The Questions Your AI Receptionist Must Never Guess At

A caller asks whether your company can send someone tomorrow at 9 a.m. The AI receptionist wants to be helpful. But the calendar is unavailable. If it says yes anyway, the caller may rearrange their day for an appointment that does not exist. A confident voice does not make an unverified answer true. The most important feature of a reliable AI receptionist may be knowing when not to guess.

1. Never invent availability

An AI can offer confirmed appointment slots only when an authorized scheduling system returns them. If the calendar is disconnected, it should collect a preferred time as a request, explain that it is not confirmed, and route it for follow-up. A booking is complete only after the connected system verifies it.

2. Never guarantee an unapproved price

Businesses may publish fixed prices for clearly defined services. But estimates that depend on inspection, scope, parts, or professional judgment should be handled according to approved pricing rules. A receptionist should describe the estimate process instead of improvising a quote to keep a caller interested.

3. Never claim a payment or refund succeeded without confirmation

If a payment processor does not return a verified result, the AI should not announce success. It should direct the caller to an approved payment channel or staff member and avoid collecting sensitive card details in an insecure conversational workflow. The FTC recommends limiting personal data collection and securing information businesses retain. Source: https://www.ftc.gov/business-guidance/resources/protecting-personal-information-guide-business

4. Never pretend to be a licensed professional

Legal, medical, insurance, and technical advice can require professional judgment and authorization. A receptionist can explain office procedures and route questions, but should not impersonate a lawyer, clinician, insurance producer, or qualified technician. State clearly that the caller is interacting with an AI assistant and provide an appropriate human path.

5. Never make up company policies

A customer asks about cancellation fees or service guarantees. If the approved knowledge base has no answer, the agent should acknowledge uncertainty and connect the customer with someone who can verify the policy. Invented promises can become expensive disputes. Maintain versioned policies and test changes before launch.

6. Never ignore a request for a person

When a caller asks for a human, offer a real transfer or documented callback process. If no one is available, be clear about what happens next. Do not promise a callback time that the team cannot support.

Design guardrails as an operating process

For every high-stakes question, define the trusted source, permitted answer, escalation trigger, and fallback when the source is unavailable. Test scenarios such as failed calendar connections, disputed charges, sensitive disclosures, and unusual customer requests. The NIST AI Risk Management Framework provides voluntary guidance for managing AI system risks, including ongoing monitoring and governance. Source: https://www.nist.gov/itl/ai-risk-management-framework

A website agent needs the same honesty

Sensoji Auto Agent is an AI voice agent embedded on a business website. It can proactively engage visitors, qualify inquiries, guide people toward relevant services, and help sell through configured workflows. Visitors should be able to mute, stop, or dismiss it. Like a phone receptionist, it should rely on approved information and verified actions rather than guessing. Explore https://www.sensoji.co. Customers do not need an AI that always has an answer. They need one that knows which answers it can stand behind.