KGetsIt

Resources · Getting started

AI policy generator: ten questions that build your company's AI policy

Updated 2026-10-27

An AI policy generator takes a handful of answers about your business and turns them into a policy document. You do not need a paid tool or a signup to get that result. The ten questions below are the ones a good generator asks, and each answer maps to a clause you copy: which tools are approved and on which accounts, what data never goes in, who reviews output, when customers are told, what happens when AI acts on its own, who owns the policy and the work, and when it gets reviewed. Twenty minutes and you have one page that matches how your business actually works, which is the part generic generators get wrong.

What a generator does, and what it misses

An AI policy generator is a form. You type in a company name, pick an industry, tick a few boxes, and it hands you a document. The good ones ask real questions and keep the output short. The weak ones produce five pages of corporate language with your name pasted in, and the free ones are often a doorway to a compliance product you did not come for.

The document is never the hard part. The hard part is the answers, because a policy is a set of operating decisions: which tools, whose accounts, what data, who checks the work. A generator cannot know that your bookkeeper already uses a free chatbot on their phone, or that your intake form collects health details, or that you are in Massachusetts and already owe the state a written data security program. You know those things. So the useful version of a generator is a list of questions that pulls those answers out of your head and attaches the right clause to each one.

That is what this page is. Answer the ten questions, copy the clause that matches each answer, and assemble them in order. If you want the plain fill-in-the-blank version first, the small business AI policy template is the base document. This page is the question-driven way to fill it.

The ten questions

Each question has a short answer set and the clause it produces. Keep your clauses to a sentence or two. Write what is true today, not what you hope will be true next year.

1. Which AI tools does your team use right now?

Include the ones nobody approved. Ask around before you answer, because the honest list is usually longer than the owner thinks.

Clause: “We use these AI tools for company work: [tool, plan, account owner]. To use anything else, ask [name] first.”

Two or three tools is enough to start. If the list is already long, the policy is your chance to shrink it to the ones you actually want to support.

2. Are those tools on business accounts or personal ones?

Mostly personal: your first job is moving them. On the major tools, the paid business tiers keep your conversations out of model training by default, and the free personal tiers often do not. We explain the difference in is your business data safe with ChatGPT, and if the tool is ChatGPT the plan question is answered in ChatGPT Team vs Enterprise.

Clause: “Company work happens on company accounts only. Personal AI accounts are not approved for company work, even for drafting.”

3. What is the most sensitive data your business handles?

This answer decides the strictest line in the policy. Everyone takes the general clause. Then pick the one that applies to you and add it.

Everyone: “The following never goes into an AI tool, approved or not: customer names tied to sensitive details, payment or bank data, government ID numbers, passwords and keys, employee records and payroll, anything under NDA. When in doubt, describe the situation without the identifying details.”

Health information (medical, dental, therapy, anyone with patient records), add: “No patient information of any kind goes into an AI tool unless that tool is covered by a signed business associate agreement and [name] has approved it for that use.” Patient data has its own rules, and we cover the practical setup in AI for medical and dental offices.

Client confidential information (law, accounting, advisory), add: “Client matters are not entered into any AI tool that trains on inputs or shares them with third parties. Approved tools for client work are listed separately, with the account and the data terms recorded.” For a law practice the confidentiality and privilege questions deserve their own page, and we wrote it: private LLM for law firms.

Financial account data (bookkeeping, lending, payments), add: “Account numbers, balances, and transaction detail stay inside the systems built to hold them. AI tools get summaries without identifiers.”

4. Where is the business located?

Massachusetts: the state data security regulation, 201 CMR 17.00, already requires any business holding a Massachusetts resident’s personal information to keep a written information security program. Your AI policy should be a page inside that program, not a separate document that contradicts it. We walk through the regulation in IT consulting in Salem, MA.

Clause: “This policy is part of our written information security program under 201 CMR 17.00. Any AI tool that could receive personal information is reviewed under that program before approval.”

Anywhere else: check whether your state has a data security or AI-specific rule and name it. If you do not know yet, write the clause without the citation and add it at the first review.

5. Does AI talk to customers without a person in between?

No, a person reviews everything first: skip to question six. Yes, a bot answers website chat, automated replies go out, or an assistant handles intake: you need a disclosure line.

Clause: “When AI interacts with customers on its own, we say so plainly, and we always offer a path to a person. We do not describe AI as a human and we do not claim abilities our tools do not have.”

Consumer protection law does not have an AI exception. Overstating what an automated system does is the same misleading claim it would be if a person made it.

6. Who reads AI output before it leaves the building?

Everyone sends their own work: “AI output is a draft. A person reads and approves anything that goes to a customer, a vendor, or the public. Whoever sends it, owns it.”

Licensed professionals sign off (clinician, CPA, attorney, engineer), add: “Anything that carries a professional’s name or license is reviewed by that professional before it goes out, with no exception for routine items.”

7. Does any AI act on its own: sending, booking, paying, changing records?

Not yet: “AI tools draft and summarize. They do not send, book, pay, or change records on their own.” Keep that sentence in. It is the line you will consciously cross later, and you want crossing it to be a decision.

Yes, something already runs unattended: you have moved from chatbots to AI agents, and the policy needs guardrails. Clause: “Automated AI actions are listed here with their owner and limits: [task, owner, what it may do, what it may spend, what it may never do]. Every automated action is logged, and [name] reviews the log [weekly]. Anything that moves money, signs, or deletes requires a person to approve it each time.”

8. Who owns this policy?

One name, not a committee. In a two-person shop it is you.

Clause: “[Name] owns this policy, approves new tools, answers questions about it, and updates this page.”

9. Who owns AI-assisted work?

Clause: “Work produced with AI assistance on company time is company work, like any other work product.”

If you publish or sell what AI helps produce (marketing, designs, written products), add: “Material we intend to protect or sell is substantially written or made by a person, with AI used as a drafting tool, and we keep a record of that.” Purely machine-generated material is on uncertain ground for copyright, and a human hand in the work is the practical protection.

10. When will you review it?

Clause: “This policy is reviewed every six months and whenever a tool is added, a tool changes its data terms, or AI starts acting on its own instead of drafting. Last reviewed [date]. Next review [date].”

Assemble it

Put the clauses in this order, each under a one-line heading: tools, accounts, data, review, customers, automation, who owns the policy, who owns the work, review date. Add the business name, the owner’s name, and today’s date at the top. If it runs past two pages, cut. A policy people read beats a policy that covers everything.

Then do the part no generator can do for you: walk the team through it once, ten minutes, with two examples from your actual work. Here is a customer email, here is how you ask the tool for a reply without pasting the customer’s details. The policy starts existing the moment somebody uses it.

Can I just have ChatGPT generate it?

Yes, and it works better than most generator websites, as long as you feed it the ten answers instead of asking for an AI policy in one line. Paste the question list with your answers, ask for one page in plain English, and then read every line, because it will invent a clause or two you did not ask for. Do not paste anything from question three into the prompt. Describe the categories, not the data. For the broader question of what a general chatbot is and is not good for, see should you just use ChatGPT.

When a generator is the wrong tool

  • You are in a regulated field and the policy touches patient, client, or account data. Generate the draft, then have counsel or your compliance person read it before it goes out.
  • You have an HR function and an employee handbook. The AI policy belongs inside the handbook, in the handbook’s voice, and HR should own the integration.
  • You hold government or enterprise contracts with their own AI clauses. Those clauses win, and your policy has to say so.
  • You already run agents in production. One clause is not enough. The task list, limits, and logging in question seven become a living document with an owner.

What generated policies usually get wrong

  • They list tools nobody at the company uses and miss the one everybody does.
  • They ban instead of direct, so AI use moves to personal phones.
  • They skip the accounts question, which is where the data risk actually lives.
  • They have no named owner, so nobody updates them.
  • They run five pages, so nobody reads them.
  • They say nothing about AI acting on its own, which is the shift that matters most in the next year or two.
  • They have no review date, so the tool list is wrong by spring.

If you want the framework behind the questions, NIST publishes the AI Risk Management Framework and a companion profile for generative AI. Both are written for large organizations. The ten questions above are the same ideas at small-business size: know your tools, know your data, keep a person accountable.

Where this fits

A policy keeps casual AI use from costing you. It does not tell you where AI would actually pay off in your business, which is different for a plumber, a property manager, and an accounting firm. That is what the assessment is for. If the team needs to learn the tools before any policy means much, that is what we do in training.

Questions people ask

Is there a free AI policy generator?
Several, and most produce a reasonable generic draft. The limit is not the price, it is the inputs: a generator only knows what you tell it, and most ask too little. The ten questions on this page are the inputs a good generator would ask for. Answer them and you can use any tool, a chatbot, or a blank page and get the same result.
Can I ask ChatGPT to generate my AI policy?
Yes. Give it the ten answers rather than a one-line request, ask for a single page in plain English, and read every clause, because it will add a few you did not ask for. Keep the actual sensitive data out of the prompt and name the categories instead.
Does a generated AI policy need a lawyer?
Not for most small businesses, where it is a set of operating decisions written in plain English. If you handle patient, client, or financial account data, or you are in a regulated field, have counsel or your compliance person read the draft before it goes out.
How long should an AI policy be?
One page, two at most. Every clause on this page is a sentence or two for a reason: the policy only works if people read it and remember it. If a generator gives you five pages, cut it to the clauses that match your answers and drop the rest.

Sources

Want this figured out for your business?

The assessment tells you where AI is worth it for you, fixed price, and the fee comes off the build.

Get an assessment →

new guides by email

When a new guide is published, you get it.

That is the whole list. We send a confirmation email first, and leaving takes one click.