Regulation

AI policy template: what an AI use policy for employees should include

Your staff are probably using AI already, with or without approval. This guide covers the rules worth writing down, a three-colour guide to data and how to roll the policy out, and ends with a short AI policy template to start from.

What an AI use policy is and why your company needs one

Many managers assume that until the company approves an AI tool, nobody on the team is using one. More often, someone is already finishing letters or emails with ChatGPT on a personal account, sometimes with the client's name and the facts of the case pasted in, and nobody has said whether that is allowed. A short AI use policy closes that gap, and the AI policy template further down gives you a first draft.

It is a short document that tells your team which tools they may use, with what data and under which checks. Many companies only write one after months of staff using AI to summarise contracts or prepare replies. Use the company does not know about is often called shadow AI.

Gallup reported in May 2026 that 30% of US employees use AI at work several times a week or more, and 15% use it daily. Habits like these grow on people's own initiative, whether or not anyone has set rules.

The first risk is data protection. When an employee pastes a client's name, address or account history into an external tool, your company is processing personal data, and the GDPR (or UK GDPR) applies as it would to any other processing. You need a lawful basis, a provider offering adequate guarantees and a contract covering how it handles that data.

The second is quality. A Harvard Business School and BCG study of 758 consultants found that AI users produced over 40% higher quality work and finished 25% faster on suitable tasks. Used outside its strengths, AI made results worse. Your company answers for whatever reaches the client, so the policy has to say who checks the work.

What an AI use policy for employees should include

The policy breaks down into six sections. For a small business, the whole document fits on two or three pages.

  • Approved tools and accounts

    List the approved tools and the type of account allowed. Free and business versions do not always handle the data you enter in the same way, so check each provider's terms. State that personal accounts are not used for work, and explain how to request a new tool.

  • Allowed and off-limits data

    Classify information with the three colours described below. It is the part people look up most, so keep it to a single page with examples from your own work.

  • Confidentiality and intellectual property

    Confidentiality commitments to clients and partners still apply when staff use AI. Make clear who owns content produced at work, and ask for care with text, images or code that might reproduce someone else's work.

  • Human review and transparency

    Anything that leaves the company is checked by a person, who verifies figures, names and sources. Decide too when clients are told that AI was used, for example in a report or a customer service chat.

  • Security

    Require two-step verification wherever a tool offers it. Ban unapproved extensions or apps from connecting to email or the CRM. Explain how to report an incident, such as a document shared by mistake.

  • Training, contacts and review

    Say what training each person receives before using AI, who answers questions and how often the document is reviewed. Every six months is a sensible rhythm, because tools and their terms change often.

Green, amber, red: what data can go into an AI tool

With three colours, deciding what data can go into a tool takes seconds. Put it on a single page, with two or three examples from your sector under each colour.

  • Green: public or non-personal information

    Text from your website, published regulations or your own drafts with no client names or figures. Fine in any approved tool. For example, asking AI to rewrite the description of a service.

  • Amber: internal or anonymised information

    Meeting notes, internal documents or client cases that have been anonymised, meaning nothing left could identify the person. Only in tools with a business account and checked terms. For example, summarising a standard contract after removing the parties and amounts.

  • Red: data that never leaves

    Passwords, bank details, health data, confidential client documents and personal data that has not been anonymised, unless expressly approved. If you are unsure between amber and red, treat it as red and ask the named contact.

How to roll out an AI use policy in 5 steps

These five steps take the policy from a draft to the way your team works day to day.

  1. 01

    Find out what people use today

    Ask the team, without looking for culprits, which tools they use and for which tasks. A five-question anonymous survey is usually enough. You will learn what to approve, what to replace and which rules are urgent.

  2. 02

    Draft it with the people involved

    Bring together management, whoever handles data protection and one or two regular users. Start from the template on this page and adapt it to your tools, clients and contracts. Then ask your legal adviser to review the text.

  3. 03

    Present it with examples and a quick quiz

    Walk through the policy in a short session using cases from your sector. Finish with five quick questions, such as "can I paste a buyer's file?" or "who do I tell if I send something by mistake?". Wrong answers show you which rule is unclear.

  4. 04

    Train the team

    Rules are easier to follow when people know how to use the approved tools well. Hands-on training shows them how to ask, how to check answers and how to anonymise. If you operate in the EU, Article 4 of the EU AI Act also asks companies using AI systems to ensure a sufficient level of AI literacy among their staff.

  5. 05

    Review and update

    Set a review date, note questions and incidents as they come in, and update the list of tools. Announce each change with a short message and the date it takes effect.

Example

A Galway accountancy practice finds three AI assistants in use

Take a typical case: an accountancy and payroll practice in Galway with twelve staff. The managing partner suspects several people use AI to answer client emails, but has no idea which tools or what data. An anonymous survey shows three different assistants in use, mostly on personal accounts, and someone uploading payslips so the AI can explain them.

The practice approves a single assistant with business accounts and writes a two-page policy from the template. Each colour gets its own examples. Green is asking about a published Revenue rule. Amber is reviewing a draft letter with the client's name replaced by "Client A". Red is payslips, certificates and anything showing a PPS number.

The launch session takes an hour and ends with the five-question quiz. Two people get the same question wrong, about scanned documents, so the partner rewrites that rule. The team then has hands-on training on the approved tool.

To measure it, the practice follows three signals. How many people have confirmed they read the policy, how many questions reach the named contact, and whether personal accounts disappear from work in a second survey three months later. If plenty of questions arrive early on, it usually means people are asking before they paste.

An AI policy template you can adapt

This template is a starting point for a small business. It is not legal advice. Adapt it to your activity and check it with your adviser before you use it.

1. Purpose. This policy governs the use of artificial intelligence tools by all [Company] staff in the course of their work.

2. Approved tools. Only the tools on the attached list may be used, with business accounts provided by [Company]. Personal accounts must not be used for work tasks.

3. Data. Information is classified using the attached colour guide. Passwords, bank details, health data and confidential documents must not be entered into AI tools. Client and employee personal data must be anonymised before use, unless [data protection lead] has given express approval.

4. Human review. Each person is responsible for the output they obtain with AI. All content intended for clients or third parties must be checked before it is sent.

5. Transparency. Clients will be told that AI was used where the contract requires it or where [Company] considers it relevant.

6. Questions, incidents and review. Questions, and any data shared by mistake, must be reported to [contact person] as soon as possible. This policy will be reviewed every [six] months.

The first two pieces of your policy, and three mistakes to avoid

If you have an hour, start with the anonymous survey and a first colour guide with examples from your sector. With those two pieces in place, the rest of the document comes together quickly from the template.

Avoid three common mistakes. A ten-page text nobody finishes. A blanket ban with no approved tool, which pushes people onto personal accounts. And a policy with no named contact, so questions go unanswered.

Obinova is an AI agency based in Barcelona and Paris. Our training prepares your team on ChatGPT, Claude, Gemini and Microsoft Copilot using cases from their own work, on a dedicated video platform or in person. It is a natural companion to launching your policy. Have the final wording checked by your adviser, especially where it touches on data protection and client contracts.

Frequently asked questions

Is an AI use policy a legal requirement?

In general, no law requires a document with that exact name. However, data protection law requires you to protect any personal data your team might enter into these tools, and Article 4 of the EU AI Act asks companies in its scope to ensure a sufficient level of AI literacy among staff. A written policy helps show that you have taken steps. Check your own situation with your adviser.

Should we ban ChatGPT at work or allow it with rules?

For most small businesses, approving one or two tools with clear rules works better. A total ban is hard to enforce and tends to move use onto phones and personal accounts, where the company sees nothing. With an approved tool, business accounts and a colour guide for data, staff know what they can do and you keep control.

How do I adapt an AI policy template to my company?

Replace the bracketed fields with your own details, attach the list of approved tools and fill the colour guide with two or three examples from your sector under each colour. Name a real contact person and set a review date. Then have your adviser check the wording, especially on data protection and client contracts, before you present it to the team.

What data should never go into an AI tool?

As a cautious rule, passwords, bank details, health data, confidential client documents and any personal data that is not strictly needed. Where the task allows, anonymise information first, for instance by replacing the client's name with a generic reference. The exact list depends on your business and on the terms of each tool.

How long does it take to write and roll out an AI use policy?

For a small business, a first version is usually ready within a few weeks. The survey and the draft take a few days. What stretches the timeline is the adviser's review and choosing which tools to approve. Presenting it to the team fits into one short session. It is better to publish a simple version early and improve it at the first review.

What does Article 4 of the EU AI Act require?

It asks providers and companies that use AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff. It has applied since February 2025. In practice, this means training people according to their role and how they use AI. Your adviser can tell you what it means for your company.

Launch your policy with a trained team

Book a free 30-minute video call. We will look at which AI tools your team uses today and which training would help them apply your policy from day one.

Free 30-minute diagnosis →
Free diagnosisWhatsApp

WhatsApp · Reply within 2 hours during business hours