There is a term for what is happening in hiring right now, and it comes from the CEO of an applicant tracking system company. Daniel Chait of Greenhouse calls it an “AI doom loop.” Job seekers use AI to game automated screening tools. Employers use AI to sort through the flood of nearly identical applications that results. Each side’s solution makes the other side’s problem worse. Nobody wins.
The Myth That Drives the Loop
The entire cycle runs on a belief: that applicant tracking systems, known as ATSs, automatically rank and filter candidates, and that only the top 10 to 20 percent of applicants ever get seen by a human. Everyone else, the bottom 80 percent, disappears into a black box.
This belief is widespread enough that a cottage industry has grown around it. Tools like Jobscan, which costs between $30 and $50 per month, promise to optimize a resume for ATS algorithms. Users learn to swap “percent” for the ”%” symbol, avoid two-page resumes, and pack their documents with keywords. Data scientist Jodi Beggs tested one such system and found that small formatting choices moved her score up or down. Her conclusion was pragmatic: if the goal is to please automated systems, then the content that works for machines may not be the content that reflects who she actually is.
Here is what most coverage of this topic misses: the premise is not always true. Chait says there is substantial folklore around how ATSs actually work. No two systems are identical. Whether a given ATS uses AI at all depends on the specific product and which features a company has paid for and activated. Some organizations use automated ranking. Others have humans review every single application. Kim Jones, vice president of human resources at Toshiba, is explicit: her team reads every application personally. She notes that automated optimization is not going to help candidates get through her process, because her process does not work that way.
Two Experiments That Reveal the Gap
The distance between the myth and reality becomes clearest when organizations actually test the assumption. Doist, a small fully remote company that hires internationally, ran exactly that kind of internal experiment. The team took roles that had already been filled, fed the job descriptions and all saved applicant materials into an ATS ranking system, and asked a simple question: would the people they actually hired have made the AI-generated shortlist?
In two cases they tested, the answer was no. The candidates who were hired, and who were performing well roughly six months into their roles, did not appear in the AI’s shortlist at all. There was some overlap in who received interview invitations, but the people the company ultimately chose were filtered out by the automated system.
This is not a minor discrepancy. It suggests that optimizing for an ATS score and being a strong candidate for a specific role can be two entirely different things.
On the job seeker side, a design professional named James Jacobsen illustrates how far the optimization impulse can go. He spent five months in a search that consumed as much time as a full-time job. He used Claude and ChatGPT to tailor his application materials, but saw little movement. He then built what he described as the opposite of an ATS: a personal tracking system that used AI to comb listings, score opportunities against his own criteria, log rejections with notes, and surface the highest-priority roles. He also used AI to critique his portfolio, leading to a six-hour revision. He got calls. He did not get offers. Chait’s assessment of candidates like Jacobsen is direct: doing more of the same is not the answer.
What the Loop Actually Signals
The doom loop is not just a story about technology misfiring. It is a story about trust collapsing on both sides of a transaction. Job seekers report sinking enormous effort into applications and hearing nothing back. Employers report receiving hundreds of applications that look nearly identical, which is partly a consequence of AI-assisted writing tools making every document sound the same. Both sides reach for AI to manage the volume, and the volume keeps growing.
Chait frames it plainly: this is the first time he can recall when both sides are unhappy with the hiring process. The system is not working for anyone.
What gets lost in the optimization race is the behavior that actually differentiates candidates: a genuine cover letter, which Jones says almost nobody submits anymore; direct outreach to companies before a role is posted; networking, which remains underrated in a market flooded with applications. These are human signals in a system increasingly designed to filter them out.
The deeper issue is that AI tools are being applied to a problem they did not create and cannot fully solve. The job market’s dysfunction, including scarce postings, ghost jobs, and a breakdown of communication between candidates and employers, predates the current wave of AI adoption. Automation is accelerating the symptoms, not treating the cause.
In Short
AI is being used by both job seekers and employers to manage a hiring process that is already under strain. The tools each side uses to cope make the other side’s experience worse, creating a self-reinforcing cycle. The assumption that automated ranking controls every hiring decision is not universally true, which means candidates optimizing for machines may be solving the wrong problem entirely. The behaviors most likely to cut through, a real cover letter, targeted research, direct human contact, are the ones the optimization tools cannot replicate.
Based on reporting from Wired.