← All articles
AI Governance6 min read

How to write an AI use policy for your company

An AI use policy sets what staff can share, which tools are approved, and who reviews AI output. Here is how to write one your team will actually follow.

How to write an AI use policy for your company

A company AI use policy is a short document that tells employees which AI tools they may use, what data they may put into them, who approves new tools, and who is accountable for AI-assisted work. It exists so staff can use AI confidently without exposing customer data, code, or confidential plans.

Why does your company need an AI use policy?

Because the alternative is already happening: employees are using AI tools whether or not a policy exists. People paste meeting notes into free chatbots, upload spreadsheets for quick analysis, and draft client emails with whatever tool is open. Most of it is harmless. Some of it sends sensitive data to services the company has never reviewed, under terms nobody has read.

A policy is not about banning that behaviour. Bans push AI use underground, where you cannot see it at all. A clear policy names the approved tools, draws a bright line around data that must never leave the building, and gives people a fast route to approve new tools. Visibility, not prohibition, is the goal.

What should an AI use policy include?

Keep it short enough that people actually read it. A useful policy covers six things:

  • Approved tools: the AI products employees may use, and which plan — data handling differs between free and enterprise tiers.
  • Data rules: what may never enter any AI tool — personal data, credentials, unreleased financials.
  • Human accountability: AI output is a draft. The person who sends, publishes, or ships it owns its accuracy.
  • Disclosure: when clients must be told AI was involved in a deliverable.
  • Approval route: how to request a new tool, who decides, and how long it takes.
  • Review cadence: when the policy itself is revisited.

How do you classify data for AI tools?

Use tiers your staff already understand. Public information — marketing copy, published documentation — can go into any approved tool. Internal information — process documents, anonymised examples, draft plans — goes only into approved tools with contractual data protections. Restricted information — personal data, health or financial records, passwords, legal matters — goes into no external AI tool unless a system was specifically approved for it.

Write the tiers with examples, not abstractions. "Do not share confidential data" forces every employee to guess. "You may paste a redacted support transcript; you may not paste one with names, emails, or account numbers" tells them exactly where the line sits. Most policy failures we see are guessing failures, not malice.

Who should own and enforce the policy?

Give it one named owner — usually IT, operations, or a small AI governance group — not a committee that meets twice a year. The owner approves tools, answers edge-case questions, and updates the list when a vendor changes its data terms. Enforcement should be mostly structural: provide the approved tools and make approval faster than the workaround. We help companies put practical AI governance around the systems they deploy, with policies written for how people actually work, not shelf documents nobody opens.

Keep the core policy to about two pages. Put the principles and data tiers in it, and tool-by-tool detail in a living appendix the owner can update without re-issuing the whole document.

Frequently asked questions

What happens if an employee uses an unapproved AI tool?

A good policy treats a first incident as a process problem, not a crime: find out what data was shared, assess the exposure, and check whether the approved toolset failed a real need. Repeated or deliberate violations are a conduct matter. Policies that lead with punishment drive AI use underground, where it is invisible and riskier.

Does a small company really need an AI use policy?

Yes, and it needs a simpler one. A ten-person company can cover the essentials in a page: approved tools, data that stays out, and who approves exceptions. The risks — customer data in a free chatbot, unreviewed AI output sent to clients — do not scale down with headcount.

How often should an AI use policy be updated?

Review it at least every six months, and immediately when an approved vendor changes its data-handling terms or a new category of tool appears. A policy older than a year may describe tools that no longer exist in that form.

Field note: The policy drafts that fail share one trait: they were written by people who do not use the tools, for people who do. One workshop fixed it in an afternoon — we listed the AI tools staff were already using and wrote the rules around that reality instead of an imagined lockdown. Adoption stopped being the problem the same day.

Related service AI products

Building something worth shipping?

We take on a small number of AI product engagements. Tell us what you are building — we reply quickly.