The short answer.
Yes. A small company needs a written AI policy, and two pages is enough: which assistant is approved, what information may and may not go into it, that a person checks and decides before anything is sent, paid or changed, and who to ask. Do not ban AI while you write it, because your people will use it anyway on personal accounts where you can see nothing. Write the two pages this week and let people work.
A policy says what not to do. A plan says what to do. A company needs both, and the policy is the quicker of the two to write.
Do small businesses need a written AI policy?
A small business needs a written AI policy because its people are already using AI, and without one each person sets their own rules. One person pastes a customer list into a free account. Another refuses to touch the assistant the company pays for, because nobody said it was allowed. Both are reasonable guesses, and both cost you.
A written AI policy replaces guesses with one answer everyone can read. It also gives a new hire something to read on day one. Spoken rules drift. Two written pages do not.
What goes in an AI acceptable use policy.
An AI acceptable use policy for a small company answers five questions, and nothing more.
- Which assistant.The one your company pays for and administers, by name, on the company account.
- What may go in.The kinds of company information people may put into that assistant to do their work.
- What must not go in.The kinds of information that stay out, stated as categories people can recognize.
- Who decides.A person checks every output and decides before anything is sent, paid or changed.
- Who to ask.One person, by name, who answers questions and owns the policy.
Everything else, such as model choices, vendor reviews and risk registers, belongs to larger organizations or to a later version.
Do not ban AI while you write the policy.
A company should not ban AI tools while it writes its AI policy. A ban does not stop the work. It moves the work onto personal phones and personal accounts, where the company cannot see what goes in, cannot set the account's settings and cannot get anything back when someone leaves.
An interim rule works better than a ban. Send one message today: use [name of assistant] on your company account, keep customer and employee personal details out of it until the policy is out, and check everything before it leaves the building. Then write the two pages and send them within the week.
Which AI tools your people may use.
Your people may use the assistant your company pays for, on the company account, and that is the whole list to start. For most small companies that is one of ChatGPT, Claude, Copilot or Gemini, chosen because it is already bought or already sits inside the office software you run.
One approved assistant on one company account gives you three things a personal account does not: an administrator who controls the settings, work that stays with the company when a person leaves, and one place to point every future workflow. Add a second tool only when a named piece of work needs it, and add it to the policy by name. What each assistant does with your data is covered in Is it safe to put company data into ChatGPT, Claude, Copilot or Gemini?
A two-page AI policy to copy.
Copy the eight headings below, keep the sample wording or replace it, and fill in the blanks. The whole policy fits on two pages.
This template is a starting point and not legal advice. A company in a regulated field, or with duties of confidentiality to its clients, should have its own adviser read the policy before it goes out.
- Purpose.This policy says how people at [company name] use AI assistants in their work. We want everyone to use AI to do better work, safely.
- The assistant we use.Our approved assistant is [name of assistant], on the company account. Do not use personal accounts or other AI tools for company work unless they are added to this policy.
- Information that may go in.You may put in the company information you need for the task, such as drafts, internal documents, procedures and [other categories]. Use only what the task needs.
- Information that must not go in.Do not put in [categories, for example: passwords and access codes, payment card or bank details, health information, employee records, anything a client has asked us to keep confidential]. When in doubt, leave it out and ask.
- A person checks and decides.The assistant prepares; a person decides. Read and check every output before anything is sent, paid, signed, published or changed, and the person who sends it is responsible for it.
- Work for clients or customers.Follow any agreement we have with a client about how their information is used. If a client asks how we use AI, refer them to [name].
- Who owns this policy and who to ask.[Name, role] owns this policy. Ask [name] before you use the assistant for something new or if you think something went into it that should not have.
- When we review it.We review this policy every [six months] and whenever we add a tool. Last reviewed: [date].
One company writes its policy.
Illustrative example.
Take a 60-person distributor. The company pays for Copilot because it already sits in the office software the company runs, and the sales and accounts teams use it daily. Two people in the warehouse use a free assistant on their phones. Nobody has written anything down.
The operations director owns the policy. On Monday she sends the interim message: use Copilot on your company account, keep customer and employee personal details out of it, and check everything before it goes out. On Wednesday she fills in the eight headings. The list of what must not go in names customer bank details, staff records and supplier pricing agreed in confidence. She names herself as the person to ask and sets the review for six months out. The policy goes to every employee on Friday, and the two warehouse staff move to company accounts the following week.
The next question in her inbox is the useful one: which work should Copilot be doing for each team. That question belongs to the plan.
How much governance a fifty-person company needs.
A fifty-person company needs one owner, one written policy, one approved assistant and one review date. It does not need an AI committee, a steering group or a separate governance program. Most published advice on AI governance is written for large enterprises, and it does not scale down.
- The owner is an operator.The person who owns the policy runs operations or the business. IT helps with accounts and settings; the decisions about how work is done stay with the operator.
- Approval is one conversation.A new tool or a new use is a question to one named person, answered that week.
- The policy is read, not filed.Two pages get read by every employee. Twenty pages do not.
- Each workflow has its own owner.As workflows arrive, each has one named person who keeps it current. Who owns an AI workflow after the pilot? sets out the roles.
When customers ask harder questions about data, Is our data safe? covers what the AI sees, where it runs and who decides, and how we work in your systems sets out our own approach.
The policy and the plan.
The AI policy says what not to do. The plan says what to do: which work each team hands to the assistant, who owns each workflow and what gets built first. How organizations plan their use of AI walks through the plan in three steps: plan it, develop the workflows, deploy them.
FullyPowered Planner is where the plan lives. You set out your teams, their roles and the work between them, choose what to build first, and invite colleagues to add their own teams. The workflows then run in your own ChatGPT, Claude, Copilot or Gemini, on your company account, under the policy you just wrote.