AI for HR: where it helps, where a person must decide

What can AI actually do for an HR team?

Assistants such as Claude, ChatGPT, Copilot or Gemini are good at producing a competent first draft from a clear brief, restructuring text, summarising long documents and spotting patterns in a table or a pile of comments. Most of an HR professional’s week contains work of exactly that kind.

They are poor at knowing your organisation. An assistant has not read your handbook unless you give it the handbook, does not know why the last hire in that role failed, and will fill a gap in its knowledge with something that sounds right. In HR that matters more than in most functions, because what you write is read by an employee as a commitment.

So the working rule is simple: AI drafts, a person decides. The rest of this guide applies that rule task by task.

Which HR tasks does AI help with, and what must a person check?

The table below covers the common tasks across the HR cycle. The third column is the part that gets skipped when a team is busy, and the part that protects the employee and the organisation.

How should you use AI for recruitment without making hiring less fair?

Hiring is where the fairness risk is highest, because a small bias in a job description or a screening rule is repeated across every applicant. The safer pattern is to put AI to work on structure, and keep it away from judgement about individuals.

• Write the criteria first, yourself, from what the job needs. Then ask the assistant to draft the job description and interview questions against those criteria.
• Ask the assistant to review its own draft for wording that could exclude people, and for requirements that are habits and not needs.
• Use the same structured questions and the same scorecard for every candidate, so that people are compared on evidence.
• Never ask an assistant to infer anything personal about a candidate, such as age, religion, caste, marital status, health or disability.
• Keep a person responsible for every decision to progress, reject or hire, and be able to explain that decision without referring to the tool.

Can AI answer employees’ policy questions?

It can, if it answers from your documents and not from general knowledge. An assistant asked about notice periods with nothing to read will describe a plausible policy that is not yours. Give it the handbook, tell it to quote the relevant clause and to say when the handbook is silent, and its answers become checkable.

Two more things make this safe. First, test it: write ten real questions, including two the handbook cannot answer, and score the replies before anyone relies on them. Second, write escalation rules. Harassment complaints, medical matters, grievances and anything involving a named individual should go straight to a person. In India, complaints under the POSH Act have their own process, and an assistant is not part of it.

Where must a person always decide?

Some decisions should never be handed to an assistant, however good the draft looks. These are the ones that change a person’s working life, and the ones where confidentiality is the whole point.

• Hiring, rejection and promotion decisions.
• Performance ratings and calibration outcomes.
• Pay, increments and bonus decisions.
• Disciplinary action and termination.
• Harassment complaints, medical information and grievances.

What about employee data and privacy?

Treat anything that identifies an employee or a candidate as something you do not paste into an assistant unless your organisation has approved that tool for that data. For most drafting work you do not need it: replace names with roles, remove identifiers, and work with the pattern and not the person.

With people analytics, check the size of every group before you report on it. A survey theme attributed to a team of three people is not anonymous. Data protection law, including India’s DPDP Act, places duties on organisations that handle personal data; understand them at principle level and take questions about your own situation to your legal or compliance team. This guide is general professional guidance and not legal advice.

How does an HR team start using AI in a controlled way?

Start small and write things down. A team that agrees three rules and one workflow will get further than a team with a forty-page policy nobody has read.

• Pick one weekly task that is repetitive, low-risk and easy to check, such as first drafts of job descriptions.
• Write a standard brief for it: the role you want the assistant to play, the context, the goal, the constraints, an example and the format you want back.
• Decide who approves the output before it leaves the team.
• Keep a short risk log: what could go wrong, and what stops it.
• Review after a month. Keep what saved time without lowering quality, and drop what did not.

How do you build the skill properly?

Reading about this is a start, but the skill comes from doing the tasks on your own work and having someone check the result. If you teach yourself, choose one task from the table each week, build it, and ask a colleague to review it against the third column.

If you want structure and a credential, AI for HR from Bodhih is a self-paced certification built on this pattern. You produce eight work products on your own HR work, a person grades them, and you sit a proctored exam.