ChatGPT for HR: 15 prompts you can use today

A manager has sent scattered notes and wants a job posted today.

Use the approved intake notes below to prepare a one-page brief for [role]. Organize it into: business outcome, first-90-day work, essential skills, learnable skills, interview evidence, and open questions. Keep each requirement tied to a task in the notes. Mark unsupported details as missing. Do not invent salary, location, seniority, or headcount approval. Finish with five questions for the hiring manager. Notes: [paste approved role notes]

Before you use the result: Make sure every must-have skill connects to work the person will actually do.

You have an approved brief and need readable candidate-facing copy.

Draft a job description from this approved role brief for [audience]. Use sections for the role purpose, responsibilities, essential skills, learnable skills, and hiring process. Replace vague phrases such as rockstar with concrete work. Keep requirements to those in the brief. Leave placeholders for compensation, benefits, location, and employment terms unless supplied. Ask before adding any qualification. Approved brief: [paste brief]

Before you use the result: Confirm pay, benefits, location, and selection requirements before publishing.

A candidate should know what the assessment involves before starting.

Write a warm, direct assessment invitation under 180 words using only these details: [role], [task format], [expected time], [deadline and time zone], [permitted AI tools], [support contact], [accommodation contact], and [next step]. If a detail is missing, show a placeholder. Explain what the candidate will submit. Do not promise an interview or job. Return a subject line and email body; do not send anything.

Before you use the result: Check the time commitment, time zone, accessibility route, and allowed-tool policy.

Recruiters keep answering the same process questions.

Create a candidate FAQ using this approved hiring process: [paste process]. Cover stages, expected timelines, assessment format, permitted tools, preparation, accessibility requests, and who to contact. For each answer, name the supporting section. If the process does not answer a question, put it in a separate list for the recruiting lead. Do not add policy commitments.

Before you use the result: Remove answers whose policy support is missing or out of date.

The panel needs questions that do not repeat the assessment.

Using the role brief and assessment rubric below, draft a 45-minute interview plan. Allocate time for introductions, three job-relevant questions, follow-ups, candidate questions, and closing. For each question, state the skill, evidence to listen for, and a neutral follow-up. Ask about demonstrated work; do not infer personality or protected characteristics. Do not score a real candidate. Brief: [paste] Rubric: [paste]

Before you use the result: Have the panel check relevance, difficulty, and consistency before the interview.

You need an initial task to review with the hiring manager.

Draft a practical assessment for [role] from the approved brief below. Include a realistic task, supplied materials, deliverables, time expectation, allowed tools, an accessibility discussion point, and a rubric with weights adding to 100. Make every criterion observable in the work. Label the rubric as a draft for calibration, not a validated hiring test. Include two pilot checks and questions for the manager. Brief: [paste approved brief]

Before you use the result: Pilot the task, check feasibility, and calibrate the rubric with subject-matter experts.

Two interviewers interpret a score differently.

Review this draft rubric without evaluating any candidate: [paste rubric]. Find vague terms, overlapping criteria, unobservable expectations, and criteria unrelated to the role. For each issue, suggest a specific rewrite and a fictional example of observable evidence. Keep not observed separate from poor performance. End with a short panel-calibration agenda.

Before you use the result: Check that the examples illustrate the rubric rather than becoming a hidden answer key.

Several teams need to coordinate a new starter’s first week.

Prepare a first-week onboarding plan for [role] using [approved checklist] and [team working practices]. Use role placeholders instead of personal data. Return a table of day, activity, owner, dependency, and completion evidence. Separate required setup from suggested learning. Flag missing owners and access dependencies. Do not create accounts, invite people, or change calendars.

Before you use the result: Confirm owners, working hours, equipment, and access before sharing the plan.

Employees need a plain-language version of a policy.

Use only the approved policy below, version [version], effective [date], to draft an employee FAQ. Include the relevant section for each answer. Preserve eligibility conditions, exceptions, and approval requirements. If the policy is silent or contradictory, say so and name the question for the policy owner. Do not add legal interpretations or advice about an individual employee. Policy: [paste approved text]

Before you use the result: Ask the policy owner to verify every answer and confirm the current version.

A revised policy needs a clear announcement.

Compare these approved old and new policy excerpts: [old version] and [new version]. Draft a short announcement describing what changed, who the stated policy applies to, its effective date, required actions, and the approved contact. Separate changes explicitly supported by the text from questions needing clarification. Do not infer retroactive effects. Return a draft only.

Before you use the result: Verify the date, affected audience, and exceptions against the approved policy.

The team needs practice reviewing AI output.

Create a 60-minute workshop for HR colleagues learning to use AI for drafting. Use fictional role-intake notes and a fictional onboarding checklist. Include one exercise in spotting invented facts, one in checking a policy citation, and one in improving a prompt. Add facilitator notes and a simple review checklist. Do not use real employee or candidate records.

Before you use the result: Run the exercises yourself and make sure the errors are explainable.

Managers want useful conversations about a new workflow.

Draft eight open-ended questions for a manager check-in about this new team workflow: [description]. Focus on clarity, workload, blockers, support, and next steps. Avoid questions that solicit medical details or other sensitive personal information. Include a short way to explain the purpose of the conversation and record agreed actions.

Before you use the result: Choose questions appropriate to the relationship; do not turn responses into automated ratings.

The team has counts but needs an understandable weekly update.

Using this aggregate recruiting table [paste counts with stage definitions and dates], show each stage-to-next-stage rate with numerator and denominator. Check for missing stages, inconsistent cohorts, and zero denominators before calculating. Separate what the data shows from possible explanations. Draft five questions to investigate. Do not infer causes or evaluate individual people. Verify arithmetic step by step.

Before you use the result: Recalculate the numbers in a spreadsheet and confirm all rows use the same cohort.

A polished demo needs to be tested against your workflow.

Turn these requirements [paste requirements] into a demo script. For each, ask the vendor to show a concrete workflow, the human review step, required data, permissions, failure behavior, export path, and plan limitations. Include one deliberately incomplete input and one correction to a generated output. Separate currently available features from roadmap statements. Do not score a vendor without evidence.

Before you use the result: Ask to see the exact configuration you would buy, not only slides.

You need to decide whether a small trial deserves expansion.

Prepare a pilot review from these aggregate results: [baseline period], [pilot period], [task counts], [time including review], [correction counts], [software cost], and [team feedback]. State differences in scope or missing data before comparisons. Calculate results only where denominators are available. Suggest continue, revise, or stop criteria for human approval. Do not invent savings, causal claims, or guaranteed outcomes.

Before you use the result: Check comparable task difficulty, review time, and adoption before calling it a success.