AI Prompts for Recruiters: 10 Templates for Sourcing & Screening (2026)

AI is a drafting and organizing assistant in recruiting, not a decision-maker. It's great for writing outreach, turning a hiring manager's notes into a job description, generating structured interview questions, and summarizing what a candidate said into consistent notes. It must never make or recommend the actual decision — who advances, who's rejected, who's ranked highest. Those are human judgments with legal weight, and a person owns each one.

Three cautions protect you and candidates. First, bias: AI models can reflect and amplify bias in language and evaluation, so write inclusive prompts, keep everything role-related, and have a human check for skewed wording or unfair criteria. Second, consistency and compliance: structured, identical questions and rubrics across candidates are both fairer and more defensible, and several jurisdictions now regulate AI in hiring and automated decision-making — know your local employment law. Third, never paste sensitive personal data (protected characteristics, immigration status, health, full contact details) into a general AI tool; describe candidates by role-relevant qualifications only.

These prompts run on any current model. Routine drafting (outreach, JDs, follow-ups) is fine on an efficiency tier (gpt-5.4-mini, Gemini 3.1 Flash-Lite); a frontier model writes more nuanced interview kits and summaries. Prices as of June 2026 (OpenAI, Gemini).

When to use: a first-touch message to a passive candidate you want to feel human, not mass-blasted.

``` Write a short outreach message to a passive candidate for [ROLE] at [COMPANY — one-line description]. What's genuinely compelling about this role/team: [PASTE 2-3 real points]. Why this person might be a fit (role-related only): [their relevant experience — no personal/protected details]. Tone: warm, specific, respectful of their time. Under 120 words. Lead with what's in it for them, not us. One clear, low-pressure ask. No hype, no fake flattery, no 'rockstar/ninja' language. ```

Why it works: 'lead with what's in it for them' and 'role-related only' produces outreach that reads as a real person noticing real fit, which is what gets replies. Banning hype and gendered/cliché language ('rockstar') keeps it professional and inclusive. Turn winning versions into posts with the LinkedIn Post Generator.

When to use: turning a hiring manager's brain-dump into a clear, inclusive JD.

``` Draft a job description for [ROLE] from these notes: [PASTE — team, responsibilities, must-haves, nice-to-haves, level, location/remote]. Structure: short role summary, what you'll do, what we're looking for (separate REQUIRED from PREFERRED — keep required list short and truly essential), and how to apply. Use inclusive, plain language. Flag any requirement that may unnecessarily narrow the pool (e.g. arbitrary years, degree if not essential) so I can reconsider. No gendered or exclusionary wording. Don't invent details I didn't give you. ```

Why it works: separating truly-required from preferred — and flagging arbitrary requirements like unnecessary degree or years bars — directly widens and diversifies the applicant pool, since overstuffed 'required' lists deter qualified people. 'Don't invent details' keeps the JD accurate to the actual role. Use the Business Email Generator to adapt the apply instructions.

When to use: building a consistent, role-related interview guide every candidate gets — the fair and defensible way.

``` Build a structured interview kit for [ROLE]. The point is consistency: every candidate gets the same core questions. From these must-have competencies [LIST], generate: 2-3 behavioral questions per competency ("tell me about a time..."), what a strong vs. weak answer looks like for each, and 1-2 role-related follow-ups. Keep every question job-related and legally appropriate — no questions touching age, family, health, origin, religion, or other protected areas. Flag anything that could stray into protected territory. ```

Why it works: structured, identical, behavioral questions with defined strong/weak answers are both fairer to candidates and far more defensible than freewheeling chats. The explicit 'no protected-area questions, flag anything that strays' instruction is a real compliance safeguard — interviewers wander into illegal questions by accident, and this catches it up front.

When to use: turning your own interview notes into a structured summary — against the same rubric for every candidate.

``` Turn my interview notes into a structured summary against this role's competencies. Use ONLY what's in my notes — do not infer or add. For each competency [LIST], note the evidence from my notes (quote where possible) and whether it was strong / mixed / not demonstrated. List open questions for a next round. Do NOT score, rank, recommend, or decide — just organize the evidence I gathered. Keep it factual and role-related; ignore anything not relevant to doing the job. Notes: [PASTE — role-relevant only, no protected characteristics] ```

Why it works: 'use ONLY my notes, do not infer' keeps the summary tied to evidence you actually collected, and 'do not score, rank, recommend, or decide' is the hard line that keeps the human decision human. Organizing every candidate against the same competencies makes comparisons fair and consistent rather than gut-feel.

When to use: constructing search strings to find candidates on a job board or search tool.

``` Help me build search strings to source candidates for [ROLE]. Core skills/titles: [LIST]. Synonyms and adjacent titles matter — include variations people actually use. Location/remote: [...]. Seniority: [...]. Give me: a Boolean string, 2-3 looser variations to widen the net, and a list of adjacent titles/skills I might be missing. Focus on role-related skills and experience only — no proxies for age, origin, or other protected characteristics. ```

Why it works: surfacing synonyms and adjacent titles is where most sourcing leaves candidates on the table — people describe the same job a dozen ways. The 'no proxies for protected characteristics' rule keeps the search role-related and avoids building bias into the funnel from the first step.

When to use: drafting role-related screening questions for an application — humans review the answers.

``` Draft [5-7] screening questions for [ROLE] applicants. Goal: surface genuine role-related qualifications and must-haves (e.g. specific skills, eligibility to work, availability) so a human reviewer can assess fit. Keep them clear, answerable briefly, and strictly job-related. No questions about protected characteristics or proxies for them. For each question, note what a qualifying answer generally looks like — as a guide for the human reviewer, NOT an auto-screen rule. ```

Why it works: 'a guide for the human reviewer, not an auto-screen rule' is the key framing — screening questions inform a person's judgment, they don't automatically reject anyone, which keeps you clear of automated-decision rules. Keeping them strictly job-related protects both fairness and compliance.

When to use: the steady stream of scheduling, follow-up, and status emails that eat a recruiter's day.

``` Draft short, warm, professional email templates for [pick: scheduling an interview / post-interview thank-you-and-next-steps / requesting more info / keeping a strong candidate warm]. Use placeholders, no real data. Keep each under 90 words, clear about next steps and timing, and respectful — candidates remember how they were treated. Include a placeholder for a personal, specific detail so it doesn't read as a mass template. ```

Why it works: a placeholder for one specific personal detail is the small thing that keeps templated emails from feeling like spam — candidate experience is part of your employer brand. Being explicit about next steps and timing is the single most-requested thing candidates want and the most common recruiting failure. Build these out with Customer Email Templates.

When to use: declining candidates respectfully — the messages that protect your brand and reputation.

``` Draft a respectful rejection email for a candidate who [reached final round / applied but wasn't a fit]. No specifics about the decision rationale beyond what's appropriate and consistent. Tone: gracious, human, and kind without being falsely warm. Thank them specifically for their time, keep it brief, and leave the door open if genuine. Don't over-explain or give individualized feedback that could be inconsistent or legally risky — keep it consistent across candidates. ```

Why it works: 'keep it consistent across candidates, don't give inconsistent or legally risky feedback' protects you from disparate-treatment problems while still being humane. A gracious, brief rejection is what candidates remember and what keeps your talent pipeline and reputation intact.

When to use: organizing how a candidate's stated experience maps to the must-haves — to inform, not decide.

``` Map this candidate's role-related experience (from what I paste) against the role's requirements. Use ONLY what's provided; don't infer or assume. For each requirement: is there evidence it's met (quote it), partial evidence, or no evidence in what I have? List what I'd need to confirm in an interview. Do NOT score, rank, recommend for/against, or decide — surface the evidence and gaps so I can ask better questions. Keep it strictly role-related; ignore anything about protected characteristics. ```

Why it works: turning a resume-vs-requirements match into 'evidence, partial, or none — and what to confirm' makes it a tool for better interview questions, not a sorting machine. 'Do not score, rank, or recommend' keeps the model on the right side of automated-decision rules and keeps you doing the judging.

When to use: a second pass on a job description or step in your process to catch exclusionary language.

``` Review the text below for language or requirements that could discourage qualified candidates or introduce bias. Flag: gendered or coded wording, arbitrary or proxy requirements (years, degree, 'culture fit' as a vague catch-all), anything that may screen out groups without job relevance, and overly long 'required' lists. For each flag, explain why and suggest an inclusive, role-related rewrite. This is a draft review to inform my judgment — I decide what to change. Text: [PASTE — JD, screening criteria, etc.] ```

Why it works: an explicit bias pass catches coded language and arbitrary requirements that quietly shrink and skew your pool — the things you stop seeing in your own writing. Framing it as 'informs my judgment, I decide' keeps you in control while still getting a useful second set of eyes. Note AI can also miss or introduce bias, so a human still owns the final call.

For routine drafting — outreach, JDs, follow-ups, rejection notes — an efficiency tier is plenty: gpt-5.4-mini ($0.75 in / $4.50 out per 1M) or Gemini 3.1 Flash-Lite ($0.25 / $1.50). For nuanced interview kits, candidate summaries, and bias audits, a frontier model — Claude Opus 4.8 ($5 / $25), gpt-5.5 ($5 / $30), or Gemini 3.1 Pro (~$2.00 / $12.00) — reasons more carefully. Prices as of June 2026; check the live rate cards (OpenAI, Anthropic, Gemini).

The non-negotiables: never let AI decide — it drafts and organizes, a human makes every screening, ranking, rejection, and hiring decision. Keep everything role-related and consistent across candidates. Watch for bias, since AI can reflect and amplify it. Don't paste sensitive personal or protected data into a general AI tool. And follow your jurisdiction's employment laws, including any regulating AI and automated decision-making in hiring — consult counsel, not a chatbot, on compliance.

Sources and further reading: Learn Prompting, the DAIR.ai Prompt Engineering Guide, and Claude's prompt engineering overview. For self-reported compensation context, Levels.fyi aggregates self-reported data — treat any figure as directional. Pricing current as of June 2026.