Automated CV screening
Quickly screen dozens to hundreds of CVs to identify most relevant profiles while staying AI Act and GDPR compliant.
CV screening is one of the most time-consuming tasks of recruitment: 100 CVs for one role easily means a day's work. AI lets you drop to 30-60 minutes for the same volume. But recruitment is classified 'high risk' by the EU AI Act since 2026: candidate transparency obligations, human supervision, traceability, bias audits. This guide presents the workflow that industrializes without degrading ethics or compliance.
Step-by-step workflow
Define explicit evaluation grid
Before any screening: list objective criteria (skills, years of experience, qualifications) and weights. Without explicit grid, AI reproduces biases and your unconscious preferences.
Anonymize CVs before processing
Remove name, photo, age, address (keep school level and specialty). Limits discriminatory biases and stays HR best practice compliant.
Submit for screening with explicit criteria
Ask AI to score each CV on defined criteria, with justification. Not opaque global score but per-criterion score for audit.
Audit results
Verify consistency: are rejected profiles legitimately rejected? Is there bias on certain variables? Keep audit trail.
Final human decision
AI screening produces a short list, but decision (who to call, who to reject) remains human. AI Act compliance: effective human supervision is mandatory for high-risk systems.
Copyable prompts
2 tested and optimized prompts. Adapt the bracketed variables [VARIABLE] to your context.
CV batch screening
You're a senior recruiter. Here are criteria for the role '[ROLE]': **Objective criteria**: [LIST WITH WEIGHT /20] - Required skills: [LIST] - Desired skills: [LIST] - Minimum experience: [DURATION] **CVs to evaluate** (anonymized): [CV 1] [CV 2] [...] For each CV, produce: 1. **Score per criterion** /20 with short justification 2. **Global score** /100 3. **Top 3 strengths** 4. **Top 3 gaps** or points to dig in interview 5. **Recommendation**: call / pile B / reject Stay factual, anchored in criteria, no interpretation of anonymized elements.
Bias detection in shortlist
Here is the shortlist from CV screening: [LIST WITH SCORES AND REASONS] Initial pool was [N] candidates with this distribution (aggregated info available): [GENDER / AGE / SCHOOL DISTRIBUTION IF ANALYZED] Audit shortlist for potential biases: 1. **Disparities** between initial pool and shortlist by variable 2. **Suspicious patterns** in screening criteria 3. **Questionable criteria**: is their link with future performance proven, or biased proxy? 4. **Recommendations** to correct or expand shortlist 5. **Documentation** to produce for AI Act traceability
Top tools for this use case
Curated selection of the 3 best AI tools for automated cv screening.
Why for this use case: Most precise for argued screening with per-criterion justifications. Limited hallucinations on factual CV elements.
Why for this use case: Good for processing volumes in parallel (API), with Code Interpreter for shortlist statistical analyses.
Why for this use case: For the interview phase: automatic capture, transcription, and synthesis. Essential for sharing with manager.
Estimated ROI
Time saved
70-80% on screening (1h vs 4-6h for 100 CVs)
Quality gain
Objectified and traced criteria, possible bias audits
Stack cost
$30-100/month for compliant enterprise solutions
Estimates based on 2026 benchmarks and user feedback. Actual ROI depends on your context.
Frequently asked questions
Is AI screening AI Act compliant?
Conditionally. Recruitment is classified high risk since 2026: document evaluation grid, guarantee effective human supervision, inform candidates of AI use, regularly audit biases. With these conditions: yes. Without: sanction risk.
Can a candidate be rejected solely on AI decision?
No. AI Act requires effective human supervision for impactful decisions. AI screening produces a recommendation, human decides. Any automatic rejection decision without human intervention is non-compliant.
Should candidates be told AI is used?
Yes, since AI Act 2026: clear information, right to request human review, possibility to ask which criteria were used. Integrate in application policies and recruitment communication.
What biases can AI reproduce?
All those of training data and your company's historical decisions: gender, origin, age, school, linear vs atypical career. Regular audit essential. Solutions: ultra-explicit criteria, upstream anonymization, post-screening sampling validation.