MrRightSite
Back to Blog

What an AI Receptionist Should Never Say

MrRightSite Team·August 18, 2026·AI Receptionist

Every conversation about AI on the phone is about what it can do. That is the wrong half.

The boundaries are the build. A system that will attempt anything will eventually attempt something that costs you a customer, a job, or in some trades a license. Defining the refusals is most of the real work, and it is the part that gets skipped when something is built in an afternoon.

Here are nine. Take them into any demo.

1. Quote a firm price for work nobody has seen

Ranges you have approved, yes. "That usually runs between $400 and $700 depending on what we find" is useful and honest.

"That will be $450" is a commitment, and a caller will hold you to it. A price stated by your system is a price you quoted, and arriving to say otherwise starts the job with an argument.

2. Promise an arrival time

"Someone will be there between two and four" is a scheduling decision, and scheduling depends on the job before it, traffic, and what the crew found this morning.

The system can offer a booked slot from your real calendar. It should not invent a window.

3. Give professional advice

Medical, legal, financial, insurance. Even when the answer seems obvious, and especially when the caller is anxious and pushing for one.

"I can't advise on that, but I can have someone who is qualified call you back" is the whole script, and it needs to hold under pressure. This is not caution for its own sake. In several trades it is the difference between a service and an unlicensed one.

4. Claim coverage, licensing, or credentials it cannot verify

Do you cover my town, are you licensed in my state, do you take my insurance, are you certified for this brand.

If the answer is not in what it was given, the answer is "let me check and have someone confirm." Guessing yes to be helpful produces a wasted trip and a bad review.

5. Take a card number, or sensitive personal information

Card details, social security numbers, medical history, dates of birth. Not by voice, not into a transcript.

If you need payment, send a secure link. A call recording containing a card number is a problem you have created for yourself, and the rules around handling that data are not written with small businesses in mind.

6. Claim to be a person

Directly asked, it says what it is. Immediately and without hedging.

Beyond the ethics, there is a practical reason: callers work it out anyway, and the ones who feel deceived are the ones who leave reviews about it: will customers hang up if an AI answers.

7. Keep going when it is out of depth

The most common failure and the most avoidable. A system that cannot answer should stop, say so, and hand off.

What it must not do is generate something plausible to keep the conversation moving. Confident and wrong is far more expensive than "I don't know."

8. Argue

Angry callers happen. A complaint, a billing dispute, something that went badly.

The correct behavior is to acknowledge, take the details, and escalate fast. Not to defend the business, not to explain policy, not to correct the caller's version of events. An argument between your customer and your software is a story that travels.

9. Handle an emergency alone

Every trade has a category where the right response is a human, now. Gas, water, injury, security, a vulnerable person.

The system should recognize those words and route immediately, and you should be able to state exactly what that trigger list is rather than trusting it to judge.

The question that tests a vendor

Ask them, without warning: what is your system forbidden to do?

A vendor who has done this work answers in ten seconds, because the list exists and they wrote it with a client. A vendor who redirects to capabilities has not written it down, which means the first time your system meets one of these nine, it will improvise.

You are not buying what it can say. You are buying what it will refuse to.


Related: what an AI chat rep is and how many calls is your business missing.

We write the refusal list with you before anything goes live, and for regulated trades we write it first. See AI Receptionist, or tell us the call you would least like it to mishandle and we will show you how it gets routed.

Common questions

What should an AI receptionist never do?

Never quote a firm price you cannot honor, never promise an arrival time, never give medical, legal or financial advice, never claim licensing or coverage it cannot verify, never take card numbers or sensitive personal data by voice, never claim to be human, and never keep improvising once it is out of depth.

Can an AI receptionist quote prices?

It can give ranges you have approved, framed as estimates. What it must not do is state a firm figure for work nobody has seen, because a caller will reasonably treat that as a commitment.

Is an AI receptionist safe for a regulated business?

It can be, but the rules have to be written into the build rather than added afterward. Medical, legal, insurance and financial businesses need explicit refusals and a documented escalation path, and that is a conversation to have before anything goes live.

How do I know a vendor has set boundaries properly?

Ask them to list what their system is forbidden to do, without preparation. A vendor who has done the work answers immediately. A vendor who talks about capabilities instead has not written the limits down.

Ready to turn your website into your hardest-working salesperson?

Professional, built around your business, and live in 5 days to 2 weeks.

Get My New Site Live