Short answer: yes. If anyone in your company uses ChatGPT, Gemini, Claude, Copilot or AI features inside other software, and almost certainly someone does, you need an AI usage policy. It doesn't need to be long. A good one fits on two pages and answers six questions: which tools are approved, what data can go into them, when AI use must be disclosed, what must be reviewed by a human, who owns what AI produces, and how to report a problem.

Without a policy, people make their own rules. Some avoid AI entirely and lose real productivity; others paste client contracts into free consumer tools. Neither is what you want. This post explains what each part of the policy should say and why, how to roll it out so people actually follow it, and gives you a starter template in plain markdown that you can copy and adapt. It's a starting point, not legal advice; have it reviewed against your contracts and the rules where you operate.

Why does a company need an AI usage policy?

Four reasons, in rough order of how often they cause real problems:

  1. Data leakage. Confidential or personal data entered into tools with unsuitable terms, where it may be retained or used for training.
  2. Unchecked output. AI-drafted content sent to clients, published or used in decisions without review, including invented facts, wrong figures or fabricated citations.
  3. Client and contract breaches. Many client agreements, NDAs and professional rules restrict where client information can go. Staff may not know an AI tool counts.
  4. Unclear accountability. When AI-assisted work goes wrong, who was responsible?

A policy also has an upside: it gives cautious employees permission to use AI productively, because they know where the lines are.

What should an AI usage policy cover?

SectionQuestion it answersWhy it matters
Approved toolsWhich AI tools can we use, on which accounts?Business plans have better data terms and admin controls
Data rulesWhat data can go into which tools?The biggest single risk
Human reviewWhat must a person check before it's used?Catches errors and invented content
DisclosureWhen do we tell clients or the public that AI was used?Trust, contracts and some legal requirements
Intellectual propertyWho owns outputs; what about third-party content?Avoid infringing others' rights and protect your own
Prohibited usesWhat must never be done with AI here?Clear red lines are easier to follow than judgment calls
Incident reportingWhat if something goes wrong?Fast response limits damage
Ownership and reviewWho maintains the policy?Tools and rules change quickly

How should the AI usage policy handle data?

This is the section that does the most work, so make it concrete. Use a data classification people can apply in seconds, with examples from your own business:

  • Public (published website copy, public marketing material): any approved tool.
  • Internal (process documents, internal meeting notes, general drafts): approved business tools only.
  • Confidential (customer lists, pricing, contracts, financials, source code, unreleased plans): approved business tools only, by authorized staff, for business purposes.
  • Restricted (personal data of customers or staff, health or financial details, ID numbers, passwords and keys, client data under NDA, privileged legal material): not in general AI tools at all, unless a specific system has been approved for that data.

Give examples in your own terms. "Don't paste the customer export from Zoho CRM into a chat tool" is clearer than "don't share personal data." Our guide to AI data privacy and security explains the vendor terms behind these rules.

What review and disclosure rules make sense?

Review. The rule of thumb: a human must review anything AI-assisted that leaves the company, is used in a decision about a person, or states facts, figures, legal or technical positions. The reviewer is accountable for the content as if they wrote it. That means checking facts against sources, not just reading for tone. Our post on why AI makes things up explains why this matters.

Disclosure. Decide when you tell people AI was involved:

  • Customer-facing chat or voice agents should identify themselves as AI.
  • Client deliverables: follow client contracts; some require disclosure or prohibit AI use.
  • Published content: decide a house position.
  • Internal work: usually no disclosure needed, but never misrepresent AI output as independent human verification.

Some jurisdictions have or are introducing disclosure requirements for certain AI uses, so check what applies. Our guide on telling customers they're talking to AI covers the customer side.

How do you roll out an AI usage policy so people follow it?

  1. Ask before you write. Find out which tools people already use and for what. The policy should approve the useful patterns, not just ban things.
  2. Provide approved tools. Rules work when the compliant path is easy. Set up business accounts with SSO before the policy launches.
  3. Keep it short. Two pages people read beat twenty pages nobody opens.
  4. Train with examples. A 30-minute session with real scenarios from your business: "Can I paste this RFP into the AI tool? Can I use AI to draft this performance review?"
  5. Make questions easy. Name a person or channel for "is this OK?" questions.
  6. Review quarterly. AI tools, vendor terms and regulations change often.
  7. Treat mistakes as learning, unless they're deliberate. People who report incidents quickly should be thanked, not punished, or they'll stop reporting.

AI usage policy template

Copy this, replace the bracketed parts, and remove anything that doesn't apply. Have it reviewed against your client contracts, sector rules and local law before adopting it.

1. Purpose and scope

This policy explains how [Company] employees, contractors and interns may use artificial intelligence (AI) tools, including chat assistants, writing and coding assistants, AI features inside other software, and AI agents. It applies to all work done for [Company] on any device or account.

2. Approved tools

  • Approved tools are listed at [link to list], with the account type to use for each.
  • Use approved tools only through company accounts. Do not use personal accounts for company work.
  • To request a new tool, contact [owner/channel]. Do not sign up for AI services for company work without approval.
  • AI features inside approved software (for example, our CRM, helpdesk or office suite) may be used within the data rules below.

3. Data rules

Data classExamplesAllowed in approved AI tools?
PublicWebsite copy, published brochuresYes
InternalProcess documents, internal notesYes
ConfidentialCustomer lists, pricing, contracts, financials, source codeYes, for authorized staff and business purposes only
RestrictedPersonal data, health or financial details, ID numbers, passwords, client data under NDANo, unless a specific system is approved for it by [owner]
  • Remove names and identifying details where the task doesn't need them.
  • Never enter passwords, API keys or access credentials into any AI tool.
  • Check client contracts before using AI on client material. If in doubt, ask [owner].

4. Human review and accountability

  • You are responsible for any work you produce with AI assistance, as if you wrote it yourself.
  • Review AI output before it is sent outside the company, published, or used in a decision about a person.
  • Check facts, figures, names, quotes, citations and legal or technical statements against reliable sources.
  • AI must not make final decisions about hiring, firing, pay, credit, or other significant matters affecting individuals. A person decides.

5. Disclosure

  • Customer-facing AI chat or voice systems must identify themselves as AI.
  • Follow client contracts on disclosure of AI use in deliverables.
  • [Company position on disclosure in published content.]
  • Do not present AI-generated content as independently verified research or expert opinion unless a qualified person has verified it.

6. Intellectual property and confidentiality

  • Do not use AI to reproduce copyrighted material, trademarks or another party's confidential information without permission.
  • Treat AI output as a draft that may resemble existing material; check anything that will be published or sold.
  • Work created with AI tools for [Company] is company work product, subject to your employment or contractor agreement.

7. Prohibited uses

  • Creating deceptive, harassing, discriminatory or illegal content.
  • Impersonating real people or organizations.
  • Monitoring colleagues or customers in ways not approved by [owner].
  • Connecting AI tools to company systems (email, drives, CRM) without IT approval.
  • Bypassing security controls or this policy.

8. Reporting incidents

Report immediately to [contact/channel] if:

  • Restricted or confidential data was entered into an unapproved tool.
  • AI output containing errors was sent to a client or published.
  • An AI tool or agent behaved unexpectedly or took an unintended action.

Prompt reporting is expected and appreciated. Delayed reporting makes problems worse.

9. Ownership and review

This policy is owned by [role]. It is reviewed every [three] months and when significant new tools or regulations arise. Questions go to [contact/channel].

Last updated: [date]

What should the policy cover for AI agents, not just chat tools?

Chat tools are where most policies start, but agents that take actions, such as sending emails, updating records or processing documents, need extra rules: who approves a new agent, what permissions it gets, which actions need human approval, and who monitors it. If you're building or buying agents, add a short section referencing your technical controls; our guide to AI agent guardrails covers what those should be. For a wider view of governance, see responsible AI for business.

Frequently Asked Questions

How long should an AI usage policy be?

Short enough that people read it. Two pages plus an approved-tools list works for most small and mid-sized companies. Put detailed technical controls in separate IT documents.

Should we ban ChatGPT and similar tools?

Bans usually push use onto personal accounts where you have no control. Approving business versions with clear data rules is generally safer and keeps the productivity benefits.

Who should own the AI usage policy?

Usually a senior person with authority across teams, often the COO, CTO or head of operations in an SMB, with input from IT, legal or your external counsel, and HR. The owner keeps the approved-tools list current.

Does an AI usage policy need legal review?

It's wise, especially if you handle personal data, work in a regulated sector, or have client contracts with confidentiality and subcontracting clauses. A lawyer can check the policy against your specific obligations.

How often should we update it?

Review it at least quarterly at first, since tools and vendor terms change quickly, and whenever you approve a new tool, deploy an AI agent, or a relevant regulation changes.

Want help putting AI rules and tools in place?

If you'd like help choosing approved tools, setting them up safely and turning this template into a policy your team will follow, get in touch. We help small teams adopt AI with sensible guardrails. Learn more about our AI automation services.