Ever added more files to your AI Chatbot and still thought: “Why is it answering like that?” 
Sometimes the chatbot already has the facts, but needs a clear direction for what to do next. We talked about this in our Best Practices for AI Chatbot Training guide, and the idea is simple:
Knowledge Base = what the chatbot knows
Conversation Rules = how the chatbot should act
For example, if pricing depends on custom details like plan, quantity, usage, or special conditions, more pricing info may not help much.
What the chatbot really needs is a clear rule for how to handle that situation. Here’s an example:
See a Conversation Rule template
The template below shows a recommended structure for most Conversation Rules:
## [Title]
When [situation or user intent], you must [main instruction].
You should:
- [allowed or expected behavior]
- [allowed or expected behavior]
You must not:
- [restricted behavior]
- [restricted behavior]
If [exception, missing information, or unclear request], you should [fallback action].
Examples:
- User: [example user question]
You: [desired response pattern]
This example follows the recommended structure and uses Markdown to organize instructions clearly:
This helps the chatbot avoid guessing, overpromising, or giving a confident answer when more context is needed.
The same idea can work for refunds, delivery times, availability discounts, or any case where the chatbot should be careful before giving a final answer.
[!question]
What’s one rule or tweak that made your AI Chatbot answers more reliable?
