Your AI chatbot learns from content you provide. Train it on your entire website and supplement with Q&A pairs, text snippets, files, and product data. The AI is smart enough to find the right information for each visitor question, so there's no need to cherry-pick pages.
How to get there: Go to Setup → Chatbot in the top menu → click your chatbot → expand the Training section in the sidebar.
Your chatbot can learn from several types of content. Each source type has its own tab in the sidebar:
| Source | Best for | Example |
|---|---|---|
| Website | Your entire site: FAQ pages, product pages, documentation | Crawl your domain or import your sitemap |
| Files | Large documents, spreadsheets, bulk data | PDF manuals, product catalogs, resource directories |
| Text Snippets | Policies, procedures, structured reference content | Return policy, service area details, operating hours |
| Products | Structured product catalog with pricing and availability | E-commerce inventory, service packages |
| Tickets | Real agent replies turned into training data | Save a great support reply so the chatbot learns from it |
| Q&A | Specific questions with specific answers | "What is your address?" with the exact answer |
| Exact Answers | Override or correct a specific chatbot response | Fix a wrong answer so it never happens again |
Start with your website. Import your sitemap or crawl your domain, and train all discovered pages. This gives the chatbot broad knowledge about your business.
Add Q&A pairs for core business facts that visitors ask about most: address, phone number, opening hours, pricing. These are focused entries that the chatbot finds very reliably.
Use text snippets for content that isn't on your website: internal policies, detailed procedures, or instructions for how the chatbot should handle specific situations.
Upload files when you have existing documents (PDFs, spreadsheets) that contain useful information.
Add products if you have a catalog with pricing, availability, and descriptions.
Tickets let you capture great agent replies as training data, directly from the inbox.
Exact answers are for corrections: when the chatbot gives a wrong answer, you override it with the correct one.
Some pages on your website are great for SEO but harm your chatbot. Anything that mentions competitors or describes your product's weaknesses gets repeated back to prospects at the worst moment.
Easy rule: if the URL or H1 contains another company's name, exclude it.
When multiple sources match a question equally well, curated sources (Q&A, text snippets, products) get a small priority boost over scraped content (website pages, files). See Source priority for details.
Training uses a credit system based on content size. Every 5,000 characters of cleaned content counts as 1 page credit (minimum 1 credit per source). A short FAQ page uses 1 credit, while a long product guide might use 2 or 3. Each trained page or file shows its credit cost in the list.
| Plan | Page Credits | Crawl Discovery Limit |
|---|---|---|
| Free / Trial | 100 | 500 |
| Lite | 100 | 500 |
| Starter | 1,000 | 5,000 |
| Growth | 3,500 | 5,000 |
| Business | 10,000 | 5,000 |
| Scale | 15,000 | 5,000 |
If you reach your credit limit during training, remaining items are marked as "Limit reached". Upgrading your plan automatically requeues them.
For more details on managing credits, see Training credits.
Once you've added training sources, these tools help you keep things accurate and organized:
The most common mistake is providing bare facts without context. Your chatbot needs to know not just what the facts are, but how to use them.
Instead of:
Mon-Fri 9-5
Write:
Our office is open Monday through Friday, 9 AM to 5 PM. Our 24/7 helpline is available outside office hours. If someone contacts us outside office hours, direct them to the helpline.
Instead of:
$49/month, $99/month, $199/month
Write:
We offer three plans: Starter at $49/month, Professional at $99/month, and Enterprise at $199/month. If someone asks about pricing, share all three options and suggest they visit our pricing page for full details.
Give the chatbot context about what to do with the information, not just the raw data.