AI Prompts for Recruiters: Sourcing & Screen Questions

AI prompts for recruiters work when you treat the model as a sourcing and screen desk. Feed a role brief, must-haves, and a fairness fence, then demand one artifact: Boolean strings, outreach macros, or phone-screen questions with score anchors. You leave with ROLE / TASK / FORMAT kits for GPT-5.5 Instant, GPT-5.6 Sol, Claude Sonnet 5, or Gemini 3.5 Flash, plus rules that block protected-class guesses and invented job claims. This guide stays on the recruiter desk: find people, open conversations, run a consistent first screen. Job posts, policy drafts, and full panel kits live in the HR prompts guide. PromptMake /text scaffolds a rough ask at https://promptmake.net/text.

Who AI prompts for recruiters help

You fill open reqs and need faster Boolean drafts, LinkedIn or email outreach, and a phone-screen script that every candidate at a stage hears the same way. The patterns fit agency recruiters who jump between clients, in-house talent partners who own a pipeline, and sourcers who hand a shortlist to a recruiter who runs the first call. Hiring managers who want a first-pass screen before a panel also land here.

Skip this page if you want public job-post copy, handbook language, or a full interviewer scorecard for a later-stage panel. Those artifacts sit in the ChatGPT prompts for HR guide. Skip candidate-side mock interviews and thank-you emails; those live in interview prep and job-search libraries. This article stays on employer-side sourcing and the first human screen.

Company AI policy, privacy rules, and employment law still govern what you paste. Redact names, emails, phone numbers, photos, and unique facts that identify a person when you test prompts. Use Candidate A / Role 12 labels in consumer chat. Keep the official file in your ATS. Check vendor rules before you drop candidate data into a public model.

Sourcing prompts that fill the top of funnel

Sourcing fails when you ask the model to "find me senior engineers" with no must-haves, no channels, and no ban list. AI prompts for recruiters should produce search strings and outreach you can run today, tied to a ROLE_BRIEF you control. The model does not browse LinkedIn for you. It drafts Boolean, keyword stacks, and message variants from the brief you paste.

Name the channel before you prompt. LinkedIn Recruiter, GitHub, conference lists, alumni directories, and niche boards need different string shapes. One mega-prompt that mixes Boolean and a cold email in the same pass leaks screening language into public outreach. Split threads: one for search strings, one for outreach.

Keep ROLE_BRIEF and MUST_HAVES in a local file. Feed them into every sourcing prompt. Mark pay, visa, and location rules as typed-from-source or [UNKNOWN]. Guessing a remote policy you cannot honor creates outreach you must walk back.

Boolean and keyword stack prompts

Boolean skeleton:

ROLE: You are a talent sourcing assistant for [company type]. You draft search strings from ROLE_BRIEF and MUST_HAVES only. You never invent titles, tools, or credentials. You never infer race, age, gender, disability, veteran status, or other protected classes from names, schools, or photos.

TASK: Draft [N] Boolean / keyword strings for CHANNEL. Split must-have terms from nice-to-have. Include title synonyms and common misspellings only when they appear in SYNONYM_LIST or when I mark them as approved. Flag any term that acts as an age or school-prestige proxy.

FORMAT: For each string: Label, Channel, Query, Rationale (one line), Risk notes. Cap Query at [character limit if CHANNEL has one]. Prefer AND / OR / NOT structure CHANNEL supports.

ROLE_BRIEF: [title, team, location or remote rule, employment type, seniority]

MUST_HAVES: [skills and behaviors you can defend]

NICE_TO_HAVES: [optional extras]

SYNONYM_LIST: [approved title and tool synonyms]

CHANNEL: [LinkedIn Recruiter | GitHub | Google | niche board]

RULES: Ban "young," "digital native," "native speaker" unless MUST_HAVES names a language skill you will test. Ban school prestige as a knockout. Prefer [UNKNOWN] over a guess. This output is a draft for a human sourcer.

REMINDER: Never infer protected classes. Never invent tools or credentials.

Title-expansion pattern: "TASK: From ROLE_BRIEF, list 8 title variants candidates might use on LinkedIn. Mark each Must, Nice, or Drop. Drop means the title pulls the wrong seniority or function. Do not add prestige schools or age proxies."

Negative-keyword pattern: "TASK: Propose NOT terms that cut contractors, students, or wrong functions for ROLE_BRIEF. Only use negatives I can defend. Ban NOT terms that target protected classes or school names."

Outreach macros for passive candidates

Outreach dies when you ask for a "compelling InMail" with no proof points from the role and no honesty fence. Paste ROLE_BRIEF, why the role exists, and what you can say about pay or remote before you draft.

InMail / email skeleton:

ROLE: You are a recruiter writing assistant. You draft from ROLE_BRIEF and OUTREACH_FACTS only. You never invent headcount, funding, pay bands, visa sponsorship, or product claims.

TASK: Draft a first outreach message for CHANNEL under [word or character cap]. Lead with one concrete reason the role fits a person who matches MUST_HAVES. One clear ask: 15-minute call or reply with interest. Tone: plain, respectful of their time.

FORMAT: Subject under 8 words if email. Body short paragraphs. Close with one next step. No attachments mentioned unless OUTREACH_FACTS lists them.

OUTREACH_FACTS: [why the role is open, team size you may share, pay line you may share or [UNKNOWN: pay], remote rule, one product fact you verified]

MUST_HAVES: [paste]

CHANNEL: [LinkedIn InMail | email | SMS if company policy allows]

RULES: Ban "rockstar," "ninja," "excited to connect," and fake personalization ("I loved your post about X") unless OUTREACH_FACTS names the post. Use [UNKNOWN] for missing pay or visa. Never claim we already spoke unless LAST_TOUCH confirms it.

REMINDER: Never invent company or compensation facts.

Follow-up pattern: "TASK: Draft a follow-up under 80 words. Reference ROLE title and LAST_TOUCH date. One ask. No guilt trip. Ban 'just circling back.'"

Agency-to-client intro pattern: "TASK: Draft a short blurb a hiring manager can approve before I send it to Candidate A. Third person. Facts from ROLE_BRIEF only. Under 60 words. Mark any claim that needs client sign-off with [CLIENT REVIEW]."

Phone-screen question kits that stay consistent

A phone screen is the first human filter after sourcing. AI prompts for recruiters should produce the same question order and the same score anchors for every candidate at that stage. The model drafts the kit. You run the call. You own advance or hold decisions. Do not ask the model for a hire or no-hire line.

Build the kit from MUST_HAVES, not from a resume dump. Resume-first prompts invent questions about gaps, school prestige, and polish. Must-have-first prompts ask about work you can see on the job. Keep illegal topics out of the kit: age, family plans, health, religion, national origin, disability status unless a lawful interactive process applies outside this draft flow.

Store one SCREEN_KIT per req ID. Reuse it across candidates. Change only Candidate A notes after the call. Mixing kit design and live scoring in one chat thread swaps anchors mid-loop.

Screen scripts and score anchors

Phone-screen skeleton:

ROLE: You are a recruiter screen-kit assistant. You draft from ROLE_BRIEF and MUST_HAVES only. You never invent competencies. You never output a hire or no-hire decision. You never ask illegal or protected-class questions.

TASK: Draft a phone-screen kit for STAGE = recruiter screen. Include: Purpose (one paragraph), Time box, Question list in fixed order, Scoring anchors 1 / 2 / 3 per question, Recruiter do and do-not list, Closing script that sets next-step expectations without promising an offer.

FORMAT: Cap at [N] questions. Each question maps to one must-have. Ban brainteasers, "tell me about yourself" as a scored item, and culture-fit scores. Mark gaps with [UNKNOWN]. Header: Draft for recruiter review. Not a hiring decision.

ROLE_BRIEF: [paste]

MUST_HAVES: [paste]

STAGE: recruiter phone or video screen, [N] minutes

RULES: Same questions for all candidates at this stage. Score answers against anchors, not likability. Do not ask about family, health, age, accent, or social media. Prefer [HR REVIEW] if a later-stage legal topic appears.

REMINDER: Never invent competencies. Never output hire / no-hire.

Score anchors belong in the prompt: 1 = no evidence, 2 = partial evidence, 3 = clear evidence tied to the work. You write those anchors with the hiring manager. The model fills question text and keeps the anchors stable.

Closing-script pattern: "TASK: Draft a 40-word close that thanks the candidate, states the next step from PROCESS_NOTES, and avoids timeline promises not in PROCESS_NOTES. Ban 'we will get back to you soon' unless PROCESS_NOTES names a date."

Redacted resume triage before the screen

Some desks want a light resume pass before they book a call. Keep that pass thin. Map evidence to must-haves. Do not score personality. Do not invent work history.

Triage pattern:

TASK: From REDACTED_RESUME and MUST_HAVES, draft a pre-screen triage table. Columns: Must-have, Evidence (quote or paraphrase), Flag ([UNKNOWN] if missing, [BOOK SCREEN] if enough to call, [HR REVIEW] if you would skip). Do not score nice-to-haves as knockouts. Do not mention name, school rank, gaps, or photos. Do not output hire or no-hire. Cap Notes at 20 words per row.

REDACTED_RESUME: [Candidate A, work history and skills only]

MUST_HAVES: [paste]

Fairness audit after triage: "TASK: Audit TRIAGE against MUST_HAVES and FAIRNESS_FENCE. For each Flag, reply Keep, Soften, or Remove. Soften means the flag uses a proxy (school prestige, gap length, polish). Propose a rewrite. Never add a new must-have."

FAIRNESS_FENCE you paste every time: "Do not infer protected classes from names, schools, photos, voice, or gaps. Do not use age proxies. Score only against MUST_HAVES. Prefer [UNKNOWN] over a guess. This draft needs human review and is not a legal opinion."

Step-by-step recruiter prompt workflow

Use one chat thread per artifact: Role 12 Boolean, Role 12 outreach, Role 12 screen kit. If you dump every open req into one long thread, you mix must-haves and invent shared pay lines. Keep ROLE_BRIEF stable while you swap only TASK and FORMAT. Spend free ChatGPT, Claude, Gemini, or PromptMake runs on structure, then recruiter time on fairness review and the live call.

Keep candidate files in the ATS. The model is not your system of record. If your company forbids consumer tools for candidate data, stop and use the approved vendor. The prompt shapes still apply inside that vendor.

Step 1: Lock ROLE_BRIEF and MUST_HAVES

List what you may say. Example: "Role 12. Backend Engineer. Team of six. Remote US. Full-time. Pay band $140k-$165k (comp approved). Must-have: Go, Postgres, on-call. Nice-to-have: Kubernetes. Ban: ninja, rockstar, culture fit, years as a knockout unless a license needs a year count. Visa: [UNKNOWN] until immigration signs off." Ugly notes beat a polished fake brief.

Mark every pay, visa, and location line as typed-from-source or absent. Tell the model to keep absent items as [UNKNOWN]. Guessing "we sponsor" when you mean "we have not checked" creates outreach you cannot honor.

Step 2: Source strings, then outreach

Run the Boolean skeleton for your primary channel. Export the strings into your sourcer notes. Book or list candidates from the ATS or Recruiter seat. Open a new thread for outreach. Paste OUTREACH_FACTS you can defend. Cap personalization to facts you verified.

Optional scaffold: open https://promptmake.net/text, describe "recruiter sourcing Boolean strings and phone-screen kits with fairness fences," generate once, then paste ROLE_BRIEF into the returned structure. Guests get about three runs per day; registered free users get about five. Use a run to shape the prompt, then finish in your chat model.

Step 3: Build the screen kit, then audit

After the hiring manager agrees on MUST_HAVES, run the phone-screen skeleton. Save the kit next to the req ID. Before you use it live, run:

TASK: Audit SCREEN_KIT against ROLE_BRIEF, MUST_HAVES, and FAIRNESS_FENCE. Flag any question that could elicit protected-class information. Flag any anchor that scores polish or likability. Rewrite or remove. Never add a new must-have. Never add a hire decision.

After each call, log evidence in the ATS yourself. Do not paste full call recordings into a consumer chat unless your company AI policy allows it and you redact. Prefer short evidence bullets you typed.

Mistakes that wreck recruiter AI drafts

Mistake 1: Asking the model to "source candidates" with no ROLE_BRIEF. The model invents titles, tools, and a culture pitch. Paste controlled notes instead.

Mistake 2: Mixing Boolean, InMail, and a screen kit in one prompt. Split artifacts. Reuse ROLE_BRIEF; change TASK and FORMAT.

Mistake 3: Allowing culture-fit scores and age proxies in screen kits. If you did not name a job-related must-have, ban it in FORMAT.

Mistake 4: Treating a triage table as a reject decision. Put "Draft for recruiter review. Not a hire or no-hire decision." in ROLE and in the table header.

Mistake 5: Pasting unredacted resumes or offer letters into a consumer chat without an AI policy check. Redact. Use candidate labels.

Mistake 6: Using GPT-5.5 Instant or Gemini 3.5 Flash as the final fairness checker. Fast models fit Boolean drafts and first-pass scripts. Route the fairness audit to GPT-5.6, Claude Opus 5, or Gemini 3.1 Pro, then still read the kit yourself.

Mistake 7: Asking the model to invent a pay band or visa rule for outreach. Paste OUTREACH_FACTS or write [UNKNOWN]. Comp and immigration own those lines.

Mistake 8: Skipping the audit after a long paste. Screen scripts hide "where are you from" and "tell me about your family" style probes you never approved. Prompt for Keep / Soften / Remove against FAIRNESS_FENCE every time.

Model notes for recruiter prompts (mid-2026)

ChatGPT often defaults to GPT-5.5 Instant for fast chat. Instant fits Boolean drafts, outreach variants, and first-pass screen kits when you already locked MUST_HAVES. Keep prompts short: ROLE, TASK, FORMAT, ROLE_BRIEF, MUST_HAVES, RULES. Skip long chain-of-thought slogans.

GPT-5.6 (Sol in API naming as of mid-2026) fits harder edit passes: fairness audits against FAIRNESS_FENCE, contradiction checks between must-haves and nice-to-haves, and outreach that must not invent pay. Give goal, constraints, and format. Drop "think step by step" padding on reasoning-class models.

Claude Sonnet 5 handles long ROLE_BRIEF pastes and tidy screen tables well. Claude Opus 5 fits careful audits when the kit must not invent illegal questions. Claude Fable 5 is the top public tier when your workspace offers it; check your plan. Haiku 4.5 fits short Boolean reshuffles when latency matters more than a deep audit.

Gemini 3.5 Flash fits volume work: many Boolean variants and outreach A/B lines from an updated ROLE_BRIEF. Gemini 3.1 Pro fits hard reasoning over long context when you paste a thick must-have pack and need conflict flags. Hedge on exact menu names in each vendor UI. They shift. Re-check the model picker when you open a new thread.

Prompting split that holds: Instant and Flash get RTF plus a short sample of your house InMail format when you need matching structure. GPT-5.6, Opus 5, and Gemini 3.1 Pro get goal + constraints + format, with an explicit "never infer" rule, a [UNKNOWN] token, and [HR REVIEW] flags. All need your ROLE_BRIEF in the message. None replace a human fairness check before you message a candidate or score a screen.

Build AI prompts for recruiters with PromptMake

Write the rough ask in plain words: artifact type (Boolean, outreach, phone-screen kit), fairness fence, and whether you need a Keep / Soften / Remove audit. Open https://promptmake.net/text and generate a structured prompt once. Expect labeled sections you can fill with ROLE_BRIEF and MUST_HAVES.

Edit titles, pay lines, and visa claims yourself. PromptMake cannot know your req or your ATS. Paste the filled prompt into Instant or Flash for drafts, or GPT-5.6 / Opus 5 / Gemini 3.1 Pro for audits. Keep free-tier runs for scaffolding, not five synonym retries of the same weak ask.

Workflow that sticks: ROLE_BRIEF + MUST_HAVES → PromptMake scaffold → fill redacted facts → fast model draft → reasoning model fairness audit → human recruiter review → source or screen only after that review. Store one template per artifact so you do not rewrite ROLE and RULES from scratch each req.

FAQ

What are the best AI prompts for recruiters in 2026?

The best AI prompts for recruiters lead with ROLE and a fairness fence, paste a ROLE_BRIEF and MUST_HAVES you agreed with the hiring manager, then demand FORMAT for one artifact: Boolean strings, outreach macros, or a phone-screen kit with score anchors. Add a second audit prompt that marks Keep, Soften, or Remove against those packs. Match GPT-5.5 Instant or Gemini 3.5 Flash for drafts and GPT-5.6, Claude Opus 5, or Gemini 3.1 Pro for fairness audits, then still read the draft yourself before you message anyone.

How do AI prompts for recruiters differ from HR prompts?

AI prompts for recruiters target sourcing and the first screen: search strings, passive outreach, and consistent phone-screen questions. HR prompts target broader employer artifacts: public job posts, screening notes against a full competency list, panel interview kits, and policy drafts. Keep those libraries separate so a Boolean string does not leak into a handbook section, and a handbook tone does not land in a candidate InMail.

Can ChatGPT write Boolean search strings for LinkedIn Recruiter?

Yes, when you paste ROLE_BRIEF, MUST_HAVES, SYNONYM_LIST, and CHANNEL limits. The model drafts query text and rationale. It does not search LinkedIn for you or guarantee result counts. Treat every string as a draft you test in the Recruiter seat, then tighten NOT terms that pull the wrong seniority or function.

How do I keep recruiter screen questions fair?

State the ban in ROLE and again in FAIRNESS_FENCE plus a REMINDER line. Forbid protected-class questions, age proxies, and culture-fit scores. Require the same question order and the same anchors for every candidate at that stage. Follow with an audit prompt that flags probes about family, health, age, or accent. A human still runs the call and owns the advance decision.

Should I use GPT-5.5 Instant or GPT-5.6 for AI prompts for recruiters?

Use Instant for Boolean drafts, outreach variants, and first-pass screen kits from a must-have list you pasted. Use GPT-5.6 when you need a careful fairness audit, conflict checks between must-haves and nice-to-haves, or outreach that must not invent pay or visa claims. Run the same packs through both only when you measure quality for a recurring template.

Can I use an AI score to reject a candidate after a phone screen?

Do not treat an AI score as a reject decision. Put that limit in ROLE and in the scorecard header. Screen notes can map evidence to must-haves and mark [HR REVIEW]; they must not output a hire or no-hire fact. Your process, your ATS, and employment rules still require a human decision, so check your company's AI policy before you paste candidate data into a consumer tool.

Can PromptMake help with AI prompts for recruiters on the free tier?

Yes. PromptMake /text turns a rough recruiter ask into a labeled prompt you can aim at ChatGPT, Claude, or Gemini. Guests get about three generations per day; registered free users get about five. Fill in your own ROLE_BRIEF and MUST_HAVES, then paste into Instant, Flash, or a reasoning model for the audit. PromptMake does not source candidates for you and does not store your ATS as a system of record.