Skip to main content

AI Business & Productivity

AI Use Policy Template for a Small Business: One Page, Copyable

A one-page AI use policy for a small business, with wording to copy: approved tools, what data may go in, who reviews output, and who owns the grey cases.

Written bySaad Amjad
Updated
Read
~6 min
A stack of printed documents with a pen resting on the top page, laid out on a desk.

A small business AI policy that works fits on one page and names five things. The templates you will find elsewhere run to ten or twelve sections because they are written for companies with a compliance department. Nobody in a six-person business reads those, which means they protect nothing.

Below is the short version, with the actual wording to copy, plus the part the templates skip: how to decide what goes on the forbidden data list, which is the only genuinely hard decision in the document.

What the policy is actually for

Not compliance theatre. Three concrete jobs:

  • It makes one answer available to everyone. Somebody is about to paste a customer list into a chatbot to tidy it up. The policy is what they check in the ten seconds before they do.
  • It makes you answerable. When a client, insurer or tender asks how you handle AI, you send a document instead of improvising. That question is routine now and improvised answers sound exactly like improvised answers.
  • It stops the quiet drift. Without a written rule, a business’s actual AI practice is the sum of what each person privately decided was fine, and nobody knows what that is.

If none of those is a problem for you yet, the data rules section is still worth writing. The rest can wait until you have a second person.

The five sections

What goes in a one-page AI policy, and whyScroll sideways to see every column.
What goes in a one-page AI policy, and why
SectionWhat it saysWhy it is there
Approved toolsWhich tools, on which accounts, and that anything else needs askingA named list is enforceable. A principle is not
Data rulesWhat may never go in, and that everything else is fineThis is the section that prevents an actual breach of an actual contract
Human reviewNothing leaves the business without a named person reading itIt is what makes your accountability claim to clients true
Prohibited usesThe handful of things that are not allowed at allRemoves the argument before it happens
Owner and questionsOne person to ask, and how often this is reviewedGrey cases go somewhere instead of being guessed

The policy, to copy

Replace anything in square brackets. Keep it on one page — if it grows past that, something in it is not a rule you will enforce.

1. Purpose and scope

This policy covers how AI tools may be used for work at [Business name]. It applies to everyone who does work for us, including employees, contractors and freelancers. It covers general AI assistants, AI features built into software we already use, and anything that sends our information to a third party for processing.

That last clause is the one that earns its place. People read “AI tools” as “ChatGPT” and do not think of the AI summariser now switched on inside their note-taking app.

2. Approved tools and accounts

The AI tools approved for work are [tool], [tool] and [tool]. They must be used on the company account we provide, not on a personal account. Any other AI tool — including free ones, browser extensions and AI features inside other software — must be approved by [name] before it is used for work.

Two points worth being firm about. Personal accounts are the problem rather than the tool: the work and its history belong to the individual, and you cannot see or set the privacy options on them. And “free ones” has to be named explicitly, because free tools do not go through procurement and therefore never get noticed.

3. What may and may not go into an AI tool

Never put the following into any AI tool: customer or client records and contact details; anything covered by a confidentiality or non-disclosure agreement; payroll, HR or staff information; financial records that are not already public; passwords, keys or access credentials; and anything a client has told us to keep in a specific system.

Everything else is fine. That includes our own public content, your own drafts and notes, published material, and anything you would be comfortable seeing outside the business.

4. Checking the output

Anything produced with AI assistance is reviewed by a named person before it is sent to a customer, published, or used to make a decision. The reviewer is responsible for its accuracy, including any facts, figures, names, prices, dates or quotations in it. If you cannot verify something, remove it rather than soften it.

The last sentence is the useful one. The standard failure is not an obvious invention — it is a plausible figure that survives because hedging it felt like checking it.

5. Not allowed, who to ask, and review

AI must not be used to produce anything presented as the work of a named person who did not review it, to make a final decision about hiring, dismissal, credit or pricing for an individual, or to generate images or audio of a real person without their permission.

Questions about anything this policy does not cover go to [name]. This policy is reviewed every six months, and whenever we add a tool or a client gives us their own AI requirements. Last reviewed [date].

The data list is the hard part

Everything else in the policy writes itself. This section is where a small business has to think, and there are two traps in it.

The first is writing an allowed list instead of a forbidden one. An allowed list is unbounded — you cannot enumerate every harmless thing — so people hit something not on it every day and either ask you or guess. The forbidden list is short and finite, which means “if it is not on this list, go ahead” is a sentence you can actually stand behind.

The second is forgetting that the rule travels with the data, not with the tool. A customer’s address is confidential whether it is typed into a chatbot, pasted into an AI-powered spreadsheet function, or sitting inside the transcript that a meeting notetaker just produced. Write the list about information, not about software, and it keeps working when somebody adopts a tool you have never heard of.

If you only do one thing

Best for
Write the forbidden data list, in one sitting, naming your own actual records. It is five lines, it is the part that prevents a contractual problem, and it is useful even in a one-person business
Not for
Downloading a twelve-section template and filling it in. The length is what stops it being read, and an unread policy protects nobody

Putting it in place

  1. Write the forbidden data list first

    Five lines, naming your real records rather than categories you copied. If a line does not describe something your business actually holds, delete it.

  2. Fix the accounts before you issue the policy

    A rule saying “use the company account” fails on day one if there is no company account. Set up the workspace first — see ChatGPT Business vs Plus for whether you need one.

  3. Name the owner, out loud

    One person, and tell everyone who it is. This is the difference between a policy and a file.

  4. Walk through two real examples with the team

    Take something somebody did last week and something borderline, and talk through where each lands. Ten minutes of this beats any amount of acknowledgment-signing.

  5. Put it where work happens

    Shared drive, onboarding pack, pinned in your team chat. A policy filed somewhere nobody opens has the same effect as no policy.

  6. Diary the review

    Six months out, and treat a vendor renaming a plan or changing its free tier as a trigger to look again sooner.

Where to go next

Share this guide

XLinkedIn

Questions

Frequently asked questions

  • Does a small business actually need a written AI policy?

    If anyone other than you does work for the business, yes — and the reason is practical rather than legal. Without one, the answer to "can I put this in ChatGPT" depends on who is asked and what mood they are in, and the answer you give a client about where their data goes is a guess. One page removes both problems. If you genuinely work alone, write the data list anyway and skip the rest; that is the part that protects you.

  • How long should an AI policy be?

    One page. The templates circulating online run to ten or twelve sections because they are written for companies with a compliance function and an auditor to satisfy. A policy nobody reads is worse than a short one everybody does, and a small business needs five things in writing — the tools, the data rules, the review requirement, the prohibited uses, and one named owner.

  • What are the five things an AI policy must cover?

    Approved tools and which accounts they run on. What data may and may not go into them. That a named human reviews anything before it leaves the business. The uses that are not allowed at all. And one person to ask about anything the policy does not cover. Everything else is optional detail that can be added later if someone asks for it.

  • Should the policy name specific AI tools?

    Yes, and this is where most policies go soft. "Use approved AI tools responsibly" is not a rule anybody can follow or break. Name the tools, name the account they must be used on — a company workspace rather than a personal login — and say that anything not on the list needs asking first. A short allowed list is enforceable; a principle is not.

  • How do I decide what data staff may put into AI tools?

    Write the forbidden list first, because it is shorter and it is the one that matters. Customer and client records, anything covered by a confidentiality agreement, payroll and staff information, financial records that are not public, credentials, and anything a client has told you to keep in a particular place. Then say that everything else is fine, which is most of what people actually want to do.

  • Does an AI policy need to mention the NIST AI Risk Management Framework?

    Only if someone is going to ask. The NIST AI Risk Management Framework is a real reference that an insurer, auditor or enterprise client may recognise, and pointing at it costs you one sentence. It is not a requirement for a small business and it does not make a one-page policy any better at the thing it is for, which is helping your own people make a decision on a Tuesday afternoon.

  • Who should own the AI policy?

    One named person, and in a small business that is usually the owner. The point of naming them is not authority, it is that grey cases have somewhere to go instead of being resolved privately by whoever hit them. A policy with no owner quietly becomes a document that describes what people were doing a year ago.

  • How often should I review it?

    Twice a year is enough for most small businesses, plus any time you add a tool or a client imposes their own rules. The thing that dates a policy is not the technology, it is the tool list — tiers get renamed, features move between plans, and a product you approved on a free plan starts training on your content when the free plan changes.