Custom instructions give the AI Agent the policy details, examples, and exceptions it cannot infer from setup fields alone.
When to add instructions
Add custom instructions when the AI Agent needs to follow a support policy, use a specific threshold, avoid certain cases, apply a preferred tag, or treat similar tickets differently.
Write the instructions
- Open the AI Agent setup or configuration page.
- Find the custom instructions, policy, or guideline field.
- Describe what the AI Agent should look for.
- Add examples of tickets where the AI Agent should act.
- Add exceptions where the AI Agent should not act.
- Save or continue after confirming the instructions are specific enough for review.
Good instruction pattern
Use direct rules such as:
- Apply the tag only when the customer explicitly asks for a refund.
- Do not escalate if the customer is only asking for a status update.
- Treat warranty dates in the next 30 days as urgent.
- Add a private note when more information is needed before action.
What happens next
The AI Agent has clearer rules for interpreting tickets, reducing incomplete output and unnecessary follow-up.
Test the AI Agent against realistic examples before deploying it broadly.
After writing the instructions, use Train and Test before deployment.