Generative AI in HR

What generative AI does differently

Older HR software followed rules: if the candidate answers "no" to the shift question, send rejection template B. Generative AI produces new content based on patterns learned from huge amounts of text. Ask it to write a job ad for a night-shift forklift operator and it will write one, in seconds, in whatever tone you ask for.

That makes it good at language tasks and unreliable at facts. An LLM will confidently write a policy summary that cites a leave entitlement your company does not offer, or state a law that does not exist. Every use case should be judged by one question: what happens if the output is wrong and nobody catches it?

Generative AI is a component of the wider shift to AI in recruitment. When an LLM is given tools and allowed to take actions, such as messaging candidates or booking interviews, it becomes agentic AI, which raises the stakes on oversight.

Practical use cases

Recruiting

• Drafting job descriptions and ads from an intake call, then editing for accuracy and inclusive language.
• Writing personalized outreach and follow-up messages at scale.
• Generating structured interview questions and scoring anchors for a competency.
• Summarizing interview notes or transcripts into scorecard evidence.
• Answering candidate questions about the role, pay range, benefits and process through chat or text.

HR operations and employee experience

• An HR assistant that answers policy questions from the handbook, with links to the source.
• First drafts of policies, letters and FAQs for HR to review.
• Summarizing engagement survey comments into themes.

Learning and development

• Turning SOPs into training modules, quizzes and job aids.
• Role-play simulations, for example a customer calling about a billing error, for contact-center agents to practice.

The use cases that deliver most are high-volume and low-risk: content a human reviews before it goes out, or answers grounded in your own documents.

Risks and how to manage them

• Fabrication. LLMs invent facts. Ground HR assistants in approved documents, show sources, and route anything about pay, leave or discipline to a person.
• Bias. Models can reproduce bias from their training data, for example in how they describe or rank candidates. Do not let a general-purpose LLM score or rank applicants without testing for AI bias and adverse impact.
• Privacy. Pasting resumes, medical notes or investigation details into a consumer chatbot may send personal data to a third party. Use enterprise tools with contracts that bar training on your data, and set a clear policy on what may be entered.
• Over-reliance. Recruiters who paste AI-written screening notes without reading the transcript lose the judgment they are paid for.
• Candidate use. Applicants use the same tools to write resumes and answer screening questions, which makes polished text a weaker signal. Our article on the AI resume revolution covers how recruiters are adapting.

Legal notes

Using generative AI to help write a job ad carries little legal risk. Using it to screen, score, rank or reject candidates is different. Title VII and the ADA apply to hiring tools whether or not they use AI, and the employer is responsible for discriminatory outcomes even when a vendor built the tool. NYC Local Law 144 requires a bias audit and candidate notice for automated employment decision tools used for NYC jobs. Colorado's AI law, as replaced in May 2026, requires notice, explanations of adverse decisions and human review from January 1, 2027 when automated tools materially influence employment decisions. Under the EU AI Act, AI systems used for recruitment, selection and evaluating candidates are classed as high-risk, with requirements including human oversight, transparency and logging. Rules vary by state and country and are changing quickly.

How to roll it out

Start with a written policy: approved tools, data that must never be entered, which uses need human review, and who owns AI decisions in HR. Pick two or three use cases with clear time savings and low risk, such as job ad drafts, outreach and interview summaries, and measure the before and after. Keep a human in the loop for any output that affects a candidate's or employee's outcome, and review vendor documentation on testing, data handling and audits before buying.