AI-Powered Applicant Screening

Recruiters spend an average of 7 seconds per resume during initial screening. With 200+ applications per position, this is neither thorough nor fair. AI-powered screening promises objectivity and efficiency — but also carries significant risks.

Resume Parsing

What AI Can Do

Modern resume parsers extract structured data from any format:

Contact data: Name, email, phone, LinkedIn
Work experience: Positions, companies, duration, description
Skills: Technical and soft skills, certifications
Education: Degrees, institutions, grades
Languages: Level and certificates

Accuracy 2026: Top parsers (Textkernel, HireAbility) achieve 95%+ for standardized resumes, 80–85% for unconventional formats.

Matching Algorithms

AI compares parsed profiles with the job posting:

Keyword matching: Simple but imprecise ("Java" ≠ "JavaScript")
Semantic matching: Understands meaning ("Projektleitung" ≈ "Project Management")
Skill gap analysis: Which skills are missing? Are they quickly learnable?
Culture fit scoring: Based on company culture keywords — very controversial

Ranking output: Top 20% of applications as "shortlist," middle 60% as "review," bottom 20% as "rejection suggestion."

Bias Risks — The Biggest Danger

Historical Bias

AI learns from historical data. If your company predominantly hired men in the past, the AI reproduces this bias.

Famous example: Amazon's AI recruiting tool (2018) systematically disadvantaged women — it was discontinued.

Proxy Variables

Even when gender/age/ethnicity are removed:

University name correlates with social background
Zip code correlates with ethnicity
Hobbies correlate with gender ("football" vs. "yoga")
Resume gaps disadvantage parental leave, illness, migration

Countermeasures

Bias audits: Regularly check whether certain groups are systematically disadvantaged
Diverse training data: Not just successful hires, but also successful employees
Transparent criteria: Which factors feed into the score?
Human-in-the-loop: AI recommends, recruiter decides — never fully automated
Regular re-evaluation: At least quarterly

Legal Framework (EU)

GDPR: Applicants have the right to explanation of automated decisions (Art. 22)
EU AI Act: Recruiting AI is classified as "high risk" — strict documentation requirements
AGG (Anti-Discrimination Act): No screening by age, gender, ethnicity, religion, disability

Recommendation: Use AI in screening as an assistant, not a decision-maker. The shortlist from AI, the final decision from humans. And: Document everything.