Can we use a few cookies?

They help us see which pages and ads are useful, so we can keep improving the site. We never sell your data, and you can change your mind any time in Cookie settings at the bottom of the page. Cookie Policy

All news

News

What should a small business AI policy include?

By Daniel Rosie, Managing DirectorNovember 20267 min read
What should a small business AI policy include?

In short

The eight sections a small business AI policy needs, who should sign it off, and how to train staff so the policy becomes everyday habit.

A small business AI policy needs to cover eight things: what you use AI for, which tools are approved, what information can and can't go into them, how outputs are checked, which decisions stay with people, how staff ask for a new tool, what to do when something goes wrong, and who owns the policy and when it's reviewed. Keep it as short as your AI use allows, have a director sign it off, and train every member of staff on it before they use AI for work.

That last part matters as much as the wording. A policy that sits in a shared drive changes nothing; one your team has practised with on their own tasks becomes how they work. This article is general information, not legal advice, so check the official guidance and take advice for your own situation, especially where AI will handle personal data.

Why does a small business need a written AI policy at all?

Because without one, every member of staff writes their own rules, one prompt at a time. The National Cyber Security Centre's blog on the hidden risks of shadow AI, published in September 2026, says that "organisations’ policies and guidance, which should reflect and manage the risks associated with using these new technologies, have not always developed at the same pace." It cites one study in which nearly three-quarters of employees (71%) reported using AI tools their employer had not approved.

The NCSC's answer isn't a ban. It says plainly that it isn't recommending people stop using AI, and it describes the goal as reducing risk rather than assuming it can be eliminated. A written policy does that job. It gives your team an approved way to use AI that is easier than the unapproved one, and it gives you something to train against, check against and update.

We've covered whether staff use of ChatGPT and Copilot is safe in a separate article. This one is about the document itself: what goes in it, who signs it and how it reaches your team.

What sections should the policy have?

Eight short sections cover a small business. Each one answers a question your staff will otherwise answer for themselves:

  • Purpose and scope: what you use AI for, and who the policy applies to, including contractors.
  • Approved tools: the named tools, the plan you pay for and the rule that staff sign in with work accounts.
  • Information rules: what never goes in, what goes in only to approved tools, and what's fine anywhere.
  • Checking outputs: who checks what before it leaves the business, and when you tell customers AI was used.
  • Decisions kept by people: the decisions AI drafts for but never makes.
  • New tools: who approves a new tool and what they check first.
  • Mistakes and incidents: who to tell if personal or confidential information goes into the wrong tool, and how fast.
  • Ownership and review: the named owner, the sign-off date and the next review date.

The information rules section deserves the most care, because it's the one staff use every day. Write it with real examples from your business, not abstract categories. "Customer names and addresses never go into a personal account" is a rule people remember. "Personal data must be handled appropriately" isn't. Three tiers work well: a red list that never goes into any AI tool, an amber list that goes only into your approved tools on work accounts, and a green list of public or general material that's fine to use.

The checking section needs names, not good intentions. Say who checks a client proposal, who checks figures in a report and who checks anything that goes on your website. If a person's name is against each type of output, checking happens.

How long should a small business AI policy be?

As long as your AI use needs, and no longer. The Information Commissioner's Office says in its guidance on AI and data protection, in a chapter last updated in March 2023, that "Your governance and risk management capabilities need to be proportionate to your use of AI", which is the right test for a small business.

For a ten-person firm using one approved tool to draft emails, proposals and social posts, two pages does the job. The length grows when AI starts handling personal data in a bigger way: summarising customer complaints, working through HR records, or feeding a tool that makes or shapes decisions about people. At that point the policy points to a data protection impact assessment, and your data protection lead decides what's needed before that use starts. The ICO notes that this guidance is under review because of the Data (Use and Access) Act, so check the current version when you write or update your policy.

Who should write the policy and who signs it off?

A director or the owner signs it off and owns it. That person answers staff questions, approves new tools and makes sure the review happens.

The same ICO chapter explains why ownership matters. It says organisations need to line up their internal structures, roles and responsibilities, training requirements, policies and incentives with their overall AI governance and risk management strategy. Put simply, the policy, the training and who's responsible for what all have to match.

Draft it with two or three of the people who use AI the most, because they know where the grey areas are. Ask your IT provider to confirm the tool settings and account controls you're describing exist on your plan. Then have whoever handles data protection in your business read the information rules before the director signs.

How do you get staff to actually follow it?

Train them on it, using their own work. Emailing a PDF and asking people to confirm they've read it gets you a signature, not a change in habits.

Good policy training does three things. It walks through each rule with an example from your business, so people see what the red, amber and green lists mean for the documents on their desk. It gets everyone practising on the approved tool, so the approved route is also the one that gets them good results. And it covers the checking step on real outputs, so people learn what a confident but wrong answer looks like.

This is where we come in. Our private AI training for staff is shaped by your own AI policy and the tools you license, whether that's Microsoft Copilot, ChatGPT, Claude or another approved tool. If you're writing your first policy, our AI for Business session suits organisations creating an AI policy and teams new to AI or using it informally. It covers accuracy, privacy, bias and keeping people in the loop, and finishes with an action plan for your first uses. Private training runs for groups of up to 12, at your premises anywhere in the UK, at a venue or online, with one fixed price per group confirmed before you book. Attendees rate our workshops 5/5, and we've signed the Responsible AI Use Campaign's 10 principles.

When should you review the policy?

Put a date on the front page and review it at least once a year. Review it sooner when any of these happens: you approve a new tool, your provider changes its terms or features, someone reports a mistake, a team starts using AI on personal data, or the ICO or NCSC publishes new guidance that affects you.

Each review is a short meeting, not a rewrite: ask staff what's unclear, check the tool list matches what people use and update the examples. When the policy changes, run a short refresher.

Frequently asked questions

Can we use a template AI policy?

Yes, as a starting point. Rewrite every section with your own tools, your own examples of red, amber and green information and the names of the people who check and approve.

Who should own the AI policy in a small business?

A named director or the owner. They sign it off, approve new tools and make sure it's reviewed.

Does an AI policy replace our data protection policy?

No. It sits alongside your data protection and IT policies and points to them, so staff follow one set of rules for personal data whichever tool they use.

Should staff be trained on the AI policy?

Yes. Training on the approved tools with your own tasks turns the policy into everyday habits. New starters need it in their first week, and everyone needs a refresher when the policy changes.

Train your team on your AI policy

Send us your policy, or tell us you're still writing one, and we'll build a session on your team's real work and your approved tools, with a fixed price before you book. Call +44 1382 699595 or plan your team's AI training.

Share