How to write a customer service policy your team and AI can both follow

It happens in a lot of small businesses. The owner answers the same customer question for the fifth time this month, worded the way she always words it. In the other room an employee answers the same question a different way, because nothing ever wrote the answer down. Then someone pastes the question into an AI tool and gets a third version that matches neither of them. Customers notice, and they notice fastest at the moments that matter most, like a complaint or a refund ask.

A short written policy, built from your own questions and rules, fixes the drift. It gives both your people and your AI drafts one place to check, so whoever or whatever replies, the customer hears one voice.

What the page is for

The policy is a one page document. Its job is not to predict every question. Its job is to hold the answers that repeat, the rules for how the business sounds, and the cases that must go to a person. When a new hire reads it, they reply like the owner. When an AI tool reads it, it drafts like the owner. The page works because it is short enough that people actually open it, and specific enough that there is no guessing.

It also protects the business. AI tools change often and can produce wrong answers, so the policy is the check. A draft from AI is never sent as-is. Someone reads it against the policy, confirms the facts, and sends it. That habit is the whole difference between AI as a time saver and AI as a liability.

Who this fits

This fits owner-led businesses where the same questions arrive again and again: hours, service areas, quotes, deposits, turnaround times, returns, what happens when something goes wrong. It fits teams where more than one person answers customers, in email, text, or a web chat. It does not fit a business whose answers are regulated in ways that require a formal compliance document. In that case the page here is a starting point, and the formal policy still needs a professional review.

It also does not replace a person for the questions that need judgment. That is what the escalation section is for.

Build it in one sitting

Load it into the AI tool

Once the page exists, it goes to work twice. A person can open it and reply from it. An AI tool can use it to draft. In ChatGPT or Claude, start a conversation with a short instruction: reply to customer questions using this policy, keep to its tone rules, and say when a case belongs to escalation instead of inventing an answer. Then paste the policy below the instruction and send.

After that, each new question goes in as a normal message. The draft comes back in the policy's voice with the policy's facts. The person sending still checks three things: does the draft answer the actual question, does it match a written answer on the page, and is it a case the policy says to escalate. If any check fails, the draft does not go out.

For questions the policy does not cover, the honest answer is a saved draft, not a send. Write the new answer once, decide whether it belongs on the page, and add it if it will repeat. The policy grows out of real questions, not out of a template nobody wrote.

Keep it current

Set a short review on the calendar every few months. In the review, drop answers that no longer apply, add the new repeat questions, and check the tone rules still sound like the business. Also review after anything changes: a price, a service area, a return window, or a rule that came out of a hard customer moment. A policy that is out of date is worse than no policy, because everyone trusts it.

A business that wants help with this kind of work can bring in a person. AZ AI Lab gives Arizona business owners an AI Concierge, one of the team who learns the business, works on real tasks on two 1-hour calls a month, and answers voice messages and email between calls. The first step is a free 25-minute fit call on Zoom that looks at two tasks that repeat in the owner's week and picks a first result worth having.