per agent are available.

“Training” a business chatbot is not machine learning. It is writing down what your customers need to know in a form the bot can use. Good training is mostly good writing.

Start with your website

Paste your site link and the agent reads your homepage and the most useful linked pages first: FAQ, pricing, services, contact and downloads. If you have a sitemap.xml it can read up to 30 pages. Make sure those pages contain real text, not only images, and open without a login.

Write Knowledge as full sentences with the facts inside

The bot retrieves short passages, so each passage should stand alone. Write “Teeth whitening costs $250 per session and takes 45 minutes” rather than “Whitening: see price list”. Include units, currency, days and conditions in the same sentence.

Cover the questions customers ask repeatedly: price, duration, location, hours, how to book, payment, cancellation, delivery and warranty. One topic per block makes retrieval more accurate.

Use key facts for anything exact

Key facts are structured entries (type, name, value) for prices, hours and contact details, up to 60. Questions about them are answered exactly from the fact and they override your website text. If your website still shows an old price, the key fact wins.

PDFs and scans

Upload .txt, .md or .csv files, or PDFs with selectable text. A scanned PDF is a picture of text with no text layer, so run it through the free PDF to Text tool first and paste the result. Tables in price lists are best converted to one sentence per line.

Keep it fresh

After a price or page change, press Re-read my website. It replaces text you added by hand, so re-add your Knowledge afterwards, which is another reason to keep exact items in key facts. A recrawl reminder can prompt you every 30, 90 or 180 days.

Then close the loop with the unanswered-questions list in Overview. Each entry is a gap. Add the answer once and it is fixed for good.

QUICK TIPS

  • ✓ One topic per Knowledge block, with the number and unit in the same sentence.
  • ✓ Exact data goes in key facts.
  • ✓ Convert scanned PDFs to text before uploading.
  • ✓ Re-read your website after changes and re-add manual text.

Frequently asked questions

How much can I teach it?+
Up to 160 knowledge blocks per agent; the website crawl uses at most 100 of them so there is room for your own text.
Can it read a PDF price list?+
Yes if the PDF has selectable text. Scanned PDFs must be converted to text first.
What if my website is only images?+
Create the agent from any page with text and paste your details into Knowledge.
Does it learn from customer chats automatically?+
No. It improves when you add answers for questions it could not answer. That keeps you in control of what it says.

Sources and further reading

Cite this page: Clipaist Editorial, "How to Train an AI Chatbot on Your Own Website Content", Clipaist, October 2026. https://clipaist.com/guides/train-ai-chatbot-on-your-website-content/

P

Clipaist Editorial

We test every AI tool and prompt before publishing. Questions or corrections? Reach us on the Contact page.

Keep reading

Your next viral post
starts with one sentence.

Free. No signup. Ready in seconds.

Try Text to Image — It's Free →

Running a business? Put a free AI agent on your website →