AI hiring algorithms create a doom loop of mutual distrust
The resume was technically flawless. Two pages, clean formatting, quantifiable achievements, and not a single typo. Yet the system still penalized it. Jodi Beggs, a data scientist navigating the current job market, watched as her carefully crafted application materials were dissected by an algorithm she could not see and whose rules she could not fully decipher. Her experience, reported by Business Insider, illustrates how opaque these systems have become. 3. The feedback was maddeningly specific: her resume was too long by one page, her use of a middle initial created an inconsistency, and writing out the word “percent” instead of using the symbol cost her precious points. She was not applying to a person anymore. She was applying to a machine that had been programmed by other machines, and the entire process felt less like a job search and more like a bizarre game of digital Twister.
Beggs is part of a growing cohort of job seekers who have discovered that their applications are being filtered by artificial intelligence before any human eyes ever see them. These systems, known as Applicant Tracking Systems or ATS, have become the gatekeepers of the modern labor market, promising to sort through hundreds of identical-looking applications to find the top 20 percent of candidates worth a second glance. The pitch is efficiency: in a market flooded with resumes, who has time to read them all? The reality, as Beggs discovered, is a feedback loop of mutual distrust where both sides are deploying AI to outsmart the other, creating what one industry executive calls a doom loop of escalating automation with no clear winner.
The tools that have emerged to help job seekers navigate this algorithmic minefield operate on a simple premise: if the machines are screening you, then you need to know what the machines want. Services like Jobscan, which costs between thirty and fifty dollars per month, compare a candidate’s resume against a specific job description and provide a compatibility score, suggesting keywords to add and formatting tweaks to make. Beggs found herself in a strange position, optimizing her human career history for machine consumption, wondering if she was polishing her materials for a robot that might not even be there. The uncomfortable truth she stumbled upon is that the entire industry of ATS-optimization services is built on a foundation of folklore and fear, not necessarily on how hiring actually works in most organizations.
The Unproven Premise of Automated Ranking
The belief that an algorithm is pre-screening applications and discarding the bottom 80 percent before a human sees them has become an article of faith among job seekers. It drives millions of dollars into the pockets of resume-optimization companies and fuels an entire cottage industry of career coaches who promise to help candidates beat the bots. Yet when you talk to the people who actually build and sell these systems, a more nuanced picture emerges. Daniel Chait, the CEO of Greenhouse, a company that sells ATS software to employers, is refreshingly candid about the limitations of his own industry. 1 He describes much of what job seekers believe about these tools as folklore, noting that no two systems are alike and that the technology changes so rapidly that an AI feature might function one way today and completely differently tomorrow.
The reality is that automated ranking absolutely happens in some organizations, but in many others, hiring is still managed personally by humans from start to finish. The division does not correlate with company size or the volume of applications received. It seems to be driven more by corporate philosophy and the personal preferences of whoever is running the human resources department. Kim Jones, the vice president of human resources at Toshiba, is adamant that her team reviews every single application that comes through their system. She acknowledges that candidates using AI to polish their materials is fine, but she insists it will not help them get through the ATS, because there is no automated culling happening in the first place. The screening comes down to job requirements, salary expectations, and whether the person would be eligible for rehire.
This discrepancy between what job seekers believe and what employers actually do creates a profound inefficiency in the labor market. Candidates spend hours tailoring their resumes to match keywords they think an algorithm wants to see, while hiring managers complain that they receive hundreds of applications that look nearly identical because everyone has used the same AI tools to optimize for the same perceived criteria. The result is a paradox: the more candidates try to game the system, the more they all look the same, which in turn pushes some employers to actually implement the automated ranking that candidates feared in the first place. It becomes a self-fulfilling prophecy where the fear of AI screening creates the conditions that make AI screening necessary.
When the Algorithm Gets It Wrong
Nadia Vatalidis, the head of people at Doist, a fully remote company that hires internationally, decided to test whether the AI ranking features in her ATS would actually make better hiring decisions than her human team. 2 She took roles that had already been filled successfully and fed the job descriptions along with all the applicant materials from the hiring process back into the system. The question she wanted to answer was straightforward: would the AI short-list the same candidates that her human team had ultimately chosen to interview? The results were troubling. In two instances, the person who was ultimately hired and who had been working out great for over six months did not even make the AI’s short list.
The experiment revealed a fundamental limitation of using pattern-matching algorithms to predict human potential. The AI could identify candidates who looked good on paper, who had the right keywords and the right experience, but it could not identify the candidates who would thrive in the specific culture of Doist, who would collaborate well with the existing team, or who would bring the kind of creative problem-solving that comes through in a human conversation but rarely translates to a resume. Vatalidis found that there was some overlap between what the AI selected and what her team selected, but the most important hires, the ones who truly made a difference, were the ones the algorithm missed entirely.
This is not an isolated anecdote. Across the industry, recruiters and HR managers are discovering that AI ranking systems are remarkably good at one thing: identifying candidates who are good at writing resumes for AI ranking systems. The optimization arms race has created a situation where the signal that these algorithms are trying to detect has been corrupted by the very tools designed to boost it. When everyone uses the same AI to polish their materials, the resumes all start to look the same, and the algorithms that are supposed to differentiate between candidates end up seeing a sea of indistinguishable text. The system becomes circular: AI generates content to please AI, and the humans who are supposed to be making hiring decisions are left with less information, not more.
The Job Seeker’s Counterattack
James Jacobsen, a design professional who started his job search five months ago, represents a new kind of resistance to the AI hiring machine. Rather than simply using AI to polish his resume and cover letter, he decided to use it to build his own tracking system that would work in his favor. He instructed Claude, an AI assistant, to comb through job listings, analyze descriptions, and log them in a structured database. He developed an elaborate scoring system based on the type of work, the seniority level, and his salary requirements, which varied depending on whether the position was in-person, hybrid, or remote. The AI assistant would highlight the top jobs he should apply to and immediately remind him if a position he had previously rejected was reposted weeks later.
What Jacobsen built was essentially the inverse of an ATS: a system designed not to filter him out but to filter opportunities in. He spent less time searching and more time applying to high-quality positions that actually matched his criteria. The system was not perfect, and he did not land a job through it, but it changed the calculus of his search. Instead of feeling like he was throwing his resume into a black hole, he had a sense of agency, a way to manage the overwhelming volume of job postings and focus his energy where it was most likely to pay off. He also used AI to critique his portfolio, which led to a six-hour revamp that he credits with getting him a call from a prospective employer, though it still did not result in an offer.
The arms race between job seekers and employers is not just about resumes and keywords. It has extended into the interview process itself. Kim Jones at Toshiba has noticed a troubling trend where candidates are using AI to answer interview questions in real time. She describes hearing a pause, perhaps the sound of typing, followed by a verbose and polished answer that seems to come from nowhere. This creates a new kind of challenge for interviewers, who must now try to determine whether they are evaluating the candidate or the candidate’s AI assistant. The technology that was supposed to make hiring more efficient has instead made it more opaque, introducing new layers of uncertainty and distrust on both sides of the table.
The Collapse of Trust in the Labor Market
The current economic environment has amplified these problems to a breaking point. Job postings are scarce, and the ones that do exist are often met with a flood of applications, many of which are generated by AI tools that can customize a cover letter and resume for any position in seconds. Scams and ghost jobs, positions that are posted but never actually filled, have become so prevalent that some states are considering new laws to combat them. Candidates sink hours into applications and hear nothing back, concluding that their materials were rejected by an algorithm. Employers, meanwhile, receive hundreds of applications that look identical, leading them to believe that candidates are just spraying their resumes everywhere without any genuine interest in the position.
This breakdown of trust is what Daniel Chait from Greenhouse calls the AI doom loop. 1 Each side has a problem, and they are using AI to solve their own problem in ways that make the problem worse for the other side. Candidates use AI to apply to more jobs more quickly, which floods employers with low-quality applications. Employers respond by using AI to filter those applications more aggressively, which makes candidates feel like they need to use AI to optimize their materials even more. The more AI use begets more AI use, and the system gets worse for everyone. Chait says this is the first time he can remember when both sides are unhappy with the hiring process, and he does not see an easy way out of the loop.
For job seekers trapped in this system, the advice from experts is surprisingly low-tech. Chait suggests that doing more of the same is not the answer, and that candidates should spend time researching companies where they might want to work, even if those companies are not the first names that come to mind. Jones at Toshiba notes that she almost never sees a cover letter anymore, and that applicants who bother to submit one immediately stand out from the crowd. Networking, that old-fashioned skill of talking to actual humans, has become underrated in a job market dominated by digital applications and automated screening. The people who are getting hired, it seems, are the ones who find ways to bypass the system entirely and connect with the humans who are making the decisions.
The Human Cost of the Doom Loop
The stakes of this algorithmic arms race are not abstract. Every rejected application represents a person who needs to pay rent, feed their family, or maintain their health insurance. Every hour spent optimizing a resume for a machine is an hour not spent on actual skill development or meaningful networking. The emotional toll of applying to dozens of jobs and hearing nothing back, of suspecting that your carefully crafted materials were never seen by human eyes, is immense. The current system has created a generation of job seekers who feel like they are screaming into a void, and the tools that were supposed to help them navigate the market have become another source of anxiety and expense.
Beggs, the data scientist whose resume was dinged for being two pages long, captures the absurdity of the situation perfectly. She admits that if her job is to please the robots, then maybe the machines do prefer AI-generated content, even if she does not like it. This is the ultimate irony of the doom loop: candidates are being forced to produce content that they do not believe in, optimized for systems that they do not understand, in the hope of impressing employers who may not even be using the automated screening they fear. The entire edifice of ATS optimization is built on a foundation of uncertainty and folklore, and yet it shapes the behavior of millions of job seekers who cannot afford to ignore the possibility that the machines are watching.
The voice that should be heard last in this conversation is that of the job seeker who is doing everything right and still getting nowhere. Chait has a message for these people, and it is worth repeating: it is not you, it is the system, and the system stinks. The AI doom loop is not a failure of individual effort or a reflection of personal worth. It is a structural problem created by the intersection of economic scarcity, technological hype, and mutual distrust. Until employers and job seekers find a way to break the cycle, to trust each other enough to communicate directly and honestly, the machines will continue to grade the machines, and the humans will be left to pick up the pieces.
Sources
1. Greenhouse
2. Doist