
Somebody on your team used ChatGPT today. Maybe they wrote a customer email with it. Maybe they pasted last month's sales numbers in and asked for a summary. Maybe they uploaded a client contract and asked it to "make this shorter." You probably don't know which one. That's the problem.
Zapier surveyed more than 500 small and midsize business owners this year and found that 76% have adopted some form of automation, and 30% are using AI that makes decisions across multiple systems with limited human input. That's not dabbling. That's AI running real parts of real businesses, usually without anyone writing down the rules.
You don't need a 20-page compliance document to fix this. You need an AI policy for your small business that fits on one page, answers four questions, and that your team will actually read. You can write it this week.
I hear "we don't really use AI here" a lot. Usually from the owner. Rarely from the staff.
AI tools are free, they live on everyone's phone, and nobody needs a purchase order to sign up. So your office manager is drafting emails with one, your marketing person is writing captions with another, and your bookkeeper is quietly asking a chatbot to explain a spreadsheet formula. None of that is bad. Some of it is great. But none of it is happening under rules you chose.
And no, the answer isn't banning it. A ban doesn't stop anyone from using AI. It just stops them from telling you about it. Now you've got the same risk with less visibility.
The point of a policy isn't to slow anybody down. It's so the person using AI at 4:45 on a Friday knows what's off-limits without having to guess.
Most AI policy templates online are written by compliance vendors and law firms for companies with an IT department and a legal team. They have sections on ethical principles, bias mitigation, and governance committees. Fine if you have 5,000 employees. Overkill if you have 12.
Call it an AI acceptable use policy, employee AI use guidelines, or just "the AI rules." The name doesn't matter. For a small business, your policy has to answer four questions:
Which tools are we allowed to use?
What never goes into them?
When does a person have to check the output?
Who do I ask when I'm not sure?
If your policy answers those clearly, you've covered the problems that actually happen in small businesses. Everything else is nice to have.
Name the tools. Not "approved AI platforms." Actual names. "Our ChatGPT Team account. Claude. The AI features in Canva. Copilot inside our Microsoft 365."
Account type matters as much as the tool. Business and team plans usually give you more control over whether your data gets used to train the model, and personal free accounts usually give you less. Check the privacy settings on each one before you approve it, and tell your team to use work accounts for work.
Then pick one person who decides. Not a committee. In most small businesses that's the owner or the operations manager. If someone wants to try a new tool, they send that person three things: the tool name, what they'd use it for, and what information it would touch. They get a yes or no within a week. That's what AI governance for small business actually looks like: one person, one simple request, one answer.
Don't forget the AI you already pay for. Your email platform, your CRM, and your design software probably all added AI features this year. And meeting note-takers deserve a specific mention, because a bot that records and summarizes client calls is handling some of the most sensitive conversations you have. Decide whether it's allowed, and whether clients get told it's there.
This is the section that does the most work. The rule is simple: if it would be a problem showing up in a stranger's inbox, it doesn't go into an AI tool.
Spell it out, because "be careful with sensitive data" means something different to everyone. The never list should include:
Customer names paired with contact info, account numbers, payment details, or health information
Employee pay, performance reviews, and personal records
Passwords, logins, and API keys
Financials that aren't public: bank statements, payroll, client-specific pricing
Anything under an NDA or marked confidential in a contract
Then give people a workaround, or they'll ignore the rule. Most of the time you can still get help by stripping out the identifying details. "Write a polite reply to a customer who's upset about a late delivery" works just as well without the customer's name, address, and order number.
One honest caveat. If you're in healthcare, financial services, legal, or you handle a lot of personal data, you've got industry rules on top of this (HIPAA, for example). That's the point where you do call the lawyer. Bring them your one-page draft. It'll be a shorter, cheaper conversation.
AI is confident. It is not always right. It will make up a statistic, invent a policy you don't have, or quote a price you've never charged, and it'll say it like it means it.
So your policy needs a short list of things that always get human eyes before they go anywhere:
Anything going to a customer, vendor, or partner
Anything posted publicly
Anything with numbers, dates, prices, or legal language
Anything used in a hiring, pay, or discipline decision
Any automation that sends messages or updates records on its own
That last one matters more every month. Remember the 30% of businesses running AI with limited human input? Those workflows need someone to sign off before they go live, and someone to spot-check them after.
Add one line that settles every argument: whoever hits send owns it. "The AI wrote it" isn't an excuse with customers, so it shouldn't be one internally either.
Now the counterpoint. Don't require review on everything. Brainstorming, rough first drafts for yourself, summarizing your own notes: let people do that freely. If your policy requires sign-off on every prompt, people will just stop telling you they use it.
Name a person and a way to reach them. "Ask Kim in the AI channel." "Text me." Whatever your team actually uses.
Then say, in writing, that nobody gets in trouble for asking. That sounds soft. It isn't. If asking feels risky, people guess. And guessing is how a customer list ends up pasted into a free chatbot.
Copy this, fill in the brackets, and delete anything that doesn't apply to you.
[COMPANY NAME] AI USE POLICY
Last updated: [date] Policy owner: [name]
WHY WE HAVE THIS
We use AI to save time and do better work. These rules protect our
customers, our team, and the business.
1. APPROVED TOOLS
You may use these for work: [tool + account type, e.g. ChatGPT Team]
AI features built into [Microsoft 365 / Google Workspace / Canva / CRM]
are approved.
Use work accounts for work. Not personal ones.
Want to try something else? Send [name] the tool, what you'll use it
for, and what information it would touch. You'll hear back within
[5 business days].
2. NEVER PUT THIS INTO ANY AI TOOL
- Customer names with contact info, account numbers, payment details,
or health information
- Employee pay, reviews, or personal records
- Passwords, logins, or API keys
- Financials that aren't public
- Anything under an NDA or marked confidential
Need AI help with something that includes this? Remove the identifying
details first, or ask [name].
3. A PERSON CHECKS IT BEFORE IT GOES OUT
Review AI output before it:
- goes to a customer, vendor, or partner
- is posted publicly
- includes numbers, dates, prices, or legal language
- is used in a hiring, pay, or discipline decision
Automations that send messages or change records on their own need
[name]'s approval before they go live.
Whoever sends it owns it.
4. WHEN YOU'RE NOT SURE
Ask [name] by [Slack / text / email]. Asking never gets you in
trouble. Guessing might.
5. IF SOMETHING GOES WRONG
Put something in an AI tool you shouldn't have? Tell [name] the same
day. We'd rather fix it in an hour than find out in a month.
We review this policy every [6 months].
I've read and understand this policy.
Name: ____________ Signature: ____________ Date: ________
That's it. One page. If yours is turning into three, you're probably writing rules for problems you don't have.
Don't email the policy with "please read." Nobody will.
Start by asking your team what they already use. A quick anonymous survey works: which AI tools, for which tasks, on which accounts. You'll learn more from that than from any template, and it tells you which tools to approve first.
Fill in the template with what you learn. That takes about an hour.
Then hold one 20-minute meeting. Walk through the page and show two real examples from your own business: one use that's totally fine, and one that would break the rules. Concrete examples stick. Abstract principles don't.
Have everyone sign it, and add it to onboarding so new hires get it on day one. Then put a reminder on your calendar to revisit it every six months, because the tools are changing faster than your policy will.
You don't need a lawyer to write the first version of your AI policy. You might want one to review it, especially in a regulated industry. But a one-page policy your team actually follows beats a perfect one sitting in a shared drive nobody opens.
Want help picking the right AI tools and setting up the rules around them? [Let's Talk]
Connect With Me