AI Use Policy for Your Shop: The One-Page Rulebook

Picture this: your best tech is on a roof, unsure about a wiring question, and asks a chatbot on his phone. The answer sounds confident. He wires it that way. Now picture the same tech pasting a customer's name, address, and credit card details into that same chatbot to "draft the invoice." One of those is smart tool use. The other is a privacy disaster you own. Both are happening right now in shops without an AI policy — and the 8-system framework only works when the tools inside it have guardrails.

Why you need this now, not later

AI tools are already inside your business. The dispatcher uses one to rewrite a nasty customer email. A tech uses one to translate a Spanish text from a customer. Someone in the office uses one to summarize a 40-message text thread before you read it. None of this is bad. All of it is happening without rules.

The risk isn't that AI makes things up — though it does. The risk is that without a policy, your team invents their own rules, and the first time you learn what those rules are is when something goes wrong: customer data in a tool's training data, an AI-written estimate that promises something you can't deliver, or a tech following generated instructions on a safety-critical repair.

A one-page policy fixes this in an afternoon. It doesn't ban AI — banning it just drives the use underground, where you can't see it. It sets the lanes: what's encouraged, what's fine, what needs approval, and what's never okay.

The three lanes: Green, Yellow, Red

The simplest policy structure that actually gets followed: three lanes. Every AI use falls into one. Print it on one page, post it in the office, go over it in a team huddle.

Green lane: use it freely

These uses are encouraged. No approval needed. If your team isn't doing these yet, start here:

• Rewriting customer messages. Turning a rushed, blunt tech text into a professional update before it goes to the customer. No customer data beyond first name and job details.
• Summarizing threads. A 50-message group text about a job becomes a 5-bullet recap for the office.
• Drafting SOPs (standard operating procedures). Talking through a process and getting a first draft to clean up — dictating beats typing every time.
• Translating customer communications. Bilingual markets run on this; a tech reading a Spanish text and replying in English is normal daily work.
• Brainstorming marketing ideas. Subject lines, post ideas, seasonal campaign angles — low stakes, high leverage.
• Math and estimates. Double-checking material quantities, converting units, comparing equipment specs.

The rule of the green lane: no real customer personal data goes in. First names, addresses, job details are fine. Anything on the red list (below) is not.

Yellow lane: use it, but verify before acting

These uses are fine, but the output must be checked by a human before it leaves your hands:

• Technical how-to answers. Wiring diagrams, torque specs, refrigerant pressures, code questions. AI tools confidently state wrong numbers. The tech's license and judgment are the final authority, always.
• Quote and estimate language. Great for drafting scope descriptions, but every number, timeline, and warranty term gets owner or manager review before it reaches a customer.
• Responses to reviews or complaints. Drafting is fine; posting requires approval. A wrong tone on a public 1-star review is a headline you don't want.
• Contract or legal-ish language. Payment terms, liability waivers, lien language. General drafting help only — anything that binds the business goes through you, and real legal questions go to your attorney.

The yellow-lane mantra to teach your team: "The AI drafts, a human decides." AI is an intern with a photographic memory of the internet. Smart, fast, and wrong often enough that you never let it sign its own work.

Red lane: never do this

Hard lines. No exceptions, no "just this once":

• Never paste customer personal data into a free AI tool. Full names with addresses, phone numbers, credit card numbers, driver's license info, gate codes — any of it. This is the big one. Free tools can train on inputs; you cannot un-ring that bell.
• Never upload a full customer list, contract, or financial document to a tool you haven't vetted. QuickBooks data, employee payroll info, signed contracts — treat them like cash.
• Never send an AI-generated message to a customer without a human reading it first. No auto-replies written by AI that nobody reviewed, no AI-written estimates that went straight out.
• Never present AI output as your own verified work on safety, code, or warranty matters. If the tech isn't sure, he calls the senior tech — he doesn't screenshot a chatbot answer and keep wiring.
• Never use AI to write fake reviews — for your business or to slam a competitor. That's fraud, it violates Google's policy, and it will surface.

The copy-paste one-page policy

Here's the actual document. Print it, have everyone sign it, keep it in the office binder next to the handbook:

Roll it out in one week

Day 1: You read this post and customize the policy with your business name. Day 2: Walk through it in a 15-minute team huddle — read the red lane aloud, because that's the part that matters most. Day 3: Print copies, have everyone sign, file them. Day 4: Add it to your new-hire packet so the next employee sees it on day one. That's it.

Revisit the policy once a year or when a tool changes — the AI landscape moves fast, and a policy from 2026 will need updating by 2027. But the lanes don't change: encourage the useful, verify the important, and never let customer data leak into the machine. Pair it with a one-day training system and your newest hire learns the lanes before they learn the dispatch board.

Related guides in this series

Your systems are only as strong as the people running them. Read the 8-system framework overview to see where AI guardrails fit, then train new hires in one day so the policy sticks from day one. And if a system ever stops earning its keep, use AI to draft review responses the right way — with a human in the loop on every send.

The done-for-you version

This post is part of the Leveraged Owner 8-system framework — the done-for-you Starter Pack with the full step-by-step SOPs, every script, the worksheets, and screen-by-screen setup instructions for all 8 systems.

Stats sourced as labeled: review trust data from BrightLocal 2026 consumer survey.

Keep building your systems

Related guides from the Leveraged Owner blog: