How AI resume screening actually works (and where it fails)

Most resume screeners, including most AI ones, are keyword matchers. They scan for '5 years React' and ignore candidates who write '5+ years building React applications'. Same meaning, different words, instant reject. That's the dirty secret of the industry.

Hireotrak works differently. We feed the full resume text AND the full job description to our own language model, served from our own infrastructure with no third-party API calls, and ask it to score fit on a 0-100 scale with reasoning.

What this gets right: candidates with non-standard titles ('Tech Lead' vs 'Engineering Manager'), career switchers, and people who describe impact instead of responsibilities.

What it still gets wrong: very senior roles (10+ years) where the model sometimes overweights recency, and roles with hyper-specific technical requirements (e.g. 'must have built a Kafka cluster from scratch') where the model can't distinguish 'used Kafka daily' from 'built one'.

The honest answer: AI screening is better than keyword matching, and worse than an experienced human. We aim for the gap in the middle — faster than humans, smarter than keywords, never a final say.

Try Hireotrak free