AI for HR: practical uses, prompts and limits

How can HR use AI for everyday tasks?

The safest and most useful HR tasks for AI are the ones where you supply an approved source and want it reshaped: a policy turned into a plain-language summary, a handbook section turned into an FAQ, a training outline turned into a session plan. The source keeps the output honest, and you can check every sentence against it.

Recruiting has its own guide, because screening and selection raise specific legal questions. This page covers the rest of the people function.

What HR work should AI not decide?

AI should not decide anything that changes someone's working life: performance ratings, promotions, pay, task allocation based on personal traits, disciplinary outcomes or dismissal. It can help you prepare, for example by organising your own notes or drafting a letter after a decision is made, but the decision and the reasoning must belong to a person who can explain and defend them.

Regulators are treating this area seriously. The EU AI Act lists AI systems intended to make decisions affecting terms of work-related relationships, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate performance and behaviour, among its high-risk uses in Annex III. If your organisation operates in the EU, that classification brings obligations worth understanding before any such tool is introduced. This is general information, not legal advice.

What does a reviewed AI policy summary look like?

Consider Lindenfield Studio, a fictional 40-person design studio. The HR lead wants a one-page staff summary of the studio's existing parental leave policy, which runs to eight pages.

The first attempt asked: 'Summarise our parental leave policy for staff.' Without the policy attached, the tool produced a confident summary of what such policies often contain, including a statement that employees receive full pay for a set number of weeks. Lindenfield's policy says nothing of the kind; its enhanced pay terms depend on length of service. A summary like that, circulated to staff, creates expectations the studio has not agreed to and can become a dispute.

The second attempt pasted the full policy and asked: 'Summarise this for employees in plain English in under 300 words. Use only what the policy says. For each point, give the section number. List any question an employee is likely to ask that the policy does not answer.' The result tracked the source closely, and the list of unanswered questions turned out to be the most valuable part: it showed two gaps the HR lead then took to the studio's employment adviser. The final summary carried a line pointing staff to the full policy as the version that applies.

Can AI summarise employee survey comments safely?

It can help, with care. Grouping free-text comments into themes is tedious work that AI handles reasonably well, but in a small company anonymity is fragile. A comment mentioning 'the only person in the Bristol office who works part-time' identifies someone, even with the name removed, and a generated summary can repeat that detail.

Remove identifying details before the comments go anywhere, use an approved tool and plan, and ask for themes with approximate frequency rather than quotes. Then check the counts yourself against the comments, because AI tools can miscount and overstate how common a view is. Report themes to leadership; do not report individuals.

What employee information can go into an AI tool?

Less than people assume. Employee records contain personal data, and some of it, such as health, ethnicity or trade union membership, is treated as especially sensitive under data protection law in many countries. Whether any of it can go into a particular tool depends on your legal basis, the vendor's terms for that plan, where data is processed and your own policy.

A practical rule for most HR teams: draft from policies and templates, not from employee files. If a task genuinely needs personal data, such as summarising a grievance file, treat it as a decision for whoever owns data protection in your organisation, not as a shortcut to take mid-task. Removing a name does not by itself make information anonymous.

How should HR check AI-drafted policies and guidance?

Check the draft against the source document, then against the law and the people affected. The items below are a starting checklist, not a substitute for professional review of policies with legal effect.

Which AI tools for HR should a small team consider?

For drafting and summarising, the general assistant your company already approves is usually enough. Many HR information systems now include AI features for answering policy questions or drafting documents. Those can be convenient, but ask the vendor what data the feature reads, whether your content is used to train their models, how long it is kept and who can see outputs.

Be most cautious with tools that score, rank or monitor employees. Those move from drafting help into decisions about people, which brings the legal and fairness questions above.

When is AI the wrong tool for HR work?

AI is the wrong tool for sensitive conversations, investigations and any message where tone could be read as the organisation's position on a person. A generated letter about a performance concern may sound reasonable and still be wrong for that individual and that history.

It is also the wrong tool when there is no approved source to work from. Asking a general assistant what your leave entitlements or notice periods should be produces a plausible answer, not your policy and not the law that applies to you.

What role does HR play in AI training for everyone else?

In many small companies HR owns the AI rollout, so it helps to practise first. HR is often asked to write the acceptable use policy, plan training and answer staff questions. For organisations covered by the EU AI Act, Article 4 asks providers and deployers of AI systems to take measures to support AI literacy among their staff, taking account of their knowledge and the context of use.

Tosheo.ai is being built to support that work with a planned HR path, role paths for the rest of the team and a manager view of skill progress that does not show chat logs. It is pre-build, so treat this as planned scope. The policy guide linked below is a better first step if you have no AI policy yet.

Sources and scope

European Commission AI Act Service Desk, Regulation (EU) 2024/1689 Annex III, high-risk AI systems ↗

European Commission AI Act Service Desk, Regulation (EU) 2024/1689 Article 4, AI literacy ↗