Job Seekers and Employers Trapped in AI Screening Arms Race

A data scientist recently discovered her résumé lost points for being two pages long and for inconsistent use of her middle initial. Changing "percent" to the % symbol boosted her score. These weren't human preferences—they came from an online tool designed to optimize application materials for applicant tracking systems.

The scenario illustrates a growing problem in hiring: job seekers increasingly use AI to tailor their materials for automated screening systems, while employers deploy AI to filter through floods of similar-looking applications. The result is what one industry executive calls an "AI doom loop" where both sides use technology to solve their problems, only to make things worse.

The Folklore Around Applicant Tracking Systems

Tools like Jobscan operate on the premise that most companies use AI to automatically rank candidates, with only the top 10 to 20 percent of applicants ever reviewed by humans. To beat these systems, candidates pay $30 to $50 monthly for services that promise to optimize their résumés with the right keywords and formatting.

The problem: this premise is often wrong. Daniel Chait, CEO of the applicant tracking system company Greenhouse, says automated ranking happens in some organizations, but many still rely on human review from start to finish. No two systems work the same way, and whether AI screening is used at all depends on which features a company purchases and activates.

Yet as long as applicants believe AI gatekeepers exist, they'll continue trying to game them with their own AI tools.

When AI Screening Fails the Test

Nadia Vatalidis, head of people at Doist, ran an experiment to see if automated ranking could help her remote company handle high application volumes. Her team fed completed job descriptions and all saved applicant materials into an AI screening system to see if it would identify the same candidates they had interviewed and hired.

The results were revealing: in two test cases, the employees they had hired—people who were performing well after six months—didn't make the AI-generated short list.

Kim Jones, vice president of human resources at Toshiba, takes a different approach entirely. Her team reviews every application manually. She doesn't mind if candidates use AI to polish their materials, but says it won't help them get through the system because culling happens based on job requirements and salary expectations, not keyword optimization.

The Mutual Breakdown of Trust

Chait describes the current situation as unprecedented: both job seekers and employers are deeply unhappy. Candidates invest enormous time in applications and hear nothing back. Employers receive hundreds of nearly identical applications and turn to AI screening for quick differentiation, which only encourages more AI-generated applications.

One design professional spent five months treating his job search like a full-time role, using Claude and ChatGPT to streamline everything from tracking listings to revamping his portfolio. Despite the effort and a few callbacks, he received no offers.

Chait's advice moves away from technological solutions: research companies thoroughly, write cover letters (Jones says she almost never sees them anymore), and prioritize networking over spray-and-pray application strategies.

Why it matters

The AI arms race in hiring represents a fundamental market failure where both sides deploy technology to solve legitimate problems—application overload for employers, opacity for job seekers—but each solution triggers an escalating response that benefits neither party. The breakdown suggests that technological optimization may be reaching its limits in contexts that fundamentally require human judgment and connection.

These details were first reported by WIRED.