---
title: "How to Write an AI Use Policy for Work: Approved Tools, Data Rules, Review Steps and Training"
url: https://businesslang.com/guides/how-to-write-an-ai-use-policy-for-work/
published: 2026-11-15
updated: 2026-11-15
---

# How to Write an AI Use Policy for Work: Approved Tools, Data Rules, Review Steps and Training

> Write a practical AI use policy for your company: approved tools, what data staff may enter, when a human must review output, training and enforcement.

- URL: https://businesslang.com/guides/how-to-write-an-ai-use-policy-for-work/
- Type: Buying Guides
- Category: AI Tools
- Author: Matthew Collins, Business Operations and Management Expert; Isabella Rossi, Human Resources Director
- Published: 2026-11-15
- Last updated: 2026-11-15
- Topics: ai policy, acceptable use policy, generative ai, ai governance, data protection, employee training, shadow it, ai tools
- Cite as: Matthew Collins and Isabella Rossi, "How to Write an AI Use Policy for Work: Approved Tools, Data Rules, Review Steps and Training", Businesslang, November 15, 2026, https://businesslang.com/guides/how-to-write-an-ai-use-policy-for-work/

**Short answer:** An AI use policy tells employees which AI tools they may use, what information they may put into them, when a person must check the output before it is used, and what training they need. To write one, list the AI tools already in use, sort them into approved, allowed with limits and not allowed, classify your data into three or four levels with a clear rule for each, define which outputs need human review and who signs off, assign an owner, train everyone before rollout, and review the policy every six to twelve months. Keep it to two or three pages so people actually read it.

Most employees already use AI tools at work, whether or not the company has said anything about it. They draft emails, summarize documents, write code, clean up spreadsheets and record meetings. That is often a good thing. The risk is that nobody has decided which tools are safe, what information can go into them and who checks the output before it reaches a customer.

An AI use policy fixes that in two or three pages. This guide shows small and mid sized businesses how to write one: what it should cover, how to build the approved tool list and the data rules, when a person must review AI output, how to train staff and how to keep the policy current. It ends with a template you can copy.

## What an AI use policy is, and what it is not

An **AI use policy** (or AI acceptable use policy) is an internal rulebook for how employees use AI tools in their work. It is practical and short. It answers the questions people actually have: Can I use this tool? Can I paste this document into it? Do I need someone to check this before I send it?

It is not the same as an AI strategy, which decides where AI should create value, or a full AI governance program, which large companies run to manage AI systems they build or deploy at scale. For a broader framework, the NIST AI Risk Management Framework organizes AI risk work into four functions, govern, map, measure and manage, and NIST's Generative AI Profile (NIST AI 600-1) applies it to generative tools. Both are voluntary, but they are useful checklists once your basic policy is in place.

| Document | Question it answers | Typical length |
|---|---|---|
| AI use policy | How may employees use AI tools today? | 2 to 3 pages |
| AI strategy | Where should AI create value for the business? | A plan or slide deck |
| AI governance program | How do we manage risk for AI systems we build or deploy at scale? | A set of processes and roles |
| Vendor AI review | Is this specific tool safe to buy? | A checklist per vendor |

## What an AI use policy should cover

A good policy has eight short sections. Each one should fit on half a page.

| Section | What it says | Example rule |
|---|---|---|
| Purpose and scope | Who the policy applies to and which tools count as AI | Applies to all employees, contractors and interns, on company and personal devices used for work |
| Approved tools | Which tools are approved, allowed with limits or not allowed | Use the company business plan of the approved assistant; do not use free consumer versions for client work |
| Data rules | What information may be entered into which tools | Never enter customer personal data, passwords or unreleased financials into any tool not on the approved list |
| Human review | Which outputs a person must check before use | Anything sent to a customer, published or used in a decision about a person needs review by a named person |
| Disclosure | When to say AI was used | Tell the reviewer when a draft was AI generated; follow client contract terms on disclosure |
| Intellectual property | Who owns output and how to avoid copying | Do not enter third party confidential material; check AI generated code and text before reuse |
| Training | What training is required and when | Complete the 30 minute AI basics session before using any approved tool |
| Ownership and enforcement | Who owns the policy, how to ask questions, what happens if rules are broken | Ask the policy owner before using a new tool; breaches are handled under the normal disciplinary policy |

## Step 1: Find out which AI tools people already use

Do not write the policy from a blank page. Start with a short, no blame survey that asks every team three questions:

- Which AI tools do you use for work, including free ones and features inside software you already have?
- What do you use them for?
- What kind of information do you put into them?

Then check your own systems. Many tools you already pay for now include AI features: office suites, CRM, helpdesk, meeting and design tools. Your software admin consoles and expense reports will show paid AI subscriptions bought on company cards. If you already keep a software inventory as part of your [small business purchasing process](https://businesslang.com/guides/small-business-purchasing-process/), add a column for AI features.

The result is a list of tools, use cases and data types. It usually shows two things: people are getting real value, and some of them are pasting sensitive information into consumer tools without knowing the risk.

## Step 2: Sort tools into approved, limited and not allowed

Use a three tier list. It is simpler than a long rulebook and easy to update.

| Tier | Meaning | Typical examples |
|---|---|---|
| Approved | Use for work, including internal and confidential data within the data rules | Business or enterprise plans with a signed contract covering data use, retention and security; AI features inside core systems you already vetted |
| Allowed with limits | Use only with public or non sensitive information | Free or personal plans of mainstream assistants, browser extensions, AI image tools |
| Not allowed | Do not use for work | Tools with unclear ownership, no published data terms, or terms that allow training on your inputs with no opt out |

To move a tool into the approved tier, review it like any other vendor. Ask whether your inputs are used to train models, how long data is retained, where it is processed, who can access it, whether admins can manage accounts and see usage, and whether the vendor offers a data processing agreement. Our [vendor risk management guide](https://businesslang.com/guides/vendor-risk-management-b2b-purchasing/) and [SaaS contract checklist](https://businesslang.com/guides/saas-contract-checklist/) list the questions and clauses. Get data use commitments in the contract, not only on a web page.

Approving at least one good general purpose assistant is important. If the policy only says no, people will keep using whatever they like on personal devices, and you lose visibility.

## Step 3: Write simple data rules

The data section is the heart of the policy. Most AI incidents in small companies are not dramatic: someone pastes a customer list, a contract or a salary file into a tool that was never vetted. Prevent that with a table that matches data levels to tool tiers.

| Data level | Examples | Approved tools | Limited tools |
|---|---|---|---|
| Public | Published website text, public product information, press releases | Yes | Yes |
| Internal | Internal process notes, non sensitive meeting notes, draft marketing copy | Yes | No |
| Confidential | Contracts, pricing, financials, unreleased product plans, client documents | Yes, if the use case is approved and the client contract allows it | No |
| Restricted | Customer and employee personal data, health data, payment card data, passwords, API keys, security details | Only in tools specifically approved for that data, with the policy owner's sign off | Never |

Add three plain rules under the table:

1. Never enter passwords, API keys or other credentials into any AI tool.
2. Check client contracts before using AI on client material. Some prohibit it or require disclosure.
3. When in doubt, remove names and identifying details, or ask the policy owner.

If you operate in several countries, privacy rules for personal data differ. This guide is not legal advice; have counsel review the restricted data rules if you handle personal data at scale.

## Step 4: Decide when a person must review AI output

AI tools produce fluent text that can still be wrong, out of date or invented. The policy should say clearly which outputs need human review, who does it and what they check.

| Use case | Risk | Review rule |
|---|---|---|
| Brainstorming, outlines, internal first drafts | Low | No formal review; the user is responsible |
| Internal summaries and meeting notes | Low to medium | User checks names, numbers and decisions before sharing |
| Customer emails, proposals, published content | Medium | A second person or the manager checks facts, figures, tone and commitments |
| Code that goes into production | Medium to high | Normal code review and tests apply; reviewer is told which parts are AI generated |
| Contracts, legal, tax or regulatory text | High | Qualified person reviews; AI output is a draft only |
| Decisions about people: hiring, performance, pay, discipline | High | AI can assist with admin tasks only; a person makes and documents the decision |

Hiring and HR deserve extra care. Several jurisdictions regulate automated tools in employment decisions, and the EU AI Act treats many AI uses in recruitment and worker management as high risk. Our guide to the [EU AI Act rules for HR and hiring software](https://businesslang.com/guides/eu-ai-act-hr-hiring-software-delay/) explains the timeline. The safe default for a small business is that AI never makes or ranks people decisions on its own.

Reviewers should check four things: facts and figures against a source, names and commitments, anything that sounds confident but is not cited, and whether confidential information slipped into an external document.

## Step 5: Set rules for disclosure, ownership and records

Keep this section short and practical:

- **Disclosure inside the company.** Tell the reviewer when a draft was largely AI generated, so they know to check it carefully.
- **Disclosure to customers.** Follow client contracts and any law that applies. Some clients require you to disclose AI use or prohibit it on their work.
- **Meeting recording and transcription.** Tell participants when an AI notetaker is recording, and follow local consent rules. Our guide to [AI meeting notes tools](https://businesslang.com/guides/claude-meetings-ai-meeting-notes-tasks/) covers what to check before turning one on.
- **Ownership and copying.** Treat AI output as a draft you are responsible for. Do not ask tools to reproduce competitors' or third parties' copyrighted material, and check code licenses before reuse.
- **Records.** Keep AI generated content that becomes a business record (a contract clause, a customer reply) in the same system as the rest of your records.

## Step 6: Assign an owner and an approval route for new tools

New AI tools appear every week. The policy needs a simple route for adding them:

1. An employee requests a tool using a short form: tool name, use case, data involved, cost.
2. The policy owner checks the vendor's data terms and security information, using your normal [software trial or proof of concept process](https://businesslang.com/guides/how-to-run-a-software-trial-proof-of-concept/) for anything significant.
3. The owner decides the tier and records it on the approved tool list with the date.
4. The list is published where everyone can see it, such as the intranet or the employee handbook.

In a small business, the owner is usually the operations lead or IT lead. HR, finance and legal or outside counsel should review the policy before launch.

## Step 7: Train everyone before rollout

A policy nobody understands does not reduce risk. Plan a short, practical session of 30 to 45 minutes for all staff, plus a longer session for heavy users and managers.

| Audience | Training content | When |
|---|---|---|
| All staff | What the policy says, the approved tool list, the data table with examples, how to ask questions | Before or at launch, and in onboarding |
| Heavy users | Good prompting, checking facts, spotting invented citations, handling confidential data | Within the first month |
| Managers and reviewers | What to check in AI assisted work, people decisions, how to handle breaches | Within the first month |
| Policy owner | Vendor review, incident handling, keeping the tool list current | Before launch |

If you operate in the EU, the AI Act also requires providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy among staff who use them. That duty has applied since February 2, 2025. A documented training session and attendance record is a sensible baseline wherever you operate.

Add the policy to your [employee onboarding checklist](https://businesslang.com/guides/employee-onboarding-checklist/) so new hires read it and complete the training in their first week.

## Step 8: Launch, measure and review

Roll the policy out like any other change. Our guide on [how to roll out new software to employees](https://businesslang.com/guides/how-to-roll-out-new-software-to-employees/) has a communication plan that works here too. Announce the approved tools at the same time as the rules, so the message is "here is what you can use" rather than only "here is what you cannot".

Then track a few signals:

- Number of approved tool requests and how long approval takes.
- Use of approved tools from admin consoles, and spend on unapproved AI subscriptions in expense reports.
- Incidents and near misses reported.
- Training completion.

Review the policy at least once a year, and review the approved tool list every quarter. Update both when a vendor changes its data terms, when you adopt a major new tool or when a new law applies to you.

## Common mistakes to avoid

- **Banning everything.** Staff keep using AI on personal devices and you lose visibility. Approve a good option instead.
- **Writing a 20 page policy.** Nobody reads it. Keep the main policy to two or three pages and link to details.
- **Vague data rules.** "Do not share sensitive data" is not enough. Give a table with examples.
- **No named owner.** Requests stall and the tool list goes stale within months.
- **Forgetting AI features in existing software.** CRM, helpdesk and office suites now include AI. Include them in the inventory and the rules.
- **No training.** People cannot follow rules they do not understand.
- **Treating it as an IT document only.** HR, finance and client facing teams need to shape it, because they handle the riskiest data.

## AI use policy template

Copy this outline and fill in your details. Keep each section short.

| Section | Fill in |
|---|---|
| 1. Purpose and scope | Who it applies to, which tools count, effective date, owner |
| 2. Approved tools | Table of approved, allowed with limits and not allowed tools, with date last reviewed |
| 3. Data rules | Your data levels table, plus the three plain rules on credentials, client contracts and anonymizing |
| 4. Human review | Your review table by use case, and who reviews |
| 5. Disclosure and records | Internal disclosure, client disclosure, meeting recording, record keeping |
| 6. Intellectual property | Ownership of output, no copying of third party material, code license checks |
| 7. Requesting a new tool | Link to the request form and approval steps |
| 8. Training | Required sessions by role and deadlines |
| 9. Incidents | How to report a mistake, such as data pasted into the wrong tool, and who to tell |
| 10. Enforcement and review | Link to the disciplinary policy, review schedule |

Put the finished policy in your [employee handbook](https://businesslang.com/guides/how-to-create-an-employee-handbook/) and ask every employee to confirm they have read it.

## The bottom line

A good AI use policy is short, practical and positive. Find out what people already use, approve at least one safe tool, write a clear table of what data may go where, require human review for anything that leaves the company or affects a person, name an owner, train everyone and review the rules regularly. Your team keeps the productivity gains, and you keep control of your data and your reputation. Start this week with the tool survey; the rest of the policy follows from what it shows.

## Frequently asked questions

**Q: What is an AI use policy?**

A: An AI use policy, sometimes called an AI acceptable use policy, is a short internal document that sets the rules for how employees use AI tools at work. It covers approved tools, what data may be entered, when output must be reviewed by a person, disclosure, training and what happens when the rules are broken.

**Q: Does a small business need an AI policy?**

A: Yes, if any employee uses AI tools for work, and most do. A short policy reduces the risk of confidential or personal data leaking into consumer tools, inaccurate output reaching customers and unclear ownership of AI generated work. It can be two pages long.

**Q: Should we ban ChatGPT or other public AI tools?**

A: A blanket ban rarely works because people keep using the tools on personal devices. A better approach is to approve one or two business plans whose contracts cover data use and retention, and restrict free consumer versions to public, non sensitive information.

**Q: Who should own the AI policy?**

A: One named person, usually the operations lead, IT lead or a senior manager in a small business, with input from HR, legal or your outside counsel, and finance. The owner approves new tools, answers questions and runs the yearly review.

**Q: How often should an AI policy be updated?**

A: Review it at least once a year and whenever you approve a major new tool, a vendor changes its data terms or a new law applies to you. Because AI tools change quickly, many companies review the approved tool list every quarter.

**Q: Do we need employee consent to use AI tools at work?**

A: Consent rules depend on where you operate and what the tool does, especially for tools that record meetings or assess employees. This guide is not legal advice; check recording, privacy and employment rules with counsel before rolling out tools that process employee or customer personal data.


## Sources

1. [NIST: AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework)
2. [NIST AI 600-1: Artificial Intelligence Risk Management Framework, Generative Artificial Intelligence Profile](https://doi.org/10.6028/NIST.AI.600-1)
3. [EUR-Lex: Regulation (EU) 2024/1689 (Artificial Intelligence Act)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)
