How to Use AI in HR Safely: A Practical Starting Point for UK HR Teams

AI can save HR teams a lot of time. Drafting, summarising, analysing survey comments, preparing for meetings, building training outlines.

But HR holds some of the most sensitive information in any organisation. So using AI safely matters more for us than for most teams.

The short version? Start with low-risk tasks, use approved tools, keep personal data out unless you're sure it's protected, and always keep a human making the decisions.

I rolled out Microsoft Copilot to an HR team in my last in-house role, so this is based on what I learnt doing it for real. Here's the practical starting point I'd suggest.

Where can AI genuinely help HR?

In my experience, AI is most useful for the tasks that eat time without needing much judgement.

Things like:

• First drafts. Policies, letters, FAQs, job adverts and manager guidance.

• Summarising. Long documents, meeting notes or email chains.

• Making sense of comments. Grouping survey or exit interview feedback into themes.

• Preparing for conversations. Thinking through how to structure a difficult meeting and what questions might come up.

• Training. Building a first outline for a workshop or induction session.

• Plain English. Turning something dense or technical into something people will actually read.

I've shared some of the prompts I use in Five Microsoft Copilot Prompts Every HR Professional Should Know.

Where should HR be careful?

This is where it gets more serious.

I'd be very cautious about using AI for:

• Decisions about individuals. Recruitment shortlisting, performance ratings, disciplinary outcomes or redundancy selection.

• Anything involving special category data. Health information, for example, which HR handles a lot.

• Sensitive casework. Grievances, investigations and disciplinaries, unless you're using an approved tool and have thought through the data risks.

AI tools can get things wrong. They can sound very confident while being completely wrong. And they can reflect bias in the data they've learnt from.

That doesn't mean HR can never use AI in these areas. But it does mean a person needs to stay firmly in charge of the decision, and you need to understand the risks first.

Step 1: Find out what people are already using

Before you do anything else, find out what's already happening.

Some people will be using approved tools like Copilot. Others might be using ChatGPT, Claude or something on their phone.

The biggest risk I see is employee information going into a personal or unapproved AI account. Usually not because anyone is being careless, but because nobody has told them where the line is.

Step 2: Talk to IT early

HR and IT need to work together on this.

IT will know what tools are approved, what's blocked and what settings are in place. You might also find you've already got AI features in systems you're paying for.

HR brings the understanding of people, processes and risks. Neither can do this well alone.

Step 3: Put simple guidance in place

People need to know what's acceptable.

A short AI policy or guidance note should cover which tools are approved, what they can be used for, what information must never be entered, and who to ask if they're not sure.

Keep it short and practical. If nobody reads it, it isn't helping. I've written more about why every organisation needs an AI in the workplace policy.

Step 4: Think about data protection from the start

Employee data is personal data. UK GDPR applies whether you're using AI or not.

If you're planning to use AI with employee information, involve whoever looks after data protection in your organisation. Depending on the risk, you may need a Data Protection Impact Assessment.

Also check what happens to the information you put in. Is it stored? Could it be used to train the tool? Different tools, and different account types, work differently.

Step 5: Train people properly

Training needs to go further than how to write a good prompt.

People also need to know:

• What not to put into an AI tool.

• How to check what it gives them back.

• When AI isn't the right tool at all.

Confidence levels will vary a lot across a team. Some people will use AI every day. Others tried it once, got a rubbish answer and gave up. Those groups need different support.

Step 6: Start small and learn

Pick one or two low-risk tasks and try them. See what works. Share what people learn.

When I rolled out Copilot, the biggest lesson was that success depended far more on people, trust and professional judgement than on the technology. I wrote about it in What I learned from rolling out AI to an HR team.

Does every HR problem need AI?

No.

I'm a big believer in using the right tool for the job. Sometimes that's AI. Sometimes it's a feature in your existing HR system, a simpler process or a better template. And sometimes it's a person picking up the phone.

AI is one option. It shouldn't be the answer to everything.

What about AI features in your HR system?

It's easy to think of AI as ChatGPT or Copilot. But a lot of HR systems now have AI built in.

Recruitment platforms that rank candidates. Systems that draft job adverts or summarise performance reviews. Chatbots that answer employee questions.

These features can be really useful. But they need the same thinking as any other AI tool. What data does it use? Is it making or influencing decisions about people? Can you explain how it works if someone asks?

If you're not sure what AI your HR systems already have, ask your suppliers. It's a fair question, and a good supplier should be able to answer it clearly.

Where I've landed

I think HR should be experimenting with AI. We can't guide the rest of the organisation if we don't understand it ourselves.

But experimenting doesn't mean putting everything into ChatGPT and hoping for the best. It means knowing what's being used, protecting employee data, giving people clear boundaries and keeping humans making the decisions that matter.

If you'd like help working out where your HR team is now and what to do next, that's what my HR AI Review is for. You can also look at The HRnetics Framework for Responsible AI, my course for HR professionals.

What's the one HR task you'd most like AI to help with?

Frequently asked questions

Is it safe to use AI in HR?

It can be, if you use approved tools, protect personal data, give people clear guidance and keep a human making the decisions. The risks come from using unapproved tools or relying on AI for decisions about individuals.

Can HR use ChatGPT?

It depends on the account type and your organisation's rules. Personal or free accounts are generally not suitable for employee data. Check with IT and your data protection lead before using any AI tool with personal information.

What should an HR AI policy include?

Which tools are approved, what they can be used for, what information must never be entered, how outputs should be checked, and who to ask for help.

What HR tasks are best suited to AI?

Low-risk, time-consuming tasks like drafting, summarising, grouping survey comments and preparing training outlines are usually the best place to start.