Most prompt lists for recruiters are the same 20 generic lines, copied from one blog to the next. Write a job description. Personalize an outreach email. Summarize a resume. They work once, then you close the tab and go back to doing it by hand, because a prompt you have to rewrite every time is slower than just writing the thing.
The prompts below are built to be reused. Each one is written for a specific stage of a real search, keeps you on the right side of US hiring law, and returns something you can paste straight into your ATS or your sequencer. There are fifteen of them, grouped by where they fit in the workflow. At the end is the part the generic lists never mention: the one job no prompt can do for you, no matter how well you write it.
What makes a recruiting prompt actually work
A weak prompt asks for a thing. A strong prompt gives the model four things: a role to play, the context it needs, the exact output format you want, and a constraint that keeps it honest.
Compare "write a candidate outreach email" with "you are a technical recruiter; here is the JD and the candidate's recent work; write a first-touch message under 90 words that opens with a specific detail from their work, no flattery, and ends with one concrete next step." The first returns a generic template. The second returns something you can send. The difference is not the model. It is the instruction.
Hold that pattern in your head as role, context, format, constraint, and every prompt below will make sense. You can also rewrite any of them for your own desk by keeping the four parts and swapping the specifics.
Intake and the candidate brief
The search goes wrong before it starts when the brief is vague. These two prompts turn a messy hiring-manager conversation into something you can source against.
1. Intake notes to a sourcing brief.
You are a senior recruiter. Read my attached intake-call notes and the JD. Produce a sourcing brief with: must-have skills (max 5), nice-to-haves (max 5), deal-breakers, ideal current title and company type, expected years of experience, and a likely comp band based on title and location. Then list the three questions still worth asking the hiring manager before I start. Flag anything in the notes that is too vague to source against.
2. The JD reality check.
Read this JD as if you were a candidate deciding whether to apply, then as a recruiter deciding whom to contact. Tell me which requirements are real filters and which are wish-list padding, which single requirement probably drives the hiring decision, and what the JD is missing that a strong candidate would want to know. Keep it under 200 words.
The second prompt is the one hiring managers never expect. A public JD usually lists skills nobody cares about and buries the one that decides the hire. Catching that on Monday saves a week of sourcing the wrong profile.
Search and Boolean
Once the brief is tight, translate it into strings you can run. These prompts assume you already know how Boolean search works; they just save the typing.
3. Boolean strings per platform.
From the attached brief, write three Boolean search strings: one for LinkedIn, one for GitHub, and one general web string. For each, use synonyms and adjacent titles for the core role, group them with parentheses and OR, and exclude the two or three terms most likely to return false positives. Explain in one line what each string is tuned to catch.
4. Plain-English intent to search terms.
I am looking for someone who has done this kind of work: [describe the person in plain language, the way you would to a colleague]. Turn that into the skills, titles, and signals I should search for, including the ones a keyword search would miss because candidates describe the work differently on their profiles than we do in the JD.
Prompt 4 exposes the weakness of keyword search directly. The words a candidate uses for their own work rarely match the words in your JD, which is the whole reason keyword matching misses good people.
Screening and shortlist review
Read this section carefully, because this is where AI use crosses into regulated territory. Under NYC Local Law 144, Illinois HB 3773, and EEOC guidance on Title VII, a tool that scores or ranks applicants to substantially drive a hiring decision can be an automated employment decision tool with audit and disclosure duties attached. The prompts here are written to keep the model producing notes and questions for you, never a decision. You read, you decide. None of this is legal advice; if you process candidate data at scale, check your setup with a privacy or employment lawyer.
5. Resume against the role rubric (notes, not a verdict).
Compare this resume against the attached role rubric. Output: where the candidate clearly matches, where the evidence is thin or missing, and three specific questions I should ask on a screening call to close the gaps. Do not give a hire or no-hire recommendation, do not score or rank the candidate, and do not consider age, gender, race, national origin, religion, or any protected characteristic. If a requirement can't be judged from the resume, say so.
6. Pool re-engagement shortlist.
Read the attached candidate-pool CSV and this new JD. Surface the candidates worth a re-engagement call for this role. For each, give: name, last contact date, two specific reasons the role might fit them now, one reason it might not, and an opening line that references something from their notes. Return a CSV. Do not surface anyone based on a protected characteristic. I will decide who to contact.
Prompt 6 has the cleanest compliance profile of anything in this article, because you are choosing which of your own past relationships to call about a new opening, not scoring a live applicant funnel. It is also the single highest-value prompt here for a busy desk. Nobody re-reads 800 old candidate notes by hand for every new req, so those people never get the call. This does it in the time it takes to read the output.
Outreach
Generic outreach is why recruiters get ignored. The fix is not a better template, it is a real hook per person, with the model handling only the wording around it.
7. First-touch message.
You are a recruiter reaching out cold. Here is the role in two lines and one specific detail about this candidate's recent work. Write a first message under 90 words that opens on that detail, matches their apparent seniority, avoids "I came across your profile," and ends with one concrete next step (a 20-minute call by a named day, not "let me know"). No flattery.
8. A three-touch follow-up sequence.
Write a three-message follow-up sequence for a candidate who opened my first message but didn't reply. Space the value across the three: message one adds a detail about the team or the problem, message two shares something useful whether or not they interview, message three is a short and graceful close. Keep each under 70 words and change the angle each time. No guilt, no "just bumping this."
9. Subject lines worth testing.
Give me eight subject lines for this outreach email. Half short and plain, half specific to the candidate's work. No clickbait, no emojis, nothing that reads as automated. Rank them by how likely a busy senior candidate is to open, and tell me which one you would send first and why.
The honest caveat on all three: batch outreach is only as good as the hook you feed it. A polished generic message lands worse than a short honest one. Spend your time finding the real reason you are reaching out to this specific person. Let the model handle the phrasing, not the substance.
Interviews
10. A role-specific question set.
From the attached brief, write an interview guide for a 45-minute screening call: three questions on the must-have skills, two behavioral questions tied to the deal-breakers, one question that surfaces how the candidate actually thinks about the core problem of the role, and a note on what a strong versus weak answer sounds like for each. No brain-teasers.
11. Transcript to structured notes.
Read this interview transcript. Output: candidate strengths with the timestamp where each showed up, concerns with timestamps, points worth following up in the next round, and three good follow-up questions. Do not produce a hiring recommendation. These are notes for me and the hiring manager, who will make the call.
The timestamps matter. They let anyone check a claim against the actual moment in the interview instead of arguing from memory, which is where interview debriefs usually go wrong.
Market context and messaging
12. Comp and market framing.
I am about to open a search for [role] in [city]. Based on what you know, outline the questions I need real market data to answer before I talk to the hiring manager about comp and timeline: what the salary range likely is, how deep the talent pool is, and how competitive the hire will be. Do not invent specific numbers. Tell me what to go verify.
Prompt 12 is deliberately written to make the model admit what it doesn't know. A language model has no live read on this month's salaries or how many people in your city actually fit the role. It will happily invent a number if you let it, so the prompt forces it to hand you a research list instead. Where that real data comes from is the subject of the last section.
13. Rejection and nurture note.
Write a short, specific rejection message for a candidate who reached the final round but didn't get the offer. Name one real strength you saw, be honest that it was a close call if it was, and leave a genuine door open for future roles. Under 100 words, warm but not falsely encouraging.
Reporting and admin
14. The Monday pipeline summary.
Read the attached pipeline CSV. Summarize for a hiring-manager update: how many candidates are at each stage, which roles are healthy (five or more active candidates with contact in the last seven days) and which are stalling, and the two things I should flag as risks this week. Write it as a short email I can send, not a table.
15. Standardize messy candidate notes.
Here are my raw notes on ten candidates in whatever format I dumped them. Rewrite each into the same structure: current role, relevant experience, status, next action, one open question. Keep every fact I wrote, add nothing, and flag any note too thin to be useful so I know to follow up.
Prompt 15 is the quiet time-saver. The reason candidate notes stop being useful is that every recruiter records them differently on a busy day, and a search you pick back up in three weeks reads like a stranger's. This makes them consistent without inventing anything.
The one thing no prompt can do
Read back over the fifteen. Every one operates on information you already have: your notes, your JD, your pool, your transcript. That is the boundary. A language model reads what you hand it and reasons about it well. It cannot see the part of the market that isn't in the room.
It does not know who is quietly open to moving right now. It has no current read on what a Senior Platform Engineer in Denver actually gets paid this quarter. It cannot tell you that the candidate you're about to spend an hour sourcing is already three weeks into another company's process. That information exists, but it lives in live market data, not in the model's training. Prompt 12 was built to make that gap obvious on purpose.
This is the division of labor worth setting up. The language model does the reading, writing, and structuring: brief, search strings, outreach, notes. Glozo does the part the model can't reach. Glozo runs on 10M+ market signals a month and gives you three reads on a candidate that no prompt can produce: a matching summary built from the Skill Graph rather than keyword overlap, a Market Compensation Estimate that puts a live salary range on each person, and an "Open to Offers" signal that surfaces the people likely to be receptive before you spend a credit.
The clean setup is not exporting candidates out of Glozo to feed your prompts. It is the reverse. You use the prompts to prep the search, then run the actual sourcing where the live data is, and bring the results back to the model for the language work. If you want the deeper version of this, we wrote up how an AI sourcing agent uses that data and how to tell a real one from a chatbot with a search box.
Which tool to run these in
The prompts are written to work in any capable assistant. Which one you reach for depends on the job. For batch work on files, like the pool re-engagement in prompt 6 or the transcript in prompt 11, Claude handles folders and long documents best. For quick one-off drafting and rewrites, ChatGPT or Gemini, the one already sitting in your Gmail and Sheets, are fine. For the market-research framing in prompt 12, Perplexity leans research-first and cites its sources, though you still need live data for the numbers.
One setup habit applies to all of them: turn off the model-training toggle in your account settings before you paste a single resume, and give candidates a privacy notice that names AI-assisted processing. That, plus a retention policy, is what CCPA and the 2023-to-2026 wave of US state privacy laws actually expect. A fuller checklist lives in our Claude for recruiters guide, and it applies whichever tool you use.