What an AI Voice Agent Should Never Say or Do

What an AI Voice Agent Should Never Say or Do

An AI voice agent represents your business on every call it answers. Clear limits make it useful and keep it from causing problems.

Last reviewed: November 2026.

It should never

1. Pretend to be human. It should say it is an automated assistant.
2. Give professional advice. No medical, legal, financial or safety advice beyond instructions you approve, such as directing emergencies to 911.
3. Promise what it cannot guarantee. No arrival times, prices or outcomes outside the approved content.
4. Offer discounts or exceptions. Those decisions belong to a person.
5. Collect more than it needs. No payment card numbers, government IDs or health details on the call unless your setup is built and reviewed for it.
6. Discuss other customers. Ever.
7. Argue. If a caller is upset, it apologizes and hands off.
8. Keep someone stuck. There is always a way to reach a person or leave a message.

It should always

• Say what will happen next, and when.
• Confirm details back to the caller.
• Log the call and pass context to whoever follows up.

Test the limits

Before launch, call it and try to break each rule. Fix the call flow and approved answers until it holds.

How we help

We write these guardrails into every voice agent we configure. See AI voice agents, after-hours call handling and call recording and consent.