AI for HR: the gap isn't tool skills, it's work design

The tools are already in the building

CIPD's Autumn 2025 Labour Market Outlook, based on a survey of more than 2,000 UK employers, found that employees in 76% of organisations are using AI tools at work. In the public sector it was 87%, in the private sector 73%.

Read that as a fact rather than a decision. By the time most HR teams sit down to write an AI policy, the technology has been in daily use for a year. The job isn't whether to allow it. It's catching up with something already happening, which is a different and slightly less comfortable exercise.

The same survey found one in six employers expect AI to reduce headcount. That single number is why this lands on the HR desk rather than the IT desk. Tooling questions belong to IT. Questions about what work exists next year, and who does it, belong to HR.

Where AI genuinely helps a people team

The reliable wins have the same shape they do everywhere else. The model handles the language and the volume, a person owns the decision:

• Job descriptions and adverts. Drafting from a role brief, then rewriting for tone and inclusive language. Fast, low risk, easy to check.
• Policy and handbook drafting. First drafts of policies, letters and process notes built from your own existing documents rather than a generic template.
• Engagement survey analysis. Turning hundreds of free-text responses into themes you can act on, as long as you read a sample of the raw comments yourself.
• Interview preparation. Structured question sets and scoring guides built from the actual competency framework.
• Case note drafting. First-pass write-ups of employee relations meetings for the person who was in the room to correct and own.
• Learning content. Turning an existing course or process document into practice scenarios and knowledge checks.
• Explaining the technical bits. Pensions, benefits and payroll language turned into something an employee will actually read.

What none of these do is make the decision. AI can draft the dismissal letter. It cannot decide whether to dismiss anyone, and it should be nowhere near that judgement.

The gap CIPD actually found

In January and February 2026, CIPD surveyed 1,342 people professionals and business leaders with YouGov. The interesting part isn't the adoption number. It's where the confidence runs out.

HR practitioners were reasonably confident about the tools: 67% on learning to use AI tools, 64% on understanding their limitations. So the standard training pitch, teaching HR to write better prompts, is largely solving a problem the profession has already half solved.

Confidence drops sharply at the work-design end. Only around a third of HR leaders felt confident estimating future workforce needs. 40% felt able to design reskilling pathways. 46% felt confident leading job redesign. Those are precisely the tasks AI creates more of, and they are the ones HR is least ready for.

CIPD also found that gap shows up in results. Where HR leaders reported higher confidence in these skills, close to nine in ten said AI had improved workers' job performance. Where confidence was lower, that fell to about half. Confidence here isn't a soft measure, it tracks whether the organisation gets anything back.

The disclosure problem lands on you by default

There's a second thing that arrives on the HR desk uninvited. People are using AI and not saying so.

In April 2026, Wakefield Research surveyed 1,250 office professionals at companies with revenues above $500m across the UK, US, Australia and Japan on behalf of PagerDuty. Two-thirds, 66%, had used AI tools at work despite believing company policy did not permit it. 39% said they would rather use AI without telling anyone. 43% had entered work correspondence into public AI tools, 34% had entered customer data, and 31% had put in financial information or confidential company documents.

The same survey found 86% believed their company had a formal AI policy, and 81% believed the rules were applied differently to leadership. That last number is the one to sit with. A policy people believe is selectively enforced doesn't change behaviour, it moves it out of sight.

The fix isn't a stricter policy. It's a policy specific enough to follow, sanctioned tools good enough that people prefer them, and a route to ask a question without feeling caught. All three are HR work.

Recruitment is the highest-risk corner

Of everywhere HR might point AI, hiring carries the most exposure. Selection decisions affect people's livelihoods, they are regulated, and the failure mode, quietly filtering out a protected group, is invisible unless someone goes looking for it.

The UK government publishes a Responsible AI in Recruitment guide setting out assurance good practice for procuring and deploying AI in hiring. If you are buying anything that touches selection, read it before the demo rather than after the contract.

The other half of the problem runs in the opposite direction. Candidates are using AI too, and trying to detect it is largely a losing game. The more useful response is to change what you assess: work samples, structured exercises, and conversations where the thinking has to happen live.

What AI training for HR should cover

If you are buying training for a people team, this is the shape worth insisting on:

• Work design, not just prompting. Task-level analysis of real roles, so the team can answer what a job contains next year. This is the gap CIPD measured and the one most courses skip entirely.
• A policy people can actually follow. Written during the sessions, on your tools and your data classifications, rather than adapted from a template afterwards.
• Data boundaries drilled on real cases. What goes into a sanctioned tool, what never leaves the organisation, and what an enterprise agreement does and doesn't change.
• Hiring specifics. Bias testing, record keeping, candidate transparency, and the assurance questions to put to a vendor.
• Verification on HR outputs. Every summarised theme, drafted policy and case note checked against the source before anyone relies on it.
• A measure. Baseline the hours on two or three recurring tasks before the training and again a month after. The method is in how to measure AI ROI.

That's how we build AI training for teams at Fautons, on your own policies, adverts and survey data, sized from a single team upwards. The free AI proficiency assessment is a quick way to see where a people team actually stands, and AI training for finance teams covers the same ground for the function next door. If you're comparing suppliers, how to choose an AI training provider has the questions worth asking.