AI Policy for Small Business: Write One Your Team Will Actually Read

TL;DR: A good AI policy for small business fits on one page people can scan, with a fuller version behind it. It names the approved tools, lists what never goes in, says which work a person checks before it leaves the building, and names who to ask. Write every rule as "you do this", give each one an example, test it on two colleagues, and set a review date.
Your team is probably using AI already. Some of it is on tools you approved and some of it is on personal accounts you've never heard of, which is what people call shadow AI. An AI policy for small business turns that into something you can manage. Templates are easy to find. The hard part is writing one your team reads, remembers and follows on a busy Tuesday.
This article covers what the policy should include, how to word each rule so it gets followed, and how to keep it current. If you want a step-by-step template, use the companion AI acceptable use policy playbook. This isn't legal advice. Rules on AI, data protection and employment differ by country and industry, so have a qualified professional check your final policy against the law where you operate.
Why many AI policies don't get read
Many policies fail for ordinary reasons. They're long, they're written in the passive voice so nobody can tell who's responsible, and they say "use AI responsibly" without saying what that means for someone writing a client email at 4 p.m. Bans with no approved alternative send people back to personal accounts.
A policy that works answers the question someone actually has in the moment: "Can I paste this into the assistant?" If they can find that answer in ten seconds, they'll use the policy. If they can't, they'll guess.
What an AI policy for small business should cover
A useful policy covers seven things. Add or cut sections to fit how your team works.
1. Approved tools and accounts
Name the exact tools and accounts. A company account can give you different data terms from a personal one, depending on each vendor's business terms, so check them before you approve a tool. Then make the approved tool easy to get, or people will use something else while they wait.
2. What never goes in
Start with this rule. NIST's profile for generative AI lists data privacy as one of the risks the technology creates or makes worse, describing it as impacts from "leakage and unauthorized use, disclosure, or de-anonymization of biometric, health, location, or other personally identifiable information or sensitive data" (NIST AI 600-1, Generative AI Profile).
For many small teams the never list is customer and employee personal data, passwords and access keys, unreleased financials, and anything under a confidentiality agreement. Give people a way to keep working: show them how to swap names and numbers for placeholders before they paste, which our free PII redactor does in the browser.
3. What a person checks before it's used
The same NIST profile lists confabulation, "the production of confidently stated but erroneous or false content (known colloquially as 'hallucinations' or 'fabrications')" (NIST AI 600-1). That's why the policy needs a human in the loop for anything sent to a customer, published, used in a number someone relies on, or used in a decision about a person.
4. Disclosure
Say when your team tells customers or colleagues that AI helped. Check your client contracts first, since some already set a rule.
5. Who to ask, and who owns the policy
Name the one person who answers questions about tools and data. New tools appear often, and "ask Sam before you sign up for anything" handles much of what the policy can't predict.
6. What happens when something goes wrong
Tell people what to do if they pasted something they shouldn't have: who to tell and how fast. Make it safe to report, so you hear about mistakes the same day.
7. Acknowledgment and review date
A signature line, and a date when you'll look at the policy again.
How these sections line up with NIST
A ten-person business doesn't need a governance framework, but it helps to know your policy lines up with one. The US National Institute of Standards and Technology publishes the AI Risk Management Framework (version 1.0, currently under revision), which "is intended for voluntary use" (NIST AI RMF). Several of its GOVERN subcategories line up with the sections above (NIST AI 100-1):
- GOVERN 1.5 asks that "ongoing monitoring and periodic review" be planned, "including determining the frequency of periodic review". That's the idea behind your review date.
- GOVERN 1.6 reads "Mechanisms are in place to inventory AI systems". That's your list of approved tools, kept current.
- GOVERN 2.1 asks that roles and responsibilities be "clear to individuals and teams throughout the organization". That's your named owner and your "who to ask".
- GOVERN 2.2 asks that personnel "receive AI risk management training". That's the walk-through when you roll the policy out.
- GOVERN 6.1 covers risks from third parties, including "infringement of a third-party's intellectual property or other rights". That's your vendor terms check.
If your business is in the European Union, or what your team produces with AI is used there, also look at the EU AI Act's AI literacy rule, Article 4. The European Commission says it "entered into application on 2 February 2025", and its questions and answers page gives an example close to everyday office work: employees who use a chat assistant to write advertising copy or translate text "should be informed about the specific risks, for example hallucination". The same page says "there is no need for a certificate" and that organisations "can keep an internal record of trainings" (European Commission, AI literacy questions and answers). In 2026 the Digital Omnibus on AI rewrote Article 4. Businesses that use AI systems must still "take measures to support the development of AI literacy" of their staff, and the amended text says this "does not require providers or deployers to guarantee any specific level of AI literacy of any individual" (Regulation (EU) 2026/1744). Ask a qualified professional what that means for your business.
How to write rules people follow
Write every rule as "you do this"
The US government's plain language guide puts the case for active voice in two sentences: "Active voice makes it clear who should do what. It eliminates ambiguity about responsibilities" (Digital.gov, Writing for understanding). Compare "Customer data should not be entered into unapproved AI systems" with "Don't paste customer names, emails or order numbers into any AI tool except [approved tool]." The second says exactly what to do.
Give every rule an example
A rule without an example gets read in whatever way is convenient. "Don't paste confidential information" means different things to different people. "Don't paste the supplier price list, even to summarize it" doesn't.
Put the answers people need on one page
Keep the full policy for reference, with a one-page card on top. The card has three short lists: things you can do with the approved tool, things you check with a person first, and things you never do. Many people will only ever read the card, so make sure it's correct on its own.
Sort real tasks into the three lists
Ask your team for the ten things they actually use AI for, and place each one on the card. The task inventory playbook walks through scoring tasks by risk and by how easy the output is to check.
A prompt to test your draft before you share it
Use your approved assistant to check the draft for gaps and write the card. Remove anything confidential first.
Below is a draft AI policy for a team of [number] at a [type of business].
Do three things:
1. Write a one-page summary card with three lists: "You can",
"Check with [owner name] first" and "Never". Use only rules that
are in the draft. Write each line as an instruction to "you",
in active voice, with one concrete example.
2. List every rule in the draft that is vague, has no example, or
doesn't say who is responsible. Quote the rule and say what's missing.
3. Write five realistic situations an employee might face this week
(for example, summarizing a customer complaint) and say which list
each one falls under. If the draft doesn't answer a situation,
write NOT COVERED instead of guessing.
Do not add rules that are not in the draft.
Draft policy:
[paste your draft]
Why this works: The first task produces the card from your own rules, and the "do not add rules" line stops the assistant inventing policy you never decided. The second task finds the vague lines that cause misreadings. The third task is a quick test: any situation marked NOT COVERED is a decision you still need to make. Read every line of the output yourself before anyone sees it.
Test it on two people
Before you share the policy, give the card to two people who didn't help write it, along with three or four situations from the prompt, and ask which list each one belongs on. Where they hesitate or disagree, the wording needs work.
Roll it out and keep it current
- Walk through it in a meeting. Ten minutes on the card, then questions, which show you what's missing.
- Collect acknowledgments. A signature or an email reply is enough for many small teams.
- Keep the tool list current. When someone asks to use a new tool, decide, then update the policy the same week.
- Set a review date. The playbook suggests six months as a starting point. An incident or a new tool is a reason to review sooner.
- Use mistakes to improve it. When someone reports a slip, fix the gap and thank them.
If you're new to AI yourself, start with how to use AI at work in your first week.
Frequently asked questions
What should an AI policy include?
At minimum: the approved tools and accounts, the data that must never go in, which outputs a person checks before use, when to disclose AI use, who to ask, what to do after a mistake, an acknowledgment line and a review date. The AI acceptable use policy glossary entry has a short outline you can adapt.
Does a small business need an AI policy?
If anyone on your team uses AI tools for work, a short policy is worth having, because it tells people where your data may go and what needs checking. Whether the law requires one depends on where you operate and what you do, so check with a qualified professional.
How long should an AI use policy be?
Long enough to answer real questions and short enough to scan. A one-page card on top of a one-to-two-page policy works for many small teams.
How often should you update an AI policy?
Pick a fixed review date when you publish it. Six months is a sensible starting point because tools and vendor terms change often. Review it sooner after an incident, a new tool or a change in the law where you operate.
Who should write a company AI policy?
The owner or manager who can make decisions about tools and data should make the calls, with input from the people who use AI every day. An assistant can help with the wording, and a qualified professional should review the final version before you roll it out.
Terms in this article
From the library
Rather have someone set it up for you?
That's a fair call. LTFI, the agentic business operating system from Kief Studio, can set up the work these guides describe and run it with you. No website yet? Kief Studio will build you one for free.

