Safe starting point: use AI to structure drafts and questions, not to make hiring, promotion, discipline, compensation, or termination decisions. Remove personal data and require human review.

Before using ChatGPT for HR work

  • Follow your organization’s privacy, security, records, and AI-use policies.
  • Do not paste employee or candidate personal data into an unapproved system.
  • Review for bias, accessibility, local law, contractual terms, and policy conflicts.
  • Keep a qualified human responsible for every employment decision and final communication.

10 practical HR prompts

1. Draft a job description

Draft a job description from the approved role brief below. Separate outcomes, essential responsibilities, essential qualifications, and optional qualifications. Use inclusive, plain language. Do not infer credentials, years of experience, location, salary, or physical requirements. List missing approvals at the end.

2. Check job-ad clarity

Review this job ad for ambiguous expectations, unnecessary jargon, duplicated requirements, inaccessible wording, and qualifications that may not relate to the work. Do not make legal conclusions. Return quoted issues, why each may reduce clarity, and a neutral alternative.

3. Create structured interview questions

Create a structured interview guide for [role] using the approved competencies below. For each competency, provide one behavioral question, consistent follow-ups, evidence indicators, and a 1–5 anchored rubric. Exclude questions about protected or personal characteristics.

4. Build an onboarding plan

Create a 30/60/90-day onboarding plan for [role]. Use these approved outcomes, systems, stakeholders, and training resources. Separate learning, relationships, supervised practice, and measurable checkpoints. Do not invent access permissions or policy requirements.

5. Summarize anonymized feedback

Analyze the anonymized feedback below. Identify recurring themes, frequency, representative paraphrases, disagreements, and limitations of the sample. Do not infer identity, intent, demographic patterns, or performance. Suppress themes based on fewer than [threshold] responses.

6. Draft a policy outline

Create a policy outline for [topic] based only on the approved requirements supplied. Include purpose, scope, definitions, responsibilities, process, exceptions, records, review cycle, and open legal/operational questions. Label it “Draft for HR and legal review.”

7. Prepare a manager conversation

Help a manager prepare for a conversation about [workplace topic]. Provide a factual opening, open questions, listening prompts, points requiring policy confirmation, and a neutral recap template. Do not diagnose, threaten, promise an outcome, or make a legal determination.

8. Improve an employee announcement

Edit this employee announcement for clarity, empathy, accessibility, and actionability. Preserve all approved facts. Separate what is changing, why, who is affected, when, required actions, support, and where questions go. Flag any missing date, owner, or policy link.

9. Design a training outline

Create a [duration]-minute training outline for [audience] on [approved topic]. Include learning objectives, scenarios, knowledge checks, facilitator notes, accessibility needs, and items requiring subject-matter review. Do not present generated examples as policy.

10. Audit a draft for unsupported claims

Review this HR draft against the supplied policy and source documents. List claims with direct support, claims that need a source, contradictions, unclear ownership, and missing escalation paths. Quote the relevant source passage for every supported claim.

How should HR teams review AI output?

Check factual accuracy first, then privacy, fairness, accessibility, legal or policy implications, tone, and operational feasibility. For selection tools, consistent job-related criteria and documented human oversight are essential; a polished explanation is not evidence that a recommendation is fair or valid.

How do you get less generic HR output?

Supply the approved role outcomes, audience, jurisdiction for human review, internal terminology, source policies, constraints, and desired format. Tell the model what it may not infer. Ask it to list missing information separately instead of filling gaps.

Keep building: see prompt engineering fundamentals or use the prompt optimizer review checklist.