AI voice agent prompts: specify the job, boundaries and handoff

A practical prompt structure

1. Role: define the business role and conversational objective.
2. Scope: state the services or requests the agent may handle.
3. Knowledge rule: answer business-specific questions only from approved sources.
4. Qualification: collect the minimum information needed for the next action.
5. Actions: define when booking, transfer, message or CRM update is appropriate.
6. Boundaries: identify questions that require a human or should not be answered.
7. Recovery: say what to do when speech is unclear, a tool fails or no appointment is available.

Do not write a telephone screenplay

Long scripts can make an AI sound rigid and can conflict with the agent’s ability to respond to unexpected wording. Give the system clear objectives and rules, then use examples for especially important edge cases.

Ground facts in a Knowledge Base

Prices, hours, policies and service details should come from maintained source material rather than being embedded in a prompt that nobody remembers to update. HighLevel supports Knowledge Bases that can be shared with Conversation AI and Voice AI workflows.

Test adversarially

• Ask for a discount that does not exist.
• Give incomplete contact information.
• Change appointment type mid-call.
• Request a human repeatedly.
• Ask a question outside the business scope.
• Provide a date with no availability.

Optimize for outcome quality

The best prompt is not the one that produces the longest conversation. It is the one that consistently moves valid callers toward the correct business action while exiting gracefully when the system should not continue.