Juliette's Learning Tribe

AI policy for employees: one your staff will actually follow

What goes in a workplace AI use policy, an outline you can copy, and how to make staff follow it: write it with them, keep it short, show examples.

  • ai-and-work
  • leadership
  • checklist
An Idea to Take With You
One next step, with you.

A good AI policy for employees fits on one page. It names the tools people may use, what must never go into them, who checks what comes out, and when to say AI helped. Write it with the people who use the tools, not for them. Put real examples in it, allowed and not allowed, side by side. Then go back to it every few months, because the tools will have changed.

This is practical advice on what to put in and how to make it stick. It is not legal advice.

Why most AI policies get ignored

Most workplace policies are written by somebody who does not do the job. They are long. They are mostly “must not”. They have no examples, so nobody can tell whether the thing they did on Tuesday was allowed.

So people carry on. They use the free app on their phone and say nothing.

That is the real risk of a bad policy. A ban does not stop people using AI. It stops them telling you.

We’re certified to ISO 27001 and ISO 42001. We have a clear brand for every product we build. We also have clear rules for how our systems, and each customer’s space, fit together. Those rules have to hold on every release. So we don’t rely on people reading them. We set up our code so that AI enforces the policies itself, on every release. Everything is documented and tested before it goes out, and we ship with far fewer mistakes.

Write it with the people who use the tools

Get five or six people in a room for an hour. Mix the jobs. Somebody in finance, somebody who writes to clients, somebody in HR, a manager.

Ask them three questions:

  1. What do you use AI for now?
  2. What would you like to use it for?
  3. What worries you about it?

Write down everything. The answers to the first question are the ones you will not hear any other way. You will also find out which tools are already in use, approved or not.

I wrote more about this part in how to introduce AI to your team when some of them do not want it. The short version: the people doing the work usually know where the risks are.

Illustration of three colleagues working together at a table. One points to a simple flow diagram in an open notebook while the other two listen.

An outline you can copy

Ten short sections. If any one of them runs past a few lines, cut it.

  1. Why we have this policy. Two sentences. We want you to use AI well, and we want to keep clients, colleagues and the business safe.
  2. The tools you may use. Name them, and say which version. The free app and the paid company version can come with different terms. Say how to ask for a new one, and who decides.
  3. What never goes in. Client names and details. Anything about a colleague. Health, pay or performance. Passwords. Anything marked confidential. Figures we have not published.
  4. What AI can help with. Real examples from your own teams. First drafts, summaries of public documents, tidying notes, ideas for a plan.
  5. What AI must not do. Make decisions about people. Who gets hired, who gets a warning, whose complaint is upheld. That stays with a person.
  6. Checking. Every number, name, date and rule it quotes gets checked against a source that is not the AI. If you cannot check it, do not send it.
  7. Saying when you used it. When you must tell a client or colleague, and how.
  8. When something goes wrong. Who to tell, how fast, and that you will not be blamed for telling.
  9. Learning. Everyone gets shown how the tools work, on their own work, before they rely on them.
  10. Who owns this, and when we review it. A name, not a department. A date.

Allowed and not allowed, side by side

This is the part people actually read. Use your own examples. Here are some to start.

  • Allowed: drafting a job advert in the approved tool, then editing it yourself. Not allowed: asking any tool to rank the CVs.
  • Allowed: summarising a public report for your team. Not allowed: pasting a client contract into a free app.
  • Allowed: asking for a polite reply to “Client A” about a late delivery. Not allowed: the same request with the client’s real name and order history.
  • Allowed: turning your own meeting notes into actions. Not allowed: recording a colleague and summarising it without telling them.

The free app question matters more than it looks. The UK’s National Cyber Security Centre explains it plainly. What you type into a public tool is visible to the company that runs it. It is stored. It will almost certainly be used to develop their service at some point. The NCSC’s advice is not to include sensitive information in queries to public tools.

Personal data has rules already

If your staff use AI on anything about a person, data protection law applies. Your policy does not switch that on. It is already on.

The Information Commissioner’s Office says how it sees this in its guidance on AI and data protection. If you use AI to process personal data, a data protection impact assessment is how you show you have thought it through. You do it before you start, not after. If you decide you do not need one, write down why.

That is a job for whoever looks after data protection, not for the policy itself. But the policy should say who that person is.

The ICO practises this itself. In 2025 it published the internal AI use policy it gives its own staff. If the regulator thinks it needs one, so do you.

How to make it stick

Launch it in a conversation. Not an email with an attachment. Ten minutes in a team meeting, with the examples on screen.

Keep it to one page. Anything longer goes in a separate note for whoever needs it.

Pair it with learning. A policy tells people the rules. It does not teach anyone to spot a confident wrong answer. That takes practice, on their own work. If you need a starting point for staff, I wrote is it safe to use ChatGPT at work? to be handed round.

Review it every three months. The tools change faster than any policy I have ever seen. Put the date at the top so everyone knows how fresh it is.

Thank the people who report problems. The first person who tells you they pasted something they should not have is doing you a favour. Big visible mistakes are the ones you can fix. The quiet ones nobody reports are the dangerous ones.

What I find exciting about this

I use AI every day. I am dyslexic, and Claude is what finally got what is in my head onto the page. I want every person in every team to have that.

A good policy is how you get there safely. It says yes to far more than it says no to.

Where we come in

Future Fluency is how we teach AI skills for work at Zavmo. It starts with working with AI in your own role, safely, and every lesson uses the learner’s own job and sector. The tutor checks what each person understands and explains it another way when it has not clicked.

If you are writing an AI policy for a team, see how Zavmo works for organisations. Or book a 15-minute team demo and bring your draft policy. We will show you how the learning fits around it.