Rules govern how the assistant communicates. Knowledge governs what it knows. Keep the two straight and everything gets easier.
Rules guide every reply, regardless of which knowledge was retrieved: tone of voice, things it must always say, things it must never say, and when to bring in a person. Leave any field blank to use the default behaviour.
Set them under Knowledge → Chatbot Rules.
What belongs in a rule rather than in knowledge?
| Put in rules | Put in knowledge |
|---|---|
| "Be warm but brief. Never use exclamation marks." | Your opening hours |
| "Never quote a price — always offer a callback." | Your price list |
| "Always mention that the first consult is free." | What the consult covers |
| "Never give clinical advice." | Your service descriptions |
A good test: if it would still be true after you replaced all your content, it's a rule.
Can I start from a recommended set?
Yes. Click Apply industry starter to fill blank fields with recommended defaults for an industry. It never overwrites something you've already written — it only fills what's empty.
If you've run a Questionnaire, we pre-fill recommended rules for your industry based on the scan. Those are suggestions: nothing is active until you review and Save rules.
How do I know a rule is working?
Use the Preview console at the bottom of the page. Type a sample visitor message — "Do you offer refunds?" — and click Preview reply. It runs against your live knowledge base and your current rules, including unsaved edits, so you can iterate on wording before committing.
Saved rules apply to new replies immediately.
What else shapes the assistant's behaviour?
Three settings that live near the rules and are worth setting at the same time:
- First-contact introduction — how it opens a brand-new conversation.
- Escalating to a human — topics it should never try to answer itself.
- Crisis safety and emergency escalation — for intake in sensitive verticals, and for real-world emergencies.