How to Use This Prompt Pack
Each prompt below names its job, the tool it runs best on, and the placeholders you replace. Two disciplines make them work. First, the extraction rule: prompts that summarize and structure are safe to feed candidate data, with job-related fields only, while prompts that draft and analyze get company context, never protected data. Second, the firewall rule: every output gets a human read before it touches a candidate or employee. The model drafts, you decide, and the bias-language sweep runs after every AI draft, not before.
The prompts follow the arc of a real hire: attract, screen, interview, offer, onboard, engage, grow, govern. Fifteen free AI HR prompts, one per job, no filler. Most run on the tools you already pay for, we mapped the full eight-tool HR stack these slots live in.
Why Prompts Beat Tool Shopping in 2026
The case for practicing prompts rather than buying platforms is now a numbers argument. SHRM found roughly one in four organizations using AI in HR, with 64% of those applying it to recruiting; Gartner found 76% of HR leaders expecting to lag without adoption; and LinkedIn found 62% of recruiting professionals optimistic about AI. Adoption is racing ahead, but every one of those deployments runs on the same bottleneck: the quality of the instruction. A team with a disciplined prompt library gets more from a generic assistant than an unequipped team gets from an enterprise platform.
That is why this pack exists: fifteen slots, each covering a recurring People Ops job, each written to hold its shape under real-world paste-and-go use. The chart below is the market context; the prompts are the response.
Sources: SHRM AI in HR survey, 2024 · Gartner HR survey, 2024 · LinkedIn Future of Recruiting, 2024
Attract: Job Descriptions and Outreach (Prompts 1-3)
The attract stage is where AI earned its reputation in HR, writing tasks with clear structure, measurable output, and no decision risk. Three prompts cover it. Run them in sequence, the rewrite first, the title map second, outreach last, because each output feeds the next: the rewritten description sharpens the title map, and the title map makes the outreach hooks specific. The sequence takes under thirty minutes for one role and is the fastest credibility win a skeptical hiring manager will ever see from AI.
Prompt 1: the job description rewrite and bias sweep
Paste your stalest posting with: "Rewrite this job description using the R.E.A.D. standard, Results (top 3 year-one outcomes, measurable), Environment (honest pace, team, tools), Advancement (where the role goes in 2 years), Deal (salary band, benefits, schedule). Keep every true requirement but cap must-haves at 5 and label the rest preferred." Follow with the bias sweep: "Scan for gendered terms, age-coded words, inflated requirements, and clichés like fast-paced family. Quote each finding, explain the risk in one line, and offer a neutral replacement." The rewrite fills the posting; the sweep is what keeps it legal and welcoming.
Prompts 2-3: title maps and three-line outreach
Prompt 2, the title synonym mapper: "I am hiring for [ROLE]. List 10 titles people actually hold while doing this job, 8 distinguishing skills, and 5 companies or industries where this talent concentrates, formatted for boolean searches." Ten minutes here transforms sourcing hit rates, because nobody's job title matches your requisition.
Prompt 3, the three-line outreach writer: "Write outreach messages (hook, evidence, ask) for these profiles: [PASTE WITH ONE SPECIFIC DETAIL EACH YOU NOTED]. Under 90 words, no flattery adjectives, the ask is a 15-minute call. Flag any profile where you could not write an honest hook." The flag instruction is the prompt's best feature, it turns the model into an honesty checker instead of a flattery machine.
Screen: CV Triage Without the Bias (Prompts 4-6)
Screening is where the stakes rise, because outputs here feed real decisions. The three prompts below keep the model on the safe side of the line, reading and structuring, never judging. The distinction is not decorative: extraction prompts are auditable, evaluation prompts are depositions waiting to happen. Keep the model's role to organized reading and the whole stage stays defensible by default.
The extraction rule that keeps prompts 4-6 safe
Prompt 4, the CV structurer: "Extract a screening summary from this CV. Output ONLY: core skills claimed, years by skill area, measurable outcomes, role context (industry, scale, team size), unexplained gaps over 6 months. Do NOT evaluate quality, recommend, or comment on anything personal." This is the extraction rule in practice, the model sees facts, produces structure, and never sees a verdict.
Prompt 5, the scorekeeper: "Here are my 5 must-have requirements: [LIST]. Here is a CV summary: [PASTE]. Score each requirement meet/partially/missing and cite the exact line justifying each score. List which requirements need human eyes because evidence is ambiguous. You are scoring, not deciding." The ambiguity list is the deliverable, it becomes your phone-screen questions. Prompt 6, the gap question writer: "For this CV summary, write the 3 most important clarifying questions for a 15-minute phone screen, job-related, open-ended, assuming good faith." Gaps are information, not verdicts; this prompt turns them into fair questions instead of silent rejections. The complete defensible workflow, including the audit steps, lives in our guide to screening resumes with AI.
Interview and Offer: Kits and Paperwork (Prompts 7-9)
Prompt 7, the kit builder: "Here are my role's must-haves and Results (paste). Derive 5-6 interview competencies; for each give a one-line definition, 2 behavioral questions about past work, and 1 follow-up probe per question. Flag any competency not reliably assessable in an interview." The flag saves you from grading stress tolerance on a fifty-minute call.
Prompt 8, the mock candidate: "Play a candidate for [ROLE]: strong on A, thin on B, left last job in a reorg. Answer realistically, sometimes concise, once evasive. Afterward grade my questioning: did I follow the kit, probe openly, avoid leading, capture quotes?" Fifteen minutes of rehearsal beats a quarter of drift for new interviewers.
Prompt 9, the offer letter drafter: "Draft an offer letter using this structure: [YOUR STANDARD SECTIONS]. Details: [TITLE, BAND, START DATE, MANAGER, LOCATION]. Warm, precise, no legal padding. Flag any placeholder with brackets." Then the consistency check on Claude: "Compare letter, contract summary, and benefits sheet. Are defined terms, names, dates, and salaries identical across documents? List every discrepancy." The discrepancy list is what a signature actually rests on.
Onboard and Engage: The First 90 Days (Prompts 10-12)
Prompt 10, the blueprint builder: "Build a 30-60-90 onboarding skeleton for [ROLE FAMILY]: for each gate, 4-6 observable milestones split into Learn/Deliver/People, explicit success measures, and the top 3 risks of that month. A stranger could verify each milestone." Personalize per hire with: "Adjust these milestones to this person's gaps and strengths (paste profile); add two stretch items if senior." The full method, gates, check-ins, and the buddy contract, is in our guide to 30-60-90 onboarding plans.
Prompt 11, the check-in scripter: "Write scripts for the five onboarding conversations (day 1, week 1, day 30, day 60, day 90): 3 manager questions, 1 manager commitment, and blocked/adjust/intervene decision rules. Under 150 words each so managers actually use them." The word cap is the feature: scripts get used when they fit on a phone screen.
Prompt 12, the verbatim clusterer: "Cluster these survey comments (paste). Output 5-8 themes with counts, 3 representative quotes each (paraphrase anything identifiable), sentiment skew, and a surprises section for themes the scores would not predict. Do not summarize away specifics." The surprises section is where the real findings hide, scores tell you where to look, verbatims tell you why.
Grow and Govern: Development and Policy (Prompts 13-15)
Prompt 13, the IDP drafter: "Draft a 70/20/10 development plan: target rung (paste), review file summary (paste), manager's ambition (paste). 2-3 on-the-job assignments with real stakes, 1-2 named mentoring relationships, 1 formal learning item chosen last. End with: success by [DATE] means [OBSERVABLE CHANGE]." Development plans fail when they are course lists; this prompt forces the 70 first.
Prompt 14, the plain-language rewriter: "Rewrite this policy organized by employee moments (I need time off; when something goes wrong). Plain sentences, every step with owner and timeline, summary first with full policy linked. Keep every substantive rule EXACTLY as-is, flag anything ambiguous in the original." The keep-as-is instruction separates a language rewrite from an accidental policy change.
Prompt 15, the consistency sweep: "Compare these [N] policies. Find: defined terms used differently, contradictory rules, duplicate sections that will drift, and broken cross-references. Numbered fix list with exact locations." Run it quarterly; contradictions breed disputes, and disputes breed the policies that cause them. For the conversation habits that make development plans stick, our guide to weekly feedback rituals pairs directly with prompts 11 and 13.
The Fairness Firewall: The Numbers Behind the Rules
Every prompt in this pack carries guardrails, and the public data explains why they are non-negotiable. Pew Research Center found 71% of US adults oppose employers letting AI make final hiring decisions and 66% would not apply to an employer that uses AI in hiring decisions, while Harvard Business School and Accenture's Hidden Workers study found 88% of employers agree their own filters already vet out qualified candidates who do not match exact criteria. The prompts encode the middle path: AI reads and structures, humans decide, and every output stays auditable.
Treat the firewall as three habits rather than a document: extraction prompts never evaluate people; protected data never enters the chat; and every output that touches a human gets a named human read. Written on one page and reviewed quarterly, those habits are the difference between a prompt library and a liability library.
Sources: Pew Research Center, 2023 · Harvard Business School Hidden Workers, 2021
| Firewall habit | What it blocks | When it runs |
|---|---|---|
| Extraction prompts never evaluate people | Bias entering a hiring decision path | Every screening prompt run |
| Protected data never enters the chat | Compensation, health, or demographic exposure | Before every paste |
| Every output gets a named human read | Unreviewed AI text touching a candidate | Before anything is sent or filed |
| One-page quarterly review | Habit drift as models and rules change | First week of each quarter |
Source: Firewall rules compiled from EEOC and NYC Local Law 144 bias-audit guidance, 2023-2025
Turn Fifteen Prompts Into an Operating System
A closing note on what this pack is and is not. It is the working skeleton of a full People Ops operating system, start with prompts 1, 4, and 10 in week one, add the interview and offer prompts when your next requisition opens, and put the survey and policy prompts on a quarterly calendar. It is not a substitute for judgment: every prompt ends with a human decision, and the fairness firewall runs whether or not anyone is watching.
If you want the complete version, scorecards, kits, audit protocols, and downloadable packs for all fifteen days, AI-Powered HR & Talent Mastery teaches the full system across fifteen days, and day 1 is free with any account. It will hand you a People Ops map and your baseline hiring numbers in one sitting, which is exactly the context that makes these fifteen prompts work hardest.
| Moment | Prompts to install | Operating cadence |
|---|---|---|
| Week 1 | Prompts 1, 4, and 10: description rewrite, CV structurer, onboarding blueprint | Run once on a real role to prove the loop |
| Next requisition opening | Prompts 7-9: interview kit, mock candidate, offer letter | Every new role that opens |
| Quarterly calendar | Prompts 12-15: survey clustering, IDP drafting, policy rewrite, consistency sweep | Survey cycle and policy review dates |
| Every run, always | The fairness firewall: extraction only, protected data out, human read | Before anything touches a candidate |
Source: Cadence compiled from ZLearning AI-Powered HR & Talent Mastery course materials, 2026
Frequently asked questions
▸Do these prompts work with free versions of ChatGPT or Claude?
Yes, every prompt runs on any current chat model, free or paid. Paid tiers help with volume and long documents (a full policy stack fits better in Claude's long context), but the drafting, summarizing, and analysis prompts here produce their full value on free tiers. If you want structure around the practice, see what a genuinely useful free AI HR course should deliver before committing an evening.
▸What makes an HR prompt safe to use with candidate data?
Three rules: the prompt extracts facts rather than evaluates people; the input data stays job-related (strip names, photos, and dates of birth where your process allows); and nothing involving compensation, health, or protected characteristics goes into general-purpose tools. A prompt that asks the model to score a human is a prompt your lawyer will meet someday, the full workflow in our guide to screening resumes with AI shows the boundaries in practice.
▸How do I customize these prompts for my company?
Every prompt carries bracketed placeholders, replace them with your specifics and add one line of company context: industry, size, location. The structure does the work; the context makes it yours. If the output sounds generic, your context line was too thin, and adding two sentences about your stage of growth usually fixes it.
▸Can I share these prompts with my team?
Please do, that is the point. The strongest adoption pattern is one shared document with your company context pre-filled into each prompt, plus your AI ground rules on page one: AI drafts, humans decide, bias check before anything touches a candidate. Teams that share one canonical version avoid the drift that happens when everyone keeps private copies.
▸Which model should I use for which prompt?
Default to whichever capable model you already pay for. The extraction and clustering prompts (4, 12) suit any strong model; the long-document prompts (14, 15) benefit from Claude's context window; the high-volume drafting prompts (1-3) run best on the model your team uses daily, because speed of habit beats marginal quality differences. Switching tools mid-pipeline costs more than it gains.
▸How often should I update my prompt library?
Quarterly, with a light touch. Models improve, so instructions that compensated for weaknesses can be simplified; regulations move, so the fairness language in screening prompts deserves a review; and your own process changes, so placeholders drift out of date. Fifteen minutes per quarter with the shared document is all the maintenance a library like this needs.
▸Are these prompts compliant with AI hiring regulations?
The prompts are written to make compliance easier, not to be a compliance program. They keep the model on the extraction side of the line, which is what bias-audit frameworks expect, but the audit, the disclosure, and the human checkpoints still have to exist in your process. Pair the library with a quarterly review of where AI touches your funnel, and check the rules in your jurisdiction before any tool enters a decision path.