Answers

What should an AI assistant never say about your business?

Ascent Vision · Published October 28, 2026

An AI assistant should never state a price or discount you have not agreed, promise a result or a date, give legal, medical or financial advice, comment on competitors, share another customer's details, discuss your staff, or invent a policy. Write that list down before it talks to customers, and give it a plain line to use instead, such as passing the question to a person. A written list lowers the risk; it cannot remove it.

The do-not-say list

• Prices or discounts you have not agreed. It may repeat a price you have published. It should not work one out, round it or offer a deal.
• Promises about results or timing. A line such as "we can be there tomorrow" or "that will fix it" is something a customer may hold you to.
• Legal, medical or financial advice. It can say it cannot advise on that and suggest asking a qualified professional.
• Claims about competitors. No comparisons, no opinions about other businesses, and no statements about what they charge or do.
• Personal details of other customers. No names, addresses, job details or purchases, even when the question sounds harmless.
• Anything about your staff. No comments on someone's pay, schedule, home or the reason they left. If a customer needs a particular person, it can pass the request on.
• Policies you do not have. If refunds, warranties or cancellations are not written down, it should not make up an answer.

What it should say instead

A list of "never" lines leaves a gap unless each one has a replacement. Write the replacement next to the item, in your own voice. A few that often work:

• "I'll have someone confirm that and get back to you."
• "That is a question for the owner. I can pass it on."
• "I can't advise on that. A qualified professional can."
• "I don't have that information."

Writing the list so it works

Use short, plain sentences and say what the assistant may answer as well as what it may not. Put the list wherever the assistant reads its instructions; the person setting it up can tell you where that is. Review it whenever your prices, hours or policies change, because an old list can be a source of wrong answers too.

Then test each item in several ways, including pushy ones such as asking for a discount or another customer's number. Any callback or text it promises goes out only with the person's consent.

What a list does not do

You can reduce the risk, not remove it. An assistant can still slip, so keep a person on anything a customer could hold you to. The related answer on stopping an AI assistant from making things up covers keeping it to approved information, letting it say it does not know, and testing it. The related answer on which calls an AI receptionist should take covers hand-off rules for phone calls, and the related answer on whether a person should approve what an AI system sends to customers covers when a person reads messages first.

Related answers

• How do you stop an AI assistant from making things up?
• Which calls should an AI receptionist take, and which should a person?
• Should a person approve what an AI system sends to customers?