An AI that answers customers is speaking for your business. Every word carries your name. Most of what it says will be routine: confirming a time, answering hours, acknowledging a message. The risk sits in the small number of things it should never say at all, because a wrong answer there costs money, trust or a legal problem. This article lists them. In a Staffhand build these are not guidelines. They are rules written into every desk, backed by a person who reads each message before it goes out.
A price, an estimate, a trade value or a financing term
This is the first rule and the one that matters most. An AI should never quote a price, an estimate, a trade-in value, a rate, a payment or anything that depends on credit. Not a range, not “roughly,” not “usually around.”
The reason is simple. A number given by a machine, without the context a person would have, is a number the customer will hold you to. Dealerships have learned this in public. Contractors learn it when a homeowner says “but your text said.” In every Staffhand build, these questions go to a person. The desk acknowledges the question, says who will follow up, and routes it with the details. Why AI should never quote price, trade value or financing explains the rule in depth.
A promise about an outcome
“This will fix it.” “You will lose the weight.” “Approval is no problem.” “We can definitely have it done Friday.” Each of these is a promise, and the AI has no authority to make it. A desk can say what the business offers and what the next step is. It cannot commit the business to a result or a deadline that a person has not confirmed.
Medical, legal or financial advice
A med spa desk can book a consult. It cannot say whether a treatment is right for someone or what the result will be. A dental desk cannot diagnose a toothache by text. A dealership desk cannot say whether a customer will qualify. A property desk cannot interpret a lease. In a Staffhand build for a med spa or dental practice, clinical questions go to staff, every time. The same applies anywhere a wrong answer could harm the person asking.
A guess dressed as a fact
When an AI does not know, the worst thing it can do is sound sure. “Yes, we have that in stock.” “Your order shipped yesterday.” “The class is at six.” If the fact is not in the business’s systems, the desk does not state it. It says it will check and routes the question. A status update desk reads the real record before it drafts. A guess that turns out wrong costs more than a short wait.
Anything the customer did not consent to receive
A customer who gave a phone number for an appointment did not agree to marketing texts. An AI should never send a message the business does not have consent to send, never ignore a STOP, and never text outside the quiet hours the business set. These are compliance rules and they are courtesy rules. How approval queues keep business texting compliant covers how a build enforces them.
A pretense that it is a person
An AI should not claim to be a named employee or deny being automated when asked directly. In a Staffhand build, messages are drafted by a desk and approved by a real person, so the honest answer is that a person at the business reviewed and sent it. That is true, and the build never says anything that is not.
Anything about another customer
Names, appointments, balances, vehicles, treatments. An AI should never reveal one customer’s information to another, even by accident, even when asked nicely. Desks work from the record of the person they are talking to and nothing else.
Anything nobody has read
This is the rule that enforces all the others. In a Staffhand build, nothing customer-facing goes out unapproved. Every draft lands in a single approval queue where a staff member approves it, edits it or holds it. The approver sees the context. If a draft crosses one of the lines above, it does not go out, and the rule gets tightened. The live action log records what was sent and who approved it.
People at Staffhand review what the desks do as well. Nothing runs unattended. The approval queue is described on How it works.
Frequently asked questions
If the AI cannot answer price, what does it say when a customer asks?
It acknowledges the question, says a named person will follow up, and routes the question with the customer’s details. That acknowledgement is itself approved before it goes out.
Can the AI tell a customer about a current promotion?
It can state an offer the business has already published and approved, in the exact wording. It cannot apply it to the customer’s situation or calculate what they would pay.
What if a customer is angry?
The desk drafts a calm acknowledgement and flags the conversation for a person. It does not argue, apologize for things it cannot verify, or offer compensation.
Does this make the AI less useful?
It makes it trustworthy. The desks handle the volume: acknowledgements, bookings, reminders, follow-ups, status. People handle the judgment. That split is the whole point of the build. See What we build.
Every one of these rules is written into every Staffhand build, and the digital audit is where we look at how your customer messages are handled today. Start with a digital audit.
