Generative AI in HR means using large language models and similar systems to produce new content, such as drafts, summaries, answers and translations, for HR work. The model writes; a person still decides. That separates it from an older HR chatbot that picks from scripted replies, and from predictive tools that only output a score.

Where it earns its keep today

The useful work is mostly first drafts and condensing long text:

• Writing a first cut of a job description from a hiring manager’s rough notes, which the recruiter then checks against the real job specification.
• Turning a 40-page employee handbook into plain-language answers, or a Hindi or Tamil version for a plant workforce, reviewed by someone fluent before it goes out.
• Summarising 300 free-text exit interview comments into themes, so HR reads a page instead of a spreadsheet.
• Drafting interview question banks, offer-letter covering notes and manager talking points for a difficult conversation.

Most Indian HRMS vendors now bundle some of this. Our AI in HR tools guide checks which vendor claims hold up, so this page sticks to the concept.

What goes wrong

These models produce fluent text that can be false. The US standards body NIST, in its 2024 Generative AI Profile, calls this confabulation. In HR it looks like a confident answer citing a leave rule your policy never had, or a payslip explanation with the wrong PF figure. Anything with a number, a legal reference or a promise to an employee needs a human check against the source document.

The second problem is data. Pasting a salary sheet or a disciplinary note into a free consumer chatbot sends employee personal data to a third party with no contract covering it. Use enterprise tools with clear data terms, and keep sensitive cases out entirely. Generative AI drafts text; once a system starts acting on that text inside your HRMS, you are into agentic AI, which needs tighter controls.

Frequently asked questions

Can we let generative AI shortlist candidates?

Using it to rank CVs makes it a screening tool, with all the fairness questions that brings; read algorithmic bias in hiring first, and keep a recruiter making the actual shortlist.

Does the DPDP Act stop us using these tools?

No, but employee and candidate data you feed into them is personal data; the Act’s core duties bind from 13 May 2027, and the DPDP employee data page covers what to prepare now, starting with knowing which tool holds what.

Will it replace HR generalists?

It removes drafting time, not judgement; the generalist still owns the grievance call, the policy interpretation and the conversation with the employee, which is where most of the job sits.

Is an AI-written policy legally valid?

Validity depends on the content and how it is adopted, not who typed it, so have HR and, where needed, counsel approve the final text just as you would a human draft.

Before buying, ask the vendor how its output can be traced back to sources, which our page on explainable AI in HR unpacks.