AiOS→HR / People→Attract & Hire→HR03
Screening & Candidate Shortlisting
Extract evidence from candidate materials, generate summary cards, and produce a criteria-linked shortlist with risk flags.
Separating evidence extraction from recommendation reduces screening bias and gives hiring managers a defensible, criteria-linked shortlist instead of subjective impressions.
GenAI Impact
56%
Faster
7.4
Hours saved
13.1
Hours without AI
Based on: 20 candidates evaluated against 5 must-have criteria
Enforces identical 5-criterion evaluation with verbatim evidence anchoring for every candidate, preventing the scoring drift that degrades manual screening after 10–15 resumes.
Prevents candidate PII from reaching unapproved AI tools and catches fabricated qualifications through a mandatory human cross-reference gate before the shortlist reaches the hiring manager.
Before You Start
This workflow processes candidate personally identifiable information (names, contact details, employment history). Do not paste candidate data into public or unapproved GenAI tools.
GenAI may fabricate qualifications or misattribute experience across candidates. Verify every evidence citation against the original application before sharing the shortlist.
Who's Involved
Recruiter
Screens candidates, verifies extracted evidence, and manages the shortlisting workflow.
Hiring Manager
Validates the final shortlist against role requirements and approves candidates for next steps.
Execution Steps
Before you start
Data Handling: Do not include candidate contact details, salary expectations, or identification numbers in the prompt — use candidate identifiers only.
Inputs
Prompt
Extract criterion-linked evidence from candidate materials
CONTEXT You will be provided with the following source documents: 1. Must-Have Criteria 2. Candidate Resumes 3. Screening Notes TASK For each candidate, extract specific, verbatim quotes or concrete facts from their application that relate to each must-have criterion. Produce an Evidence Extraction Table mapping every candidate to every criterion. OUTPUT FORMAT Use a markdown table with the following columns: - **Candidate** — candidate identifier - **Criterion** — the must-have criterion being assessed - **Evidence** — verbatim quote or specific fact from the application - **Source** — which document the evidence comes from (resume, cover letter, screening notes) - **Strength** — [Strong / Partial / No Evidence] Include one row per candidate-criterion pair. If no evidence exists for a criterion, enter "No evidence found" in the Evidence column and "No Evidence" in the Strength column. CONSTRAINTS Do not infer or assume qualifications not explicitly stated in the source materials. Do not paraphrase — use verbatim quotes where possible. Do not include personally identifiable contact details in the output.
Outputs
Verification: Verify that every evidence citation maps to an actual passage in the candidate’s application — GenAI may fabricate quotes.
Inputs
Prompt
Outputs
Inputs
Prompt
Produce evidence-backed candidate summary cards
CONTEXT You will be provided with a Verified Evidence Table that maps each candidate’s application evidence to the must-have criteria for the role. TASK For each candidate, generate a summary card that consolidates the evidence into a concise profile. Each card should state the candidate’s overall strength against the criteria, highlight the strongest evidence, and note any criteria with weak or missing evidence. OUTPUT FORMAT For each candidate, use this structure: ### [Candidate Identifier] - **Overall Fit**: [Strong Fit / Moderate Fit / Weak Fit] - **Strongest Evidence**: 2–3 bullet points citing the most compelling criterion-evidence pairs - **Gaps or Weak Areas**: 1–2 bullet points noting criteria with Partial or No Evidence ratings - **Screening Note**: One sentence summarizing the recruiter’s overall impression EXAMPLE ### Candidate A - **Overall Fit**: Strong Fit - **Strongest Evidence**: - Technical Skills: "Led migration of three legacy systems to cloud infrastructure" (Resume) - Experience: "8 years in enterprise platform engineering" (Resume) - **Gaps or Weak Areas**: - Core Competencies: No evidence of stakeholder management experience - **Screening Note**: Strong technical profile with a gap in stakeholder-facing experience. CONSTRAINTS Do not introduce qualifications or evidence not present in the Verified Evidence Table. Do not rank or recommend candidates — summary cards are descriptive only.
Outputs
Verification: Verify that overall fit ratings are consistent with the actual evidence counts — GenAI may over-rate candidates with sparse evidence.
Inputs
Prompt
Outputs
Inputs
Prompt
Outputs
Verification: Verify that bucket placements are consistent with the evidence counts — GenAI may assign Advance to candidates with significant gaps.
Inputs
Prompt
Outputs
Reference
Guardrails
- Evidence-Based Scoring Only — Every rating and bucket placement must cite specific evidence from the candidate’s application — no inferred or assumed qualifications.
- Separate Extraction From Decision — Complete evidence extraction and verification before generating summary cards or shortlist rankings to reduce confirmation bias.
- Consistent Criteria Application — Apply the same must-have criteria to every candidate — do not adjust thresholds mid-screening.
Pitfalls
- Pasting unredacted candidate contact details or salary expectations into the GenAI prompt.
- Accepting AI-extracted evidence without verifying quotes against the original application documents.
- Allowing the AI to infer qualifications not explicitly stated in the candidate materials.
- Skipping the risk and gaps log review before sharing the shortlist with the hiring manager.
Definition of Done
- Every candidate summary card cites specific verbatim evidence from the source application.
- The shortlist table assigns every screened candidate to exactly one bucket with an evidence-backed rationale.
- The risk and gaps log identifies every Partial or No Evidence rating across all candidates and criteria.
Unlock the Full Library
Get full access to all prompts, execution steps, and downloadable examples — for this playbook and the rest of our GenAI capability framework — AGASI AiOS.
We'll send a magic link — no password needed.
AGASI AiOS · HR03 v1.0 · Apr 7, 2026