These 14 recruiting prompts help you write job descriptions, check a job ad for biased wording, write outreach people answer, build a structured interview with a scorecard, and send clear candidate emails. They keep the model on drafting and organizing, and leave every hiring decision with a person.
How to use these prompts
Each prompt works in ChatGPT, Claude or Gemini. Replace the blanks in double braces, or keep the default after the colon: {{role:senior backend engineer}} means "senior backend engineer" unless you change it.
Give the model the real job, not the job title. The prompts ask for what the person will actually do, the team they join and the two or three skills that separate a great hire from an average one. OpenAI's prompt guide for everyday work puts it simply: "If you can't describe 'good' in one or two sentences, the model can't reliably hit it either."
Keep personal data out. Paste your notes and the job requirements, not candidates' contact details, photos or anything about protected characteristics.
Which prompts write job descriptions?
1. Job description from the real work
Starts from outcomes, so the ad describes a job instead of a wish list.
Write a job description for a {{role:senior backend engineer}} at {{company:our company}}.
What this person will do in the first 6 months:
{{first_six_months}}
Team and who they report to: {{team}}
Location and work pattern: {{location:remote, within 3 hours of UTC}}
Salary range: {{salary_range}}
Structure: a two-sentence summary of the role, what you'll do (5 bullets,
outcomes not duties), what you'll need (no more than 6 bullets), nice to
have (up to 3), how we work, and the hiring process step by step. Plain
language, no buzzwords like "rockstar" or "ninja", under {{words:500}} words.
Only use facts from my notes; mark anything you had to assume with "Check:".
2. Bias and clarity check for a job ad
Run it on every ad before posting. The EEOC page on prohibited employment practices gives an example: a help-wanted ad that seeks "females" or "recent college graduates" may discourage men and people over 40 from applying and may violate the law.
Review this job ad for wording that could discourage qualified people from
applying or that shows a preference based on race, color, religion, sex,
national origin, age, disability or genetic information. Also flag:
requirements not needed to do the job (for example a degree or years of
experience that the work doesn't require), gendered or age-coded words,
jargon an outsider wouldn't understand, and claims we can't back up.
Return a table: quote, problem, suggested rewrite. Then the full revised ad.
This is a language review, not legal advice; list anything I should check
with {{reviewer:our HR team}}.
{{job_ad}}
3. Separate must-haves from nice-to-haves
Long requirement lists shrink the pool without improving it.
Here are the requirements we drafted for a {{role:senior backend engineer}}:
{{requirements}}
And here is the work the person will do:
{{first_six_months}}
For each requirement, tell me which task it is needed for. Sort them into:
must have on day one, can learn in the first 3 months, and not needed.
Suggest how to reword "years of experience" requirements as the skill or
result we actually need.
Which prompts help with sourcing and outreach?
4. Personal outreach message
A short note that shows you read the person's work gets more replies than a template.
Write a first outreach message to a potential candidate for a
{{role:senior backend engineer}} role.
What I know about them (from their public profile or work): {{profile_notes}}
Why this role might interest them: {{hook}}
The role in two lines: {{role_summary}}
Salary range: {{salary_range}}
Rules: under {{words:120}} words, mention one specific thing from their work,
state the salary range, make it easy to say no, and end with one simple
question. No flattery, no "I came across your profile". Give 2 versions.
5. Follow-up that doesn't nag
For the polite second message a week later.
I sent this outreach message {{days:7}} days ago and got no reply:
{{first_message}}
Write one short follow-up (under 70 words) that adds something new, such
as {{new_info:a link to our engineering blog post on the system they'd own}},
and says this is the last message unless they want to talk. No guilt, no
"just bumping this".
6. Boolean search string
Turns a role into a search you can paste into a sourcing tool or search engine.
Write Boolean search strings to find candidates for a
{{role:senior backend engineer}} who has worked with {{skills:Go, PostgreSQL and event-driven systems}}
in {{location:Europe}}.
Give 3 strings: narrow, medium and broad. Use AND, OR, NOT, quotes and
brackets, and include common synonyms and job title variants. Explain each
part in one line. Target {{platform:LinkedIn Recruiter}} syntax, and tell me
if an operator I used isn't supported there.
Which prompts build a structured interview?
The US Office of Personnel Management describes a structured interview as "an assessment method designed to measure job-related competencies of candidates by systematically inquiring about their behavior in past experiences and/or their proposed behavior in hypothetical situations." OPM notes that all candidates are asked the same predetermined questions in the same order, and all responses are evaluated using the same rating scale. The next four prompts build that kit.
7. Interview plan from the role
Decide what each stage measures before anyone books a call.
Design an interview process for a {{role:senior backend engineer}}.
The job: {{first_six_months}}
Must-have skills: {{must_haves}}
Number of stages we can afford: {{stages:4}}
Map each must-have skill to exactly one stage, so nothing is tested twice
and nothing is missed. For each stage give: what it measures, format,
length, who runs it, and what a strong answer looks like. Keep the total
candidate time under {{hours:4}} hours.
8. Questions with a rating scale
The scale is the part teams skip, and it is what makes scores comparable.
Write structured interview questions for the competency
"{{competency:handling production incidents}}" for a {{role:senior backend engineer}}.
Give 3 behavioral questions ("Tell me about a time...") and 2 situational
questions ("What would you do if..."). For each question: 2 follow-up
probes, and a 1 to 5 rating scale with a short description of what a 1, 3
and 5 answer sounds like. Every question must relate to the job. No
questions about personal life, health, family plans, age, religion or
background.
9. Work sample brief
A small, realistic task beats trivia. Respect the candidate's time.
Design a work sample exercise for a {{role:senior backend engineer}} that
reflects real work from this list: {{first_six_months}}
Constraints: it takes no more than {{hours:2}} hours, needs no proprietary
knowledge, and can be done with common tools. Give: the brief the candidate
receives, what we provide, what they hand in, the scoring criteria tied to
our must-have skills, and how we will discuss it in the follow-up interview.
10. Check interview questions before use
A second pass on questions written by several interviewers.
Here are the questions our interviewers plan to ask:
{{questions}}
Flag any question that is not clearly related to the job, asks about
disability or health before an offer, or touches on race, color, religion,
sex, national origin, age, family status or other personal matters. Suggest
a job-related rewrite for each. Also flag duplicate questions across
interviewers. This is a review, not legal advice; list anything to confirm
with {{reviewer:our HR team}}.
Which prompts help after the interview?
11. Notes to evidence against the scorecard
Sorts raw notes by competency without making the call for you.
Here are my notes from an interview and our scorecard competencies.
Competencies: {{competencies}}
Notes: {{notes}}
For each competency, list the evidence from my notes as direct quotes or
close paraphrases, and note where I have no evidence. Do not give a score,
a hire or no-hire recommendation, or any comment on the person beyond what
the notes say. Flag any note that isn't about the job so I can remove it.
12. Debrief agenda when interviewers disagree
Keeps the debrief on evidence instead of whoever talks first.
Here are the written scorecards from {{count:4}} interviewers for one
candidate:
{{scorecards}}
Prepare a 30-minute debrief agenda. Show where scores agree and where they
differ by 2 or more points, quote the evidence each interviewer gave, and
list the questions the panel should answer to resolve each disagreement.
Point out any score with no evidence behind it. Do not recommend a decision.
13. Kind, clear rejection email
People remember how a "no" felt.
Write a rejection email to a candidate who reached the
{{stage:final interview}} stage for the {{role:senior backend engineer}} role.
Tone: warm and direct. Thank them for their time, give the decision in the
first two sentences, and {{feedback:offer a short call for feedback if they want one}}.
Under 120 words. No phrases like "we had many strong candidates" unless I
say so, and no reasons I haven't given you.
Reasons I'm comfortable sharing: {{reasons:none}}
14. Offer summary email
A plain summary so the candidate isn't decoding a PDF on their own.
Write an email that accompanies a job offer for {{name}}, joining as
{{role:senior backend engineer}}. Use only these terms:
{{offer_terms}}
Summarize the offer in a short table (salary, equity, start date, location,
benefits highlights), explain anything that usually confuses people, say
who to contact with questions, and give the deadline: {{deadline}}. Friendly,
no pressure tactics. Note that the formal offer letter is the binding
document.
What should you never hand to a model in hiring?
The final decision. Anthropic's Usage Policy (the version effective 15 September 2025) lists "resume screening, hiring tools, or other employment determinations" among its High-Risk Use Cases. For those it requires a human in the loop: "a qualified professional in that field must review the content or decision prior to dissemination or finalization." The updated policy that takes effect on 12 November 2026 keeps a qualified human review requirement for screening, ranking and rejecting candidates.
That is why prompts 11 and 12 ask for evidence and questions, not scores or verdicts. The model organizes; people decide.
Questions about protected characteristics. The EEOC says "Employers are explicitly prohibited from making pre-offer inquiries about disability." It also says information about race, sex, national origin, age and religion is irrelevant when deciding whether a person is qualified. Prompts 2, 8 and 10 bake that into the instructions. US rules are a floor; check the law where you hire.
Facts you didn't give it. Models fill gaps with plausible details, such as a benefit you don't offer. Anthropic's guide to reducing hallucinations recommends giving the model explicit permission to admit uncertainty, which is why the prompts say "mark anything you had to assume".
Make your favorite one a skill
If you write a job ad or an interview kit every month, turn the prompt into a skill with your company's tone, process and rules built in, so Claude, Claude Code or Codex applies them every time. Paste it into the free SKILL.md generator. For engineering roles, pair these with the technical interview prompts.
To keep a hiring kit ready for the next opening, save these prompts in Kipt with their blanks, tag them by stage (ad, outreach, interview, close) and press Cmd+K to find one. Your library lives as Markdown in a private GitHub repo you own.