How to Train a Chatbot on Your Business Data
A well-built knowledge base is the difference between a useful chatbot and a frustrating one. Here's how to do it right.
An AI chatbot is only as good as the information you provide. Without a complete and well-structured knowledge base, the assistant will give generic answers or, worse, make things up. In this guide we'll look at how to build an effective knowledge base.
The 4 training methods
SmartChatAssistant offers four ways to feed your assistant's knowledge base. You can use them individually or combine them.
1. PDF document upload
Upload brochures, catalogs, price lists, technical manuals or any business document. The AI analyzes the content, indexes it and makes it available for responses.
Best practice: upload PDFs with selectable text (not scanned images). Documents of 50 pages or fewer give the best results.
2. Automatic web crawling
Enter your website URL and SmartChatAssistant will automatically scan public pages, extracting text, titles and structure. Ideal for sites with lots of existing content.
Best practice: make sure important pages are publicly accessible. Exclude login pages, cart and restricted areas.
3. Manual FAQs
Write question-answer pairs to control exactly what the chatbot says about specific topics. FAQs have the highest priority in the knowledge base.
Best practice: add variants of the same question ("How much does it cost?", "What's the price?", "What are your rates?") to improve matching.
4. Free text
Paste any text content: company policies, procedures, service descriptions, contact information. Useful for adding information that isn't in structured documents.
Best practice: use headings and bullet points to structure the text. The AI understands organized content better.
What to include in the knowledge base
Here's a checklist of content every business chatbot should have:
- Basic information: hours, address, contacts, how to reach you
- Products/Services: descriptions, prices, features, differences between options
- FAQs: the 20-30 most frequent questions you receive
- Purchase process: how to order, payment methods, delivery times
- Policies: returns, warranties, cancellations, privacy
- Differentiation: what sets you apart from the competition
Mistakes to avoid
- Too little data: a chatbot with 5 FAQs will give poor answers. Aim for at least 30-50 items in the knowledge base.
- Contradictory information: if a PDF says "price €49" and an FAQ says "price €59", the chatbot will get confused.
- Outdated content: update the knowledge base when prices, hours or services change.
- Sensitive data: don't upload documents containing customer personal data, passwords or confidential information.
Monitor and improve
After launch, use the Analytics section of the dashboard to identify:
- Unanswered questions: add the missing information to the knowledge base
- Negative feedback: review and correct answers that don't satisfy users
- Recurring topics: if a question is asked frequently, consider highlighting it with a dedicated FAQ
Build your knowledge base
PDFs, web crawling, FAQs and free text. All in one intuitive dashboard.
Start Free