Greenhouse and Claude Users Fuel an AI Hiring Doom Loop
As job seekers use models like Claude to bypass automated screening, employers are deploying applicant tracking systems to filter them out, creating a mutually destructive AI hiring loop.
The widespread adoption of artificial intelligence in recruitment has triggered what Greenhouse CEO Daniel Chait calls an "AI doom loop," where both job seekers and employers use automated tools to outsmart each other. Candidates increasingly rely on optimization platforms like Jobscan, which cost $30 to $50 per month, to tailor their resumes for applicant tracking systems (ATS). This optimization is driven by the belief that automated filters immediately discard the bottom 80 percent of applicants, leaving only the top 10 to 20 percent for human review.
To combat this automated gatekeeping, applicants are turning to advanced large language models. For instance, design professional James Jacobsen used Anthropic's Claude and OpenAI's ChatGPT to refine his portfolio and application materials. When that failed to yield results, he used Claude to build a personalized tracking system that analyzed job listings, logged opportunities, and calculated custom scores based on seniority, role type, and salary requirements.
However, the efficacy of these automated hiring systems remains highly contested. Nadia Vatalidis, head of people at Doist, ran an experiment feeding successful past hires back through an ATS ranking tool. In two instances, the actual hires—who had proven to be excellent employees after six months—failed to make the AI-generated short list. Similarly, Kim Jones, vice president of human resources at Toshiba, emphasized that her company still uses humans to review every application, noting that AI-optimized resumes do not actually help candidates bypass their screening process.
For HR practitioners and recruiters, this cycle of mutual automation is degrading the quality of the hiring pipeline. Employers are inundated with hundreds of nearly identical, AI-polished applications, prompting them to rely even more heavily on automated ranking features. For job seekers, the reliance on algorithmic optimization can backfire; simple formatting quirks, like using a middle initial or writing "percent" instead of "%", can arbitrarily alter an applicant's score, proving that the current system of automated recruitment is deeply flawed.
This is our own summary of reporting by WIRED AI
