ChatGPT Prompts for HR: 25 That Actually Save Time

Most HR work is writing. Job adverts, interview questions, rejection emails, policy drafts, review summaries, the same onboarding note for the fortieth time. None of it is hard; all of it eats the day.

That is exactly the kind of work a language model is good at — and exactly the kind where a lazy prompt produces something you then have to rewrite anyway.

Below are 25 prompts that hold up in real use, organised by task, plus the small habits that decide whether you get a usable draft or generic filler.

The rule that makes every prompt work

A prompt without your context is just a template.

Compare these two:

• “Write a job description for a sales manager.”
• “Write a job description for a sales manager at a 40-person B2B software company in Manchester. Hybrid, three days in office. Reports to the Head of Revenue. Manages four people. Salary £55,000–£65,000. We care more about coaching ability than personal sales numbers.”

The first gives you something you have read a hundred times. The second gives you something you can edit in two minutes.

Four things to include every time:

• Company context — size, sector, location, working pattern
• Audience — who reads this, and what they already know
• Tone — plain, warm, formal; and what to avoid
• Format and length — bullets or prose, word count, sections

Hiring and job adverts

1. “Write a job advert for [role] at [company context]. Lead with what the person will actually do in their first 90 days, not a list of requirements. Keep it under 400 words, plain English, no buzzwords like ‘rockstar’ or ‘ninja’.”

2. “Here is our job advert: [paste]. Rewrite the requirements section so it separates genuine must-haves from nice-to-haves. Flag any requirement likely to discourage strong candidates who could do the job.”

3. “Review this job advert for language that may deter women or candidates from underrepresented backgrounds. Point to the specific phrases and suggest replacements.”

4. “Generate 10 screening questions for [role] that can be answered in two sentences each and that reveal whether someone has actually done the work, rather than read about it.”

5. “Write three versions of the same job advert: one for LinkedIn, one for our careers page, and one as a 60-word message for a direct approach.”

Interviews

6. “Create a structured interview guide for [role]. Five competencies, two behavioural questions per competency, and a scoring rubric from 1 to 4 with a short description of what each score looks like.”

7. “Here are the notes from a candidate interview: [paste]. Summarise the evidence for and against hiring, and list the two things we still do not know.”

8. “Suggest follow-up probes for this answer: [paste]. I want questions that test whether the candidate did the work themselves or described a team effort.”

9. “Write a 15-minute take-home exercise for [role] that can be assessed objectively and does not ask the candidate to do unpaid real work.”

10. “Rewrite these interview questions so none of them touch age, family status, health or nationality, while still assessing the same thing.”

Offers and rejections

11. “Write a rejection email for a candidate who reached final stage. Specific, warm, under 120 words, no false promise of future roles unless I say otherwise.”

12. “Draft an offer email for [role] at [salary] including start date, working pattern and probation. Confident in tone, no hedging language.”

13. “A candidate has asked for £6,000 above our offer. Draft three responses: one accepting, one meeting halfway with a review clause, and one holding firm but explaining why.”

Onboarding

14. “Build a 30-60-90 day onboarding plan for [role] at [company context]. For each phase give the outcome, who they meet, and what ‘going well’ looks like.”

15. “Write a welcome email for a new starter joining [team] on [date]. Include first-day logistics and one line about why we hired them specifically.”

16. “Turn this messy process document into a numbered checklist a new starter could follow without asking anyone: [paste].”

Performance and reviews

17. “Here are my rough notes on an employee’s year: [paste]. Draft a review that leads with specific evidence, separates results from behaviours, and ends with two development areas framed as actions rather than criticisms.”

18. “Rewrite this feedback so it is direct but not harsh, and so the person knows exactly what to do differently: [paste].”

19. “Draft talking points for a difficult conversation about [issue]. Include what I open with, the two things I must say plainly, and how I respond if they become defensive.”

20. “Write five calibration questions a manager should ask themselves before submitting a performance rating, designed to surface recency bias and halo effect.”

Policies and internal comms

21. “Draft a plain-English [policy name] policy for a [size] company in [country]. Short sentences, no legal jargon, and mark anywhere I should get professional advice before publishing.”

22. “Rewrite this policy at a reading age of about 14 without losing any of its meaning: [paste].”

23. “Write an internal announcement about [change]. Lead with what changes for the reader, then why, then what happens next. Under 200 words. Do not use the word ‘exciting’.”

24. “Anticipate the 10 questions employees will ask about this announcement, including the awkward ones, and draft honest answers: [paste].”

25. “Turn this long policy into a one-page summary with a short FAQ underneath.”

Three habits that improve every output

Give it an example. Paste a job advert or email you were happy with and say “match this tone”. One good example beats three paragraphs of description.

Ask for options, then choose. “Give me three versions: one formal, one warm, one very short.” Picking is faster than editing, and you often end up combining two.

Push back once. The first draft is rarely the best one. “This is too corporate — redo it as if explaining to a colleague over coffee” usually produces the version you actually send.

Where this genuinely should not be used

This is the part most prompt lists leave out, and it matters more than any of the prompts above.

Never paste personal data. Names, salaries, health information, disciplinary details, anything identifying an employee. Under UK and EU GDPR this is personal data, and putting it into a consumer AI tool is a disclosure you probably cannot justify. Strip it to “an employee” and “[role]” before you paste anything.

Do not let it decide. Automated decision-making that significantly affects someone — who gets hired, who gets dismissed, who gets a rating — carries specific legal obligations in the UK and EU, and the EU AI Act classifies employment-related AI as high risk. A model can help you write. It should not be the thing that chooses.

Check the law yourself. Models produce confident, wrong answers about notice periods, statutory leave and dismissal procedure, and they do not reliably know which country you are in. For anything with legal consequence, draft with AI and verify with a professional.

Keep the human judgement visible. If a hiring manager cannot explain why a candidate was rejected without pointing at a tool, the process has a problem that no prompt will solve.

A sensible starting point

Do not try to use all 25. Pick the task you do most often — for most HR teams that is job adverts or rejection emails — and use one prompt for a fortnight until the output needs almost no editing.

Then save the version that works. The prompt you have refined for your own company is worth far more than any list, including this one.

If you are building out the wider process rather than just the writing, our guide to delegating tasks properly covers the handover side, and the statement of work guide is useful when you are hiring contractors rather than employees.

Frequently asked questions

Can ChatGPT replace an HR team?
No. It replaces typing, not judgement. Every example above still needs someone who knows the company to check it before it goes out.

Is it safe to use ChatGPT for HR work?
For drafting with anonymised information, yes. For anything containing employee personal data, no — that is a data protection issue, not a preference.

Which model should I use?
Any current large model handles these prompts. What changes the output is the context you give it, not the brand.

Will candidates notice AI-written job adverts?
They notice generic ones. An advert with real specifics about the team and the work reads as human whether or not a model helped write it.

Should we tell employees we use AI internally?
Yes. A short, plain statement of where it is and is not used prevents far more trouble than it causes, and in some jurisdictions disclosure is becoming a requirement rather than a courtesy.

The short version

Give the model your actual context, ask for options rather than one answer, and push back on the first draft.

Keep personal data out of it, keep decisions with people, and verify anything legal.

Do that and this stops being a novelty and becomes what it should be — a way to get the writing done by eleven instead of four.