RecruitingHuman Resources

AI Prompts for Resume Screening Rubrics

Resume screening is the highest-volume decision in hiring and the one most exposed to bias and inconsistency. A rubric fixes what is being looked for, what evidence satisfies each criterion, and how much each counts — before any resume is opened. Calibration makes two screeners score the same resume the same way. Recording the evidence makes every advance or pass explainable.

This page keeps the original rubric prompt and adds calibration and the structured screen. The guardrail below applies to everything here: the rubric organizes evidence; people make the decision, and nothing in it may score on protected characteristics or their proxies.

Before you use these

Have these ready to replace the highlighted [variables]:

Must-haves and nice-to-haves from intake, with evidence standards
The volume of applications and who screens
Known screener tendencies (what gets over-weighted)
Legal and policy requirements for screening in your jurisdiction

These prompts structure evidence against stated, job-related criteria and draft documents. They do not make hiring decisions. Do not ask the model to rank, reject or recommend candidates on protected characteristics or their proxies (name, address, school, gaps, age indicators), and do not treat its output as a decision: a person accountable for the hire reviews every score and makes the call. Check local law before using any automated or AI-assisted screening.

The prompts

1. Build the resume screening rubric
2. Calibrate screeners on the rubric
3. Structure the screen so evidence is recorded

1. Build the resume screening rubric

Best forA weighted, evidence-anchored rubric with automatic advances and passes.

Inputs needed

Role
Must-haves

How to use itThe original prompt from this page. Feed it the must-haves from intake so the criteria match the JD.

Expected outputSix weighted criteria with strong/acceptable/weak evidence anchors, automatic advances and passes, and five common over-indexes to avoid.

Act as a recruiting operations lead. Build a resume screening rubric for [role]: 6 criteria with weights, what 'strong/acceptable/weak' evidence looks like for each, automatic advances and automatic passes, and 5 things screeners commonly over-index on that don't predict performance (pedigree, gaps, buzzwords). Output as a usable scoring table.

2. Calibrate screeners on the rubric

Best forMaking scores consistent across people before real applications are screened.

Inputs needed

Rubric
Three to five sample resumes (anonymized)
Screeners' scores

How to use itHave each screener score the samples independently, paste the scores, and the model finds the disagreements, traces them to ambiguous anchors, and proposes tighter anchors.

Expected outputDisagreement analysis by criterion, the anchor wording causing it, revised anchors, and the calibration protocol to repeat.

You are calibrating resume screeners on a rubric for [role].

Rubric: [criteria, weights, anchors]
Sample resumes (anonymized): [summaries or key facts per sample]
Scores: [screener × sample × criterion]

1. Agreement analysis: for each criterion, the spread between screeners on each sample; the criteria with the most disagreement.
2. Trace each disagreement to its cause: an ambiguous anchor, a criterion that overlaps another, evidence that different screeners weighed differently, or an over-index (pedigree, gaps, formatting).
3. Revised anchors: rewrite the ambiguous ones with concrete, observable evidence.
4. Rules to add: what counts as evidence for each criterion, what is ignored (photos, names, addresses, dates that indicate age, hobbies), and how to handle unusual backgrounds against the equivalents list.
5. Calibration protocol: how often to repeat, the sample size, the agreement threshold to reach before screening live.
6. Note any criterion that appears to correlate with a protected characteristic or its proxy; recommend removing or redefining it.

Report on the rubric and the process, not on the screeners as individuals.

3. Structure the screen so evidence is recorded

Best forA screening record where every advance or pass points to the rubric evidence, and the decision stays with the screener.

Inputs needed

Rubric
Application volume
Tools

How to use itThe model designs the screening record, the sequence (knockouts first, then scored criteria), the review of borderline cases, and the audit sample.

Expected outputScreening record template, the sequence and time budget per application, borderline review rule, the audit sample and what is checked, and the reporting that shows the funnel by stage without demographic scoring.

Act as a recruiting operations lead designing the screening process for [role] with [volume] applications and [n] screeners.

Rubric: [criteria, anchors, weights, knockouts]
Tools: [ATS, forms]
Policy: [legal requirements for screening records in our jurisdiction — to verify]

1. Screening record: per application, the evidence quoted for each criterion, the score, the knockout result, the decision (advance / hold / pass) and the screener's initials — designed so a third party can see why.
2. Sequence: must-have knockouts first (job-related, consistently applied), then scored criteria; time budget per application; the rule for stopping early.
3. Borderline rule: score bands that trigger a second screener, and how disagreements are resolved.
4. Consistency safeguards: masked fields where possible (name, photo, address, dates that indicate age), the order in which applications are screened, breaks to reduce fatigue effects.
5. Audit: a sample re-screened by a lead, what is checked (evidence quality, anchor use, consistency), and how findings feed back into the rubric.
6. Reporting: funnel by stage and source; no reporting or scoring on protected characteristics; where equal-opportunity monitoring exists, it stays separate from screening decisions.

The record supports the decision; it does not make it. Say so in the template header.

What an evidence-anchored criterion looks like

Two criteria from a rubric for a customer-operations analyst, written the way the calibration prompt tightens them. Note what is not on the list.

Criterion (weight)	Strong evidence	Acceptable evidence	Weak / absent	Not evidence
Analyzes operational data to a decision (30%)	Describes a specific analysis they ran, the decision it informed and the result; names the tools	Mentions data analysis as part of a role with some detail on method	Generic 'data-driven' claims with no example	Employer name, degree subject, years in role
Works across functions to fix a process (20%)	A named process change they led with two or more teams and a measured outcome	Participation in a cross-functional project with a stated role	'Team player', 'collaborative'	Titles, gaps in employment, seniority of previous employer

Every anchor points at something a candidate did, not at where they did it. The last column is what screeners most often score by mistake.

Related prompts

Logical next step

After this, most HR teams move on to Role Intake and Success Profiles.

Get the free HR & Recruiting AI Starter Kit → Nine of these prompts as an intake → assess → offer hiring workflow with an intake worksheet, delivered by email. See what's inside

All Human Resources prompts · Search the full library