Use the AI Builder to generate a rule

The automation assistant turns a plain-English description of what you want into an automation rule you can review and edit on the same page. Start with a new rule or open an existing one, show the assistant, describe the change you want, then review the resulting When / Only if / Then rule before you save or enable it.

The automation assistant is currently in beta. The shape of the editor and the quality of the drafts are still evolving. The output is always editable in the rule builder before you save.

Before you start

• Admin permissions to edit automation rules
• A clear answer to “what do I want to happen, and when?” — the better the prompt, the better the draft

Steps

1. Open an automation rule. Go to Settings → Automation Rules and either click Add rule or open an existing rule you want to change.

2. Show the assistant. In the rule editor, open the assistant from the sparkles button. You can use it before you have built the rule manually, or after you have already made manual edits.

3. Describe the outcome, not the mechanics. Type what you want in plain English. The good pattern is to describe behavior, not config:

   
   Good: “When a customer emails us, send them the order confirmation template in their own language.”
   Less good: “Create a trigger on conversation.created with a condition channel equals email and an action send_email_template templateId=42.”

The AI knows the schema; you don’t have to. Describe the situation and the desired outcome.

4. Let the assistant update the rule. The assistant updates the on-page rule with the trigger, conditions, and actions it thinks match your request.

5. Review the rule. Read each section in the builder:

   
   When — is the rule firing on the right event?
   Only if — are the field/operator/value choices right?
   Then — does it use the template, tag, team, or other action you actually wanted?

The builder only keeps actions the chosen trigger can actually run, so some requested actions may be omitted when the trigger lacks the needed object context.

6. Edit anything wrong. Anything the AI got wrong, fix directly in the same When / Only if / Then rule. You can also ask the assistant for another change after manual edits.

7. Undo if needed. If the assistant changes the rule in a way you do not want, undo the change and continue editing manually or try a more specific prompt.

8. Test before enabling. Use recent conversations to test whether the rule fires the way you expect before turning it on.

9. Save and enable. Once you’re satisfied, save the rule. New rules start turned off, so toggle the rule on when you’re ready for it to run.

The inverse — Explain Rule

The same AI surface works the other way. On any existing rule, click the sparkles button (Explain this rule) and the AI describes what the rule does in plain language. Useful for:

• Inheriting rules from a previous admin
• Auditing what’s actually live before a change
• Writing down what a rule does in plain words before disabling it

Prompt patterns that work well

• When/then — “When a VIP customer emails us, route to Premium and add a VIP tag.”
• Wait/then — “Three hours after a conversation is resolved, send a CSAT survey.”
• Schedule — “Every weekday at 9am, send a reminder to #support-leads in Slack to review open escalations.”
• Escalate — “When an SLA is at risk, ping #support-leads in Slack and assign a senior agent.”

Troubleshooting

• Symptom: The AI generated the wrong trigger. Fix: Add specificity to the prompt about when the rule should fire. “When a new email arrives” → trigger A customer sends a message with condition channel=email.

• Symptom: The action references a template/tag/team that doesn’t exist. Fix: Create the missing entity first (Email Studio, Tag Management, Teams) and re-prompt with the exact name.

See also

• Create your first automation — the manual path, when you know exactly what you want
• Simulate a rule before enabling