Technical Explainer · 8 Min Read

AI resume screening isn't magic.

From keyword matching to semantic vectors, how modern AI scores candidates, where it breaks down, and what Keelzo does differently.

While most ATS platforms run keyword matching dressed up as AI, Keelzo uses embedding-based vector matching, semantic AI that understands that “people management” and “team leadership” mean the same thing, and scores candidates accordingly across experience duration, seniority, industry relevance, and skill recency.

What sets Keelzo apart is transparency. Recruiters and hiring managers don’t just see a score, they see the reasoning behind it. Which skills matched, which didn’t, and why. A score you can interrogate is a score you can act on.

Keelzo’s AI is placed deliberately, at the moments that change hiring outcomes, not scattered across the platform for the sake of appearances. And it never auto-rejects. Every decision stays with the human. The AI’s job is to make that decision faster and better informed.

Binol GeorgeCo-Founder Keelzo

See how Keelzo scores candidates, and why.

Try Keelzo Free
No credit card required
Setup in 5 minutes
Free for startups

Three Generations of Resume Screening

Not all 'AI screening' is created equal. Understanding the generations helps you evaluate vendors:

  • Gen 1: Keyword matching, exact word matching against job description. Misses synonyms, context, transferable skills. Still used by most legacy ATS tools.
  • Gen 2: NLP-based parsing, extracts structured data (skills, experience, education) from resumes. Better but still rule-based at the core.
  • Gen 3: Semantic AI, understands meaning, context, and relevance. 'Led a 12-person engineering team' matches 'engineering management' even without the exact phrase.
  • Gen 3.5 (current): Embedding-based vector matching, converts both resumes and job descriptions into mathematical vectors. Similarity is computed in semantic space, not keyword space.

Semantic Understanding

Modern AI understands that 'people management' and 'team leadership' mean the same thing. Keyword matchers don't.

Contextual Scoring

AI weighs experience duration, role seniority, industry relevance, and skill recency, not just presence/absence.

Bias Awareness

Good AI screening assistively ranks candidates. It never auto-rejects. Humans make the final call.

Where AI Resume Screening Fails

Career changers: AI trained on historical data undervalues non-traditional paths
Non-English resumes, most models are English-centric; multilingual candidates are disadvantaged
Creative roles, design, writing, and artistic skills don't encode well into structured data
Over-optimised resumes, candidates gaming the system with keyword stuffing get artificially high scores
Small sample bias, with fewer than 5 applications, AI scoring is less meaningful than human review

Frequently Asked Questions

Quick answers about how ai resume screening actually works (and where it fails).

It can. Keyword-based screening disadvantages career changers and non-traditional candidates. Semantic AI reduces this but doesn't eliminate it. Keelzo's AI ranks candidates assistively, it never auto-rejects. A human recruiter always makes the shortlist decision.
Accuracy depends on the technology. Keyword matchers miss 40–60% of qualified candidates. Semantic AI (like Keelzo's) achieves 80–90% alignment with human recruiter rankings in internal benchmarks. No AI is 100% accurate, which is why human review remains essential.
Keelzo converts both the job description and each resume into semantic embedding vectors using large language models. The similarity between these vectors produces a match score. Additional signals, experience years, skills overlap, seniority fit, adjust the final score. Results are presented as a ranked shortlist, not a binary pass/fail.

Explore More