Best AI Prompts for Recruiters

AI summary

Seven AI prompts for in-house recruiters and agency searchers: job description sharpening, sourcing outreach, resume screening with EEOC discipline, interview loop design, panel feedback synthesis, offer conversations, and pipeline health audits. Built to scaffold recruiter judgment, never to make hiring decisions on its own.

Prompt 1

Job Description Sharpener

Most job descriptions sound like every other JD in the category. This prompt produces a JD that actually filters in the right candidate.

I am writing a job description for:

ROLE: [TITLE]
LEVEL: [SENIORITY]
FUNCTION: [DEPARTMENT]
COMPANY STAGE: [STARTUP / GROWTH / ENTERPRISE]

What the role actually does day-to-day:

[BULLET POINTS: not what we WISH the role did, but what the first 90 days will actually be]

What would make someone bad at this role: [WHAT WE ARE FILTERING OUT]
What would make someone good at this role: [WHAT WE ARE LOOKING FOR]
What we actually offer (be specific): [COMPENSATION RANGE, BENEFITS, TEAM CONTEXT]

Draft a JD with:

1. ROLE TITLE that filters for the right level (not inflated).
2. WHAT YOU WILL DO: 4-5 outcomes-based bullets, not task lists.
3. WHAT WE ARE LOOKING FOR: must-haves and nice-to-haves, separated.
4. WHAT YOU WILL FIND HARD HERE: an explicit section about the role's challenges. This filters more effectively than any sell.
5. WHAT WE OFFER: specific compensation range if my state requires it; meaningful detail otherwise.
6. HOW WE HIRE: brief overview of the process so candidates can self-select.

Avoid: "rockstar," "ninja," "family," "wear many hats," "fast-paced," "work hard play hard." Use EEOC-compliant language. Do not include any preference based on protected categories.

When to use: Before posting the role. · Best model: Claude (most disciplined about EEOC language and avoiding cliches).

Prompt 2

Sourcing Outreach Personalizer

Most recruiter outreach gets ignored because it is templated. This prompt produces a message that reads like it was written for one specific person.

I want to reach out to a passive candidate:

NAME: [NAME]
CURRENT ROLE: [TITLE AT COMPANY]
WHAT I KNOW ABOUT THEIR WORK: [SPECIFIC: recent post, side project, talk, paper, public output]

The role I am hiring for: [ROLE TITLE]
Why I think they specifically fit (not generic): [SPECIFIC REASON]
My company's edge for this role (the non-comp angle): [WHAT MAKES THIS WORTH THEIR ATTENTION]

Draft a 120-word outreach message that:

1. Opens by referencing something specific about THEIR work, not their job title.
2. Names the role in one sentence and the specific fit reason.
3. Acknowledges they probably get a lot of recruiter messages.
4. Asks for one small thing (a 15-min intro call, a reply on whether the role even sounds interesting, a referral if not them) not a big commitment.
5. Signs off without "hoping to hear from you!" desperation.

Do not invent shared connections. Do not pretend the company is what it is not. Do not use "reaching out," "touch base," "circle back," or "synergies."

When to use: After you have done 5 minutes of research on the candidate. · Best model: Claude (most disciplined about the no-cliche rule).

Prompt 3

Resume Screening Calibration

Most resume screening is gut-driven, which is fast and biased. This prompt structures the screen so the decision is anchored to the JD, not to vibes.

Here is a resume for a candidate:

[PASTE RESUME]

Here is the job I am screening for:

ROLE: [TITLE / LEVEL]
MUST-HAVES: [LIST]
NICE-TO-HAVES: [LIST]
KNOCKOUT CRITERIA (if any): [LIST]

Screen this resume:

1. MUST-HAVES MET: for each, the evidence on the resume (quote the specific line).
2. MUST-HAVES NOT MET or UNCLEAR: with the specific concern.
3. NICE-TO-HAVES PRESENT: relevant pluses.
4. RED FLAGS: anything that signals a likely mismatch (be specific; do not flag based on protected categories).
5. INTERVIEW QUESTIONS: 3 questions to ask in the phone screen that would resolve the unclear must-haves.
6. RECOMMENDATION: advance / decline / borderline-needs-discussion, with the one-sentence reasoning.

Do NOT flag or weight any of the following: name, age, gender, race, ethnicity, religion, nationality, marital status, family status, disability status, sexual orientation, gender identity, photo, address (beyond visa/location relevance), school prestige absent JD requirement.

This is a screen, not a hiring decision. The phone screen is where the actual evaluation happens.

When to use: Before clicking “advance” or “decline” on each resume. · Best model: Claude. The discipline about not flagging protected categories matters legally.

Prompt 4

Interview Loop Designer

Most interview loops are 5 people asking similar questions. This prompt designs a loop where each interview tests a different thing.

I am designing an interview loop for:

ROLE: [TITLE / LEVEL]
KEY COMPETENCIES the role requires: [LIST 5-8]
AVAILABLE INTERVIEWERS and their strengths: [LIST]
LOOP CONSTRAINTS: [E.G., 4 rounds max, candidate time budget, etc.]

Design a loop:

1. ROUND-BY-ROUND PLAN: for each round, which competency it tests, who interviews, what format (behavioral / technical / case / role-play / portfolio review).
2. WHY THIS ORDER: the reasoning for the sequence (warm up, escalate, save the highest-stakes for after they trust the process).
3. THE OVERLAP: any competency that gets tested in more than one round (intentionally, for calibration).
4. THE GAP CHECK: any competency that no round tests. Flag and suggest a fix.
5. THE DEBRIEF: how interviewers should write up their feedback (rubric, not narrative).
6. THE CALIBRATION QUESTION: the one question every interviewer should ask differently, so the panel sees how the candidate adjusts.

Do not propose questions; propose competencies-per-round and rubric structure. Interviewers will write their own questions.

When to use: Once per new role; refine as the role evolves. · Best model: Claude or ChatGPT. Both handle this well.

Prompt 5

Panel Feedback Synthesis

Five interviewers give you five different reads. This prompt structures the synthesis so the hire/no-hire conversation is evidence-based.

We just finished interviewing [CANDIDATE NAME] for [ROLE / LEVEL].

Here is each interviewer's written feedback:

[PASTE EACH ONE WITH THE INTERVIEWER'S ROLE]

The competencies we were testing: [LIST]

Synthesize:

1. PER COMPETENCY: what the panel observed, with the strongest piece of evidence quoted from feedback. If two interviewers disagreed, surface both.
2. WHERE THE PANEL AGREED: a high-conviction reading.
3. WHERE THE PANEL DISAGREED: the actual divergence and which interviewer is more likely correct given their role.
4. THE LONE READ: any single interviewer's read that diverges sharply, worth weighing.
5. THE OPEN QUESTIONS: what we did not get to test that would resolve the disagreement.
6. RECOMMENDED DECISION: hire / no-hire / extend / additional-round, with the reasoning.

Do NOT blend feedback into a smoothed-over average. Surface the disagreements; that is where the calibration discussion needs to happen.

When to use: Within 24 hours of the panel completing. · Best model: Claude. The discipline about surfacing disagreement instead of smoothing it matters.

Prompt 6

Offer Conversation Drafter

Sending an offer is the moment a hire wins or walks. This prompt drafts the offer call script so you go in prepared, not improvising.

I am about to make an offer to [CANDIDATE NAME] for [ROLE].

What I know about their priorities (from the interview process): [LIST]
What we can offer:

- Base: [AMOUNT]
- Equity / RSU / options: [DETAIL]
- Bonus: [DETAIL]
- Other benefits worth mentioning: [LIST]

Likely competing offers or considerations: [WHAT THEY HAVE TOLD YOU]
My walk-away: [INTERNAL ONLY, FOR PRIVATE REFERENCE]

Draft an offer conversation script that:

1. OPENS WITH THE WIN: "We want to make you an offer" delivered cleanly, no preamble.
2. WALKS THROUGH THE COMPENSATION: in the order the candidate has signaled they care about.
3. NAMES THE TEAM ANGLE: 1-2 sentences about the specific reason they fit this team, beyond comp.
4. ASKS THEIR REACTION: the open-ended question that gets them talking.
5. ADDRESSES LIKELY CONCERNS: the 2 things they might bring up, with prepared response framings.
6. CLOSES WITH A TIMELINE: when we need to hear back, what happens next.

Do not overpromise. Do not commit to negotiating items I cannot move on without my approval.

Tone: warm, professional, confident. They should hang up the call feeling wanted and respected.

When to use: 30 minutes before the call. · Best model: Claude. Tone discipline matters more than speed.

Prompt 7

Pipeline Health Audit

Most recruiters track “applicants in pipeline” and miss the actual health signals. This prompt surfaces them.

Here is the state of my pipeline for [ROLE]:

TOTAL CANDIDATES IN ACTIVE PROCESS: [NUMBER]
BY STAGE: [BREAKDOWN]
TIME-IN-STAGE for the oldest candidates: [BREAKDOWN]
SOURCING CHANNELS BREAKDOWN: [WHERE CANDIDATES CAME FROM]
RECENT DROP-OFFS (candidates we lost in the past 2 weeks): [WHY EACH ONE LEFT]

Produce a health audit:

1. THE BOTTLENECK: which stage is moving slowest and why.
2. THE LEAK: which stage is losing the most candidates and what they are losing them to.
3. THE SOURCING QUALITY: which channel is producing the best advance-rate (not just volume).
4. THE CANDIDATE EXPERIENCE FLAGS: anything in the time-in-stage or drop-off data that suggests we are losing candidates because of how we are treating them.
5. ONE PROCESS FIX worth running this week.
6. THE EARLY-WARNING METRIC I should be watching going forward.

Be direct. Pipeline data is rarely good news.

When to use: End of every week on every active role. · Best model: Claude or Grok. Both willingly point out process issues.

Connectors are now standard

Claude, ChatGPT, and Grok all support connectors that let your AI read live data from your work tools (Gmail, Notion, GitHub, Asana, HubSpot, Stripe, and many more) instead of relying on you to paste context. For recruiters this means the AI can read your Workable / Ashby / Greenhouse data, your Gmail or Outlook outreach threads, your Calendly bookings, or your Notion sourcing wiki.

For recruiters, the connectors worth pairing with these prompts:

Get Smarter About AI Every Morning

Free daily newsletter. Built for people who want to use AI well, not chase every model.

Free forever. Unsubscribe anytime.

Two ways to go further