How to Write an AI Use Policy for a Small Business
Ask a team of eight people whether they use AI tools at work and you will usually get two or three hands.

Ask a team of eight people whether they use AI tools at work and you will usually get two or three hands. Ask what they pasted into a chatbot last week and the real number appears: the bookkeeper summarising supplier emails, the sales lead rewriting proposals, the intern drafting replies to customer complaints, complete with names and order numbers. None of them did anything malicious. Nobody ever told them where the lines were.
That is what an AI use policy for a small business is for. Not a twenty page legal document, but one or two pages that tell people which tools they can use, what they must never paste in, and who checks the output before it reaches a customer. You can write a good one in an afternoon.
Start With What People Already Do
Before writing a single rule, find out how the tools are being used today. Send a short, blame free question to everyone: which AI tools have you tried for work, and for what tasks? Promise in writing that nobody gets in trouble for the answer.
You will almost always find three patterns. Someone uses a free chatbot on a personal account. Someone has installed a browser extension that reads every page they visit. And someone has connected an AI tool to the shared mailbox or calendar without telling anyone. A policy written without knowing this ends up banning things nobody does and missing the things everybody does.
Sort Your Information Into Three Simple Tiers
The core of any AI use policy is a plain answer to "can I paste this in?" The easiest way to give that answer is to sort information into tiers that a new hire can apply without asking.
| Tier | Examples | Rule |
|---|---|---|
| Public | Published prices, website copy, public job ads | Any approved tool |
| Internal | Draft proposals, meeting notes, process documents | Approved business accounts only |
| Restricted | Customer names and contact details, payroll, health data, passwords, contracts | Never, unless a named tool is approved for that exact use |
Keep the restricted list concrete. "Confidential information" means nothing at 4 p.m. on a busy Friday. "Anything with a customer's name, email, phone number or order history" is a rule people can actually follow.
Name the Approved Tools and the Account Type
A policy that says "use AI responsibly" will be ignored. A policy that says "use the business account of these two tools, signed in with your work email" will be followed, because it is easy.
The account type matters more than the brand. Many free consumer plans allow the provider to use your conversations to improve their models, while paid business plans usually switch that off by default and give you an admin view of who is using what. For a team of ten, a business plan often costs 20 to 30 dollars per person per month. Weigh that against one customer list leaking.
List each approved tool with its purpose in a single line, for example "Meeting transcription: tool X, internal meetings only, delete recordings after 30 days." If you are still choosing, the guide on comparing AI meeting transcription tools walks through what to check before paying.
Decide Who Reviews Output Before Customers See It
The biggest risk for most small businesses is not data leaking out. It is wrong information going out with your logo on it: an invented refund policy in a support reply, a made up statistic in a sales deck, a confident but incorrect answer about a product.
Write one rule that covers it: anything AI drafted that goes to a customer, supplier or the public must be read in full by a person who could have written it themselves. Not skimmed. Read. The reasoning behind this is covered well in why the human in the loop still matters, and it applies doubly when one wrong email can cost you a client.
Where review can be lighter
Internal summaries, first drafts nobody outside sees, and brainstorming lists can skip formal review. Say so explicitly, or people will either review everything and resent the policy, or review nothing because the rule felt unrealistic.
Why a Total Ban Usually Backfires
Some owners react to the first scare by banning AI tools completely. It feels safe. In practice it pushes use onto personal phones and personal accounts, where you have no visibility at all and the free plan's data terms apply.
Surveys of office workers keep finding that a large share use AI tools their employer has not approved, and bans do not change that much. A small business is better off approving one or two tools, paying for proper accounts, and drawing clear lines around restricted data. You get the productivity and keep the control.
The exception is regulated work. If you handle health records, legal case files or client money, check your sector's rules before approving anything, and keep restricted data out entirely until you have.
Cover the Awkward Cases Before They Happen
A short section of plain answers saves a lot of questions later. Include at least these:
- Can I use AI to write a job advert or screen CVs? Drafting the advert, yes. Ranking or rejecting candidates, no, a person decides.
- Do we tell customers when a reply was drafted with AI? Decide once and write it down. Many businesses disclose it on chatbots but not on human reviewed emails.
- Who owns prompts and outputs? The business, the same as any other work product.
- What if I pasted something restricted by mistake? Tell your manager the same day, with no penalty for reporting. Hidden mistakes are the expensive ones.
Keep It Short and Review It Every Six Months
Aim for two pages at most. Put the three tiers and the approved tool list on page one, where people will actually see them. Ask each person to read it and reply to confirm, which takes ten seconds and gives you a record.
Then put a date in the calendar six months out. Tools change their plans and data terms often, new tools appear, and your team will have found uses you never imagined. A quick review keeps the policy matched to reality instead of slowly turning into a document nobody opens.
Frequently Asked Questions
Does a small business legally need an AI use policy?
In most places there is no law that requires a standalone document. Data protection rules still apply to whatever staff paste into AI tools, though, and a written policy is the simplest way to show you took reasonable care.
Should the policy name specific AI products?
Yes. Naming approved tools and account types is what makes the policy usable. Update the list at each review rather than writing vague rules that try to cover every possible product.
How do I get staff to actually follow it?
Make the approved route the easy route. Pay for the business accounts, set them up for people, and keep the rules short enough to remember. Policies fail when following them takes more effort than ignoring them.
Your First Draft This Week
Send the usage question on Monday, sort your data into three tiers on Wednesday, and circulate a two page draft by Friday. It will not be perfect, and that is fine; a short policy people follow beats a thorough one nobody reads. For more practical ideas on using bots and automation in a growing company, see the business section.
More in Bots for Business
Bots for Business
Quote Follow Up Automation for a Small Trade Business
A joiner I have swapped notes with for two years sends about fourteen quotes a month. He wins roughly three.
Bots for Business
Client Onboarding Automation for a Small Service Business
A two person branding studio I traded notes with last year measured something uncomfortable.
Bots for Business
Abandoned Cart Email Automation for a Small Online Shop
Roughly seven out of ten online shopping carts are abandoned before payment, and for a small shop that is real money walking...