Claude prompts for recruiters: 10 copy-ready prompts built for long documents

The Claude prompts that work best for recruiters put the source material first, in labeled XML tags, state the task and format last, and ask Claude to quote the source before it concludes, which is the structure Anthropic's own prompting guide recommends for long documents. Below are 10 copy-ready prompts for the document-heavy parts of recruiting: intake notes, resume checks, interview transcripts, scorecards, submittals, panel debriefs and candidate emails, plus how to set up a Claude Project for each requisition and what candidate data belongs in which plan, as of October 2026.

These are not a translation of our ChatGPT prompts for recruiters, which cover intake, outreach and job ads. They lean on what Claude's guidance is built around: long inputs and answers grounded in quotes. Prompts for writing interview questions are in AI prompts for interview questions.

How should you structure a prompt for Claude?

Anthropic's prompting best practices give five rules that map directly onto recruiting work:

• Be clear and direct. The guide suggests thinking of Claude "as a brilliant but new employee who lacks context on your norms and workflows." Say exactly what you want and in what format.
• Explain why. Context such as "this goes to a client who reads on a phone" gets better output than a bare rule.
• Use XML tags. Wrapping each input in its own tag, such as <transcript> or <job_description>, "reduces misinterpretation" when a prompt mixes instructions and material.
• Long documents first. Put long inputs "near the top of your prompt, above your query." Anthropic notes that queries at the end can improve response quality by up to 30 percent in its tests.
• Ground in quotes. For long documents, ask Claude to "quote relevant parts of the documents first before carrying out its task." For recruiting this is the most important rule, because a quote can be checked against the transcript.

How do you set up a Claude Project for a requisition?

A Project is a workspace whose knowledge files and instructions apply to every chat inside it. Per Anthropic's help article, Projects are available on all plans, free accounts can create up to five, and Team and Enterprise users can share a Project with colleagues as view or edit. One Project per requisition works well: upload the job description, intake notes and scorecard rubric as project knowledge, then paste these as project instructions.

1. Project instructions for a requisition

You are helping [agency / company] recruit a [JOB TITLE] for
[client or team]. The project knowledge contains the job description,
the intake notes and the scorecard rubric.

Rules for every answer in this project:
- Use only the documents provided and anything I paste into the chat.
  If something is not in them, write [NOT IN SOURCE].
- When you state what a candidate said, quote them exactly and say
  which document the quote is from.
- Assess only against the rubric's criteria. Do not comment on age,
  health, family, nationality, accent, appearance or any other
  personal characteristic, even if it appears in a document.
- Do not recommend hiring or rejecting anyone. Lay out evidence;
  I make decisions.
- Write in plain American English, short paragraphs, no hype.

What candidate data can you put into Claude?

Anthropic's August 28, 2025 update to its consumer terms applies to Claude Free, Pro and Max: each user chooses in Privacy Settings whether their data is used to improve Claude, and retention is extended to five years for those who allow it, against 30 days for those who do not. The update does not apply to services under Anthropic's commercial terms, including Claude for Work (Team and Enterprise) and the API.

A training setting is not permission. Whether a recruiter may paste a resume or transcript into any AI tool depends on the organization's policy, client agreements, what candidates were told and privacy law. The safe default: candidate documents only in a plan your organization has approved for them, with contact details removed where the task does not need them, and nothing about health, family, immigration detail or references' personal information. The policy behind that is in how to use AI in recruiting responsibly.

Which prompts help before the interview?

2. Turn intake call notes into a role brief

<job_description>
[PASTE]
</job_description>
<intake_notes>
[PASTE YOUR NOTES OR TRANSCRIPT FROM THE INTAKE CALL]
</intake_notes>

Write a one-page role brief for recruiters and interviewers. Sections:
the problem this hire solves; must-haves (max 5); nice-to-haves;
deal-breakers; what success looks like at 6 months; process and
interviewers; pay range and logistics.
For each must-have, quote the line from the intake notes it comes from.
Where the job description and the intake notes disagree, list the
conflict under "Open questions for the hiring manager" instead of
choosing one.

Good output separates what the hiring manager said from what the old job description says. The open questions list is often the most valuable part.

3. Check a resume against the must-haves, with quotes

<must_haves>
[PASTE THE MUST-HAVES FROM THE ROLE BRIEF]
</must_haves>
<resume>
[PASTE RESUME, CONTACT DETAILS REMOVED]
</resume>

First, for each must-have, copy the exact resume lines that relate to it
into <quotes> tags. Then make a table: must-have | evidence quoted |
what is unclear | a phone screen question to resolve it.
If there is no evidence, write NOT FOUND. Do not infer skills from job
titles or company names. Do not score or rank the candidate.

This turns a resume read into a screening plan. "NOT FOUND" is not a rejection; it is a question for the screen.

Which prompts help after the interview?

4. Draft a scorecard from a transcript

<rubric>
[PASTE THE SCORECARD CRITERIA AND RATING ANCHORS]
</rubric>
<transcript>
[PASTE THE INTERVIEW TRANSCRIPT]
</transcript>

For each criterion in the rubric:
1. Quote up to three passages where the candidate gives evidence,
   word for word, in <quotes> tags.
2. Summarize the evidence in two sentences.
3. Say which rating anchor the evidence best matches and why, or
   write INSUFFICIENT EVIDENCE.
Use the candidate's words, not the interviewer's questions, as
evidence. This is a draft for the interviewer to check and change.

The interviewer still owns the score. If your interview tool already produces a scorecard with quotes checked against the transcript, as Interview Signal (which we make) does, keep this prompt for calls captured elsewhere. For the wider method, see how to use AI to summarize interviews.

5. Check every quote in a draft against the transcript

<transcript>
[PASTE]
</transcript>
<draft>
[PASTE THE SCORECARD, NOTES OR SUBMITTAL TO CHECK]
</draft>

List every quotation and every factual claim about the candidate in the
draft (dates, employers, titles, numbers, pay, notice period). For each,
say VERIFIED with the matching transcript passage, PARAPHRASED if the
meaning matches but the words differ, or NOT FOUND. Do not rewrite the
draft.

Run this on anything a client or hiring committee will read, whoever or whatever wrote it. A separate checking pass catches errors that the drafting pass made.

6. Turn a screen into a client submittal

<role_brief>
[PASTE]
</role_brief>
<transcript>
[PASTE THE SCREEN TRANSCRIPT OR YOUR NOTES]
</transcript>
<template>
[PASTE YOUR AGENCY'S SUBMITTAL TEMPLATE]
</template>

Fill in the template for the client. Lead with how the candidate meets
the top three must-haves, each supported by one short quote. Include
motivation for the move, notice period and pay expectations only as
stated in the transcript. List any gaps honestly under "To explore at
interview." Under 300 words. Leave out anything about health, family or
other personal matters.

The structure behind a good submittal is in how to turn interview notes into a submittal.

Which prompts help with debriefs and decisions?

7. Combine panel feedback before the debrief

<feedback interviewer="[NAME 1]">
[PASTE]
</feedback>
<feedback interviewer="[NAME 2]">
[PASTE]
</feedback>
[ADD ONE BLOCK PER INTERVIEWER]

Produce a debrief sheet with three sections, by competency:
- Agreement: where interviewers' evidence points the same way.
- Disagreement: where ratings differ, with each interviewer's evidence
  quoted side by side.
- Gaps: competencies nobody gathered evidence on.
Do not average the ratings or recommend a decision.

8. Compare finalists criterion by criterion

<rubric>[PASTE]</rubric>
<candidate name="A">[PASTE SCORECARDS AND NOTES]</candidate>
<candidate name="B">[PASTE]</candidate>
<candidate name="C">[PASTE]</candidate>

Make a table with one row per rubric criterion and one column per
candidate. In each cell give the strongest piece of evidence, quoted,
and its source. Then list where the evidence is thin for each
candidate. Do not rank the candidates or name a preferred one.

Asking for evidence by criterion, rather than a ranking, keeps the decision with the hiring team. It also matters legally: an AI ranking acted on without real review can bring in automated decision rules such as California's ADMT regulations.

9. Write next-round questions from the gaps

<rubric>[PASTE]</rubric>
<scorecard>[PASTE THE ROUND ONE SCORECARD]</scorecard>

For each criterion marked INSUFFICIENT EVIDENCE or rated with low
confidence, write two behavioral questions for the next interviewer,
each asking for a specific past example, plus one follow-up probe.
Do not repeat questions already asked in round one.

10. Draft feedback to a candidate who was not selected

<scorecard>[PASTE]</scorecard>

Draft a short, kind email to the candidate saying they have not been
selected for [ROLE]. Include two specific strengths from the scorecard
and, only if I have marked it as shareable, one job-related area where
other candidates showed more evidence. Do not mention other candidates
by name, personal characteristics, or anything not in the scorecard.
Under 150 words. I will review and send it myself.

Which prompts should recruiters not use?

• "Who should I hire?" or "Rank these candidates." The decision stays with a person, with a job-related reason.
• "Is this candidate a culture fit?" It invites judgments about similarity rather than the job.
• Anything that infers age, health, ethnicity, family plans or personality from a resume, photo, name or transcript.
• "Find information about this candidate online." It can pull in personal social media and protected information you should not hold.
• Prompts that send candidate data to a personal account your organization has not approved.

How do you check Claude's output before using it?

1. Every quote: does it appear word for word in the source? Prompt 5 does this systematically.
2. Every fact: dates, employers, pay and notice period against the transcript or resume.
3. Every [NOT IN SOURCE] or NOT FOUND: resolve it with a question, never by guessing.
4. Tone and fairness: nothing about personal characteristics, nothing you would not say to the candidate's face.
5. Your judgment: you are the one signing the scorecard or sending the submittal.