AI for HR works best as a fast first-draft writer: job postings, interview questions, policy FAQs, onboarding plans, survey themes and feedback wording. It works badly as a judge of people. This playbook gives HR teams seven ChatGPT prompts you can copy, the model's real answers, and what to check before you use each one. It also covers what not to paste, and why AI must never decide or rank candidates or employees on its own.
How can HR teams use AI?
HR teams use AI to draft and summarize: job postings, interview questions, employee FAQs, onboarding plans, survey themes and the wording of hard conversations. You give the facts, the AI writes a draft in seconds, and an HR professional checks it. AI should not decide, score or rank candidates or employees. People review every decision about a person, and HR or legal sets the rules for how AI is used.
AI drafts, people decide. Use it for text, never for the hiring or performance decision itself.
Give it your facts. Pay range, duties, policy text. Without them it invents plausible ones.
Every draft gets checked against the source: the approved offer, the policy, the survey comments.
Keep personal data out unless your company has approved a tool for it. Use placeholders such as [Employee].
Results vary by tool. ChatGPT, Claude, Gemini and Copilot word things differently, but the prompts work in all of them.
[02]
What AI is good at in HR today (and what it isn't)
A chat assistant writes the most likely text for the words you give it. It is very good at structure, tone and speed. It knows nothing about your company, your policies or your people unless you paste them in, and it has no judgment about whether a person is a good hire.
Good at | Not good at, or not its job |
|---|---|
First drafts of job postings and announcements | Deciding who to interview, hire, promote or let go |
Interview questions, with notes on what a strong answer includes | Scoring or ranking candidates and employees |
Turning a policy into a plain-language FAQ | Knowing your policies, or what the law requires |
Onboarding plans and checklists | Knowing your systems, people and dates |
Grouping survey comments into themes | Exact counts you can publish without checking |
Making feedback specific and kind | Formal warnings, investigations and terminations |
The left column saves real time. The right column is where AI output can hurt someone, so keep those tasks with people. For the general prompt method behind every example below, see the pillar guide AI prompts for work.
[03]
ChatGPT prompts for HR: seven tasks with real output
Each task has a prompt you can copy, the model's real answer (trimmed with "…"), and what to check. The answers come from gpt-5-mini, a small model from OpenAI. The company, Fernway, and every name and number are made up. Your tool will word things differently.
1. Write a job posting
Start with what most people type:
Write a job description for a payroll specialist.Job title: Payroll Specialist
Position overview
We are seeking a detail-oriented, organized Payroll Specialist to manage and execute accurate, timely payroll processing for our employees. …
Working conditions
- Office-based or hybrid schedule (specify as applicable)
…
Metrics / KPIs (examples)
- Payroll accuracy rate (target >99.5%)
…
Compensation & benefits
- Competitive salary (provide range if available)That answer ran to more than 600 words, with blanks to fill and targets nobody set. Give it the facts instead:
Task: Write a job posting for a Payroll Specialist.
Context:
- Company: Fernway, a 300-person logistics company with offices in Denver and Salt Lake City.
- The role runs payroll every two weeks for about 300 employees in Colorado and Utah, using ADP.
- Must-haves: 2+ years running multi-state payroll; ADP or a similar payroll system; Excel lookups and pivot tables.
- Nice-to-haves: CPP certification; payroll for hourly warehouse staff.
- Hybrid: 3 days a week in the Denver office.
- Pay range: $62,000 to $72,000. Benefits: health and dental insurance, 401(k) with a 4% match, 15 days of paid time off.
- Reports to the Payroll Manager.
Audience and tone: Experienced payroll people. Plain, specific and friendly. No buzzwords such as "rockstar" or "fast-paced".
Format: Job title, a 2-sentence intro, "What you'll do" (5 bullets), "What you need" (must-haves only), "Nice to have", "Pay and benefits". Under 250 words.
Rules: Use only the facts above. Don't add requirements, benefits or perks that aren't listed. After the posting, list anything a candidate would expect that is missing, under the heading "Missing:".Job title: Payroll Specialist
Fernway is a 300-person logistics company with offices in Denver and Salt Lake City. You will run biweekly payroll for employees in Colorado and Utah, reporting to the Payroll Manager and working hybrid (3 days/week in the Denver office).
What you'll do
- Process biweekly payroll for ~300 employees in Colorado and Utah using ADP.
- Reconcile payroll reports, verify hours and deductions, and resolve discrepancies.
- Maintain payroll data accuracy with Excel lookups and pivot tables.
- Coordinate with HR and benefits to apply taxes, deductions, and leave correctly.
- Prepare regular payroll documentation and support audits or inquiries.
…
Pay and benefits
- Salary: $62,000–$72,000.
- Benefits: health and dental insurance, 401(k) with a 4% match, 15 days paid time off.
Missing:
- Educational requirements (degree or equivalent).
…
- Exact work schedule/hours and expected start date.
- Information on paid holidays and payroll processing deadlines.What to check: The pay and benefits are right, but look at "What you'll do". The prompt asked for five duties and gave only one, so the model wrote four more. They sound reasonable, and they are still guesses. Give the real duties, or ask for fewer bullets. The "Missing:" list is useful: it tells you what to decide before you post. Have the posting reviewed through your normal HR approval before it goes live.
2. Draft structured interview questions
Task: Draft structured interview questions for a Payroll Specialist.
Context: The role runs payroll every two weeks for about 300 employees in Colorado and Utah, using ADP. The interview is 45 minutes with the hiring manager. We want to learn about three things: accuracy under deadlines, payroll for more than one state, and explaining pay problems to employees who are upset.
Format: For each of the three areas, write 2 questions that ask about past experience ("Tell me about a time..."), 1 follow-up question, and one line on what a strong answer includes. Then suggest one 10-minute practical exercise that uses no real employee data.
Rules: Every question must be about the job. We ask every candidate the same questions. Don't ask about personal life. Don't score or rank candidates.Area 1 — Accuracy under deadlines
- Q1: Tell me about a time when you had to complete payroll on a tight deadline and discovered an error late in the process. What did you do step by step, and what was the outcome?
…
- What a strong answer includes: specific example with timeline, concrete actions taken to correct or prevent errors, clear prioritization under time pressure, communication with stakeholders, and measurable outcome …
…
Area 3 — Explaining pay problems to upset employees
- Q1: Tell me about a time an employee came to you upset about a paycheck. How did you handle the conversation and resolve the issue?
…
10-minute practical exercise (no real employee data)
- Setup/instructions (given to candidate): You will be given a short anonymized payroll register (10 fictitious rows) …What to check: The questions are about the job, and the "strong answer" notes give every interviewer the same yardstick. Agree on those notes before the interviews, not after. Several questions ask two or three things at once; split them so candidates can answer one at a time. The AI doesn't know which topics your company keeps out of interviews. HR does, so HR reviews the final list.
3. Build a screening guide for recruiters (not a ranking)
Candidate screening is where AI is most tempting and most risky. Use it to write the guide your recruiters apply, not to judge applications.
Task: Create a screening guide our recruiters can use when they read applications for a Payroll Specialist role. A person reads every application and makes the decision. You are not reviewing any candidate.
Must-haves:
- 2+ years running multi-state payroll
- ADP or a similar payroll system
- Excel lookups and pivot tables
Nice-to-haves:
- CPP certification
- Payroll for hourly warehouse staff
Format: A table with the columns: Criterion | Must-have or nice-to-have | What counts as evidence in an application | What is not enough on its own. Then 3 short reminders for the reviewer.
Rules: Use only the criteria above. Don't add criteria. Don't produce a score, a ranking or a pass/fail rule.Criterion | Must-have or nice-to-have | What counts as evidence in an application | What is not enough on its own
…
Excel lookups and pivot tables | Must-have | Explicit mention of Excel functions/skills such as VLOOKUP/XLOOKUP, INDEX/MATCH, and PivotTables in relation to payroll or reporting … | A vague "proficient in Excel" or "advanced Excel" claim without specific functions or examples …
…
Three short reminders for the reviewer:
- Only use the four criteria in the table when screening—do not add other requirements or infer pass/fail.
…What to check: The table is a solid start. Now read the first reminder: "the four criteria". The table has five. A small slip, but this document guides decisions about people, so every line needs a careful read. HR approves the criteria. Recruiters apply them. Don't paste applications into a chat tool and ask it to score them.
4. Turn a policy into an employee FAQ
Task: Turn the policy below into a short FAQ for employees.
Audience: All employees, including warehouse staff who read it on their phone. Plain words, short sentences.
Format: 6 questions and answers, each answer under 40 words. Then one line: "Full policy: HR portal > Policies > Hybrid Work".
Rules: Use only the policy text. Don't add rules, numbers or exceptions. If something in the policy is unclear, add a final line that starts with "Unclear:".
Policy:
"""
Hybrid Work Policy (effective March 1)
1. Employees in hybrid roles work in the office at least three days per week. Each team chooses its office days. At least one of them must be a Tuesday, Wednesday or Thursday.
2. Fully remote roles must be approved by the department head and HR. Approved remote roles are listed in the HR portal.
3. Home office allowance: hybrid and remote employees may claim up to $400 once every two years for a desk, chair, monitor or headset. Submit the claim in Expensify with the receipt within 60 days of purchase.
4. Equipment bought with the allowance belongs to the employee.
5. Part-time employees follow these rules on a pro-rated basis.
6. Company laptops must not be used on public Wi-Fi without the VPN.
7. Managers may require extra office days during an employee's first four weeks and during quarterly planning weeks.
Questions: people@fernway.example
"""Q: How many days must hybrid employees work in the office?
A: Hybrid employees work in the office at least three days per week. Each team picks the office days. At least one must be a Tuesday, Wednesday or Thursday.
…
Q: What is the home office allowance?
A: Hybrid and remote employees may claim up to $400 once every two years for a desk, chair, monitor or headset. Claim in Expensify with the receipt within 60 days of purchase.
…
Q: Can managers change required office days?
A: Managers may require extra office days during an employee's first four weeks and during quarterly planning weeks.
Full policy: HR portal > Policies > Hybrid Work
Unclear: How pro‑rating is calculated for part‑time employees.What to check: Compare every answer with the policy, line by line. These match. The "Unclear:" line is the most useful part: "pro-rated" really is vague, and HR must answer it before the FAQ goes out. Don't let the AI guess. The FAQ also dropped the contact address, so add it back. Always link to the full policy, which stays the source of truth.
5. Draft an onboarding plan
Task: Draft a first-two-weeks onboarding plan for a new Payroll Specialist.
Context:
- Manager: the Payroll Manager. Onboarding buddy: Luis, a Senior Payroll Specialist.
- Systems: ADP (payroll), Expensify (expenses), Slack, the HR portal.
- IT sets up the laptop and accounts before day 1.
- The first payroll run they will shadow is on day 8. They run the next one with Luis watching, in week 4.
- Required training: data security (online, 1 hour) in week 1.
- The team meets every Monday at 10:00.
Format: A table with the columns Day | Goal | Activities | Owner, for days 1 to 10. Then 3 questions the manager should ask at the end of week 2.
Rules: Use only the people, systems and dates above. If an owner isn't clear, write "not stated". Don't add trainings or meetings that aren't listed, except a daily 15-minute check-in with the buddy.Day | Goal | Activities | Owner
---|---|---|---
Day 1 | Get set up and meet the team | - Confirm laptop & accounts were set up by IT (verify login to ADP, Slack, HR portal, Expensify).<br>- Join team meeting at 10:00 (weekly Monday meeting). … | not stated (IT did setup); Payroll Manager; Luis
…
Day 8 | Shadow first payroll run (Monday team meeting at 10:00) | - Attend team meeting at 10:00.<br>- Shadow the full payroll run in ADP … | Luis
…
Day 10 | Hands-on practice with supervised tasks | - Complete assigned small ADP tasks (e.g., update a test employee record, generate a payroll report) with Luis watching. … | Luis; Payroll ManagerWhat to check: Dates and days. The plan puts the Monday meeting on day 1 and again on day 8, which only works if days 6 and 7 are a weekend, yet both have work planned. Give real dates and the AI can't mix them up. Day 10 assumes ADP has a test employee record. Check whether yours does, and never practice on real employee records. The buddy and the manager adjust the plan. The AI just saves them the blank page.
6. Find themes in survey comments
Task: Find the main themes in these anonymous comments from our quarterly pulse survey (Operations team).
Format:
1. A table: Theme | Number of comments | One short quote (exact words)
2. Two sentences on the overall tone.
3. Three follow-up questions we could ask the team.
Rules: A comment can count toward more than one theme. Quotes must be copied exactly from the comments. Don't guess who wrote a comment. Use only the comments.
Comments:
"""
1. The new shift schedule is posted too late. I found out Friday night that I was working Sunday.
2. My supervisor is great, always explains why we do things.
3. Scanners keep dying halfway through the shift. We lose time hunting for chargers.
4. Shift swaps take forever to get approved.
5. Love the team, but the break room is too small for the night shift.
6. Schedules change after they are posted and nobody tells us.
7. More training on the new warehouse system please. I learned by trial and error.
8. The scanners are old. Half of them don't hold a charge.
9. Pay was wrong twice this quarter (overtime missing). It got fixed but it took two weeks.
10. Good communication from my lead.
11. I'd like to know earlier which shifts I'm on so I can plan childcare.
12. Training for new hires is rushed. They shadow for one day and then they're alone.
13. Overtime hours didn't show up on my paycheck in August.
14. Equipment is the biggest problem. Broken scanners, one forklift out for weeks.
15. Honestly things are better than last quarter.
"""Theme | Number of comments | One short quote (exact words)
---|---:|---
Scheduling/shift notice & changes | 4 | "The new shift schedule is posted too late. I found out Friday night that I was working Sunday."
Equipment (scanners, forklift) | 3 | "Scanners keep dying halfway through the shift. We lose time hunting for chargers."
Training (new system & new hires) | 2 | "More training on the new warehouse system please. I learned by trial and error."
Payroll/overtime errors | 2 | "Pay was wrong twice this quarter (overtime missing). It got fixed but it took two weeks."
Supervision / communication (positive) | 3 | "My supervisor is great, always explains why we do things."
Break room / facilities | 1 | "Love the team, but the break room is too small for the night shift."
…What to check: The themes are right and the quotes are exact. One count is off: only comments 2 and 10 praise a supervisor or lead, not three. Counting is a known weak spot, so spot-check every number before it goes into a slide. Before you paste anything, remove details that could identify a person. In a small team, even an "anonymous" comment about a night shift can point to one person. Follow your company's rules for survey data.
7. Word difficult feedback
Task: Help me turn my notes into feedback I can say to a team member in our weekly 1:1.
My notes: [Employee] missed the Friday report deadline 3 times last month. When the reports arrive they're accurate and well organized. When I mentioned it last week they got defensive. They're great with new hires, two people told me so.
Format: What I could say, in under 120 words, in a natural spoken style. Then one question I can ask to understand what's getting in the way.
Rules: Keep to the facts in my notes. Be specific and kind. Don't add consequences, warnings or judgments about their attitude. One clear ask.I want to say I appreciate how accurate and well organized your reports are and that you’re great with new hires — two people told me that. I noticed the Friday report was late three times last month, and when I mentioned it last week you got defensive. Can you commit to getting the Friday report in on time going forward? Question: what’s getting in the way of getting the report submitted by Friday?What to check: The facts are kept and the ask is clear. But "you got defensive" came straight from the notes, and saying it out loud is a judgment that will likely make the next talk harder. Cut it. Then say the rest in your own words. Notice the placeholder: the AI doesn't need the person's name to help. Use AI for wording only. Performance ratings, warnings and anything formal go through HR.
[04]
A checklist before you use any HR draft
Copy this into your team's notes and run it on every AI draft.
Before I use this AI draft:
[ ] Every fact (pay, benefits, dates, names, systems) matches my source
[ ] Nothing was added that I didn't provide (duties, perks, promises)
[ ] Counts and quotes are checked against the original comments or data
[ ] No real personal data was pasted outside a company-approved tool
[ ] The AI did not decide, score or rank any candidate or employee
[ ] Anything unclear ("Unclear:", "Missing:", "not stated") is answered by a person
[ ] The right reviewer has approved it (HR, the hiring manager, or legal)[05]
Using AI safely in HR
HR handles some of the most sensitive data in a company, so the safe-use rules matter more here than almost anywhere.
Follow your company's data rules first. If IT or HR has approved a tool for employee or candidate data, use that one, with your work account. If not, assume personal data is off-limits.
Don't paste: names with contact details, salaries tied to a person, performance reviews, health or leave details, ID numbers, résumés, or investigation notes, unless your approved tool and your company's rules allow it.
Use placeholders. [Employee], [Candidate A] and [Manager] work as well as real names for drafting.
Keep people in charge of decisions about people. AI must not decide, score or rank candidates or employees on its own. A person reviews every application and every decision, and HR or legal sets the rules for how AI may be used in hiring and performance.
Ask before you roll anything out. Some uses of AI in hiring and employment are regulated in some places. Ask your legal and compliance team what applies to you before you roll this out.
The guide to using ChatGPT safely at work covers what each vendor says it keeps and which settings to check. The idea of a person approving AI output has a name, human-in-the-loop, and HR is the clearest case for it.
[06]
What to automate next
Once your team uses the same prompts every week, the next step is usually a shared prompt library, then AI features inside the HR tools you already have. Some teams go further and want software that acts, for example an assistant that answers routine policy questions from the handbook or prepares onboarding tasks in your systems. That is an AI agent: it takes steps in other systems, not just in a chat window. It needs clear limits, approvals and logs, and it should never make decisions about people. If your team is there, read what is an AI agent? and the overview of AI agents for business leaders before you talk to vendors.
[07]
Common mistakes
Mistake | Why it fails | Fix |
|---|---|---|
Asking AI to rank or shortlist candidates | It has no judgment about people, can reflect unfair patterns in its training text, and can't reliably explain its choices | Use AI to draft criteria and questions; people review every application; HR and legal set the rules |
Asking for more content than you have facts | The model fills the gaps with invented duties, perks or targets | Give the real facts, or ask for fewer bullets |
Publishing a policy FAQ without a line-by-line check | A rule can be dropped, softened or reworded | Compare each answer with the policy; HR signs off |
Trusting survey counts | Models miscount, even on short lists | Spot-check every number before it goes into a report |
Pasting real employee records into a personal account | The data leaves your company's control | Use the approved tool, or replace details with placeholders |
Using AI wording for formal warnings or investigations | The stakes are high and the context is never complete | Keep these with HR and, where needed, legal |
[08]
Frequently asked questions
Can AI screen candidates or résumés?
Not on its own. AI can help write the screening guide, the must-have criteria and the interview questions that people use. A person should review every application and make every decision. HR and legal set the rules for any tool that touches candidate data.
What are the best ChatGPT prompts for HR?
The best HR prompts give the AI your facts, say who will read the result, describe the format, and add a rule such as "use only the facts above". Good starting tasks are job postings, interview question sets, policy FAQs, onboarding plans, survey themes and feedback wording. The seven prompts in this guide are ready to copy.
What is prompt engineering for HR?
Prompt engineering for HR means writing clear instructions so an AI tool produces drafts HR can use, such as job postings or employee FAQs. In practice it is five parts: the task, the context, the audience, the format and the rules. You don't need technical skills. See what is prompt engineering? for the basics.
Is it safe to put employee data into ChatGPT?
Only if your company has approved the tool for that kind of data and you use your work account. Otherwise, keep names, salaries, reviews, health details and ID numbers out, and use placeholders such as [Employee]. Ask IT or HR which tools are approved.
Which AI tool is best for HR work?
ChatGPT, Claude, Gemini and Copilot all handle these drafting tasks well. The best one for you is the one your company has approved, because that decides where your data goes. Many HR systems also now include AI features. Ask your vendor what they do with your data before you switch them on.
Can AI write performance reviews?
AI can help a manager turn notes into clear, specific and kind wording. It shouldn't decide the rating, and it shouldn't see more personal data than your company allows. The manager owns the review, and HR sets the process.
[09]
Next steps
Learn the method behind these prompts: AI prompts for work
Catch wrong facts and counts before you send: how to check AI answers
Know what you can paste: is it safe to use ChatGPT at work?
Write clearer instructions: clear asks, roles, constraints and format
Understand why AI invents details: hallucination basics and quick checks
The other role playbooks: AI for marketing, AI for sales and AI for finance