Skip to content
Essays & arguments, with sources

Reluctant Citizen

Future of Work

Rejected by a Robot: How AI Hiring Filters Shut Out the People We Need

By The Reluctant Citizen 12 min read

You spent an evening tailoring your résumé. You matched the job description line by line. You hit submit. Forty minutes later, at 11:52 p.m., a polite email arrives: “After careful consideration, we have decided to move forward with other candidates.”

Nobody carefully considered anything. Software did.

That small moment now plays out countless times a day, all over the world. Applicant tracking systems, résumé parsers, chatbot screeners, game-style assessments and AI video interviews have quietly become the front door to work. Some of them help. But too many of them are tuned to throw people out rather than find people who can do the job.

My view (and this is opinion): automated screening, as most employers use it today, is breaking one of the most human jobs an economy has: matching people with work they can do. It leaves out qualified people. It copies old biases and scales them up. And it is eroding trust on both sides of the hiring table, while the laws meant to rein it in lag years behind. The fix isn’t to smash the machines. It’s to stop letting them make the call.

The front door is now a filter#

This isn’t new, and it isn’t just a fringe tool. In 2021, Harvard Business School and Accenture published Hidden Workers: Untapped Talent, based on surveys of more than 8,000 hidden workers and more than 2,250 executives in the U.S., the U.K. and Germany. They found that more than 90% of employers used their recruiting software to initially filter or rank middle-skills (94%) and high-skills (92%) candidates.

The employers themselves admitted the cost. 88% said qualified high-skills candidates were “vetted out” because they didn’t match the exact criteria in the job description. For middle-skills candidates, it was 94%. The researchers estimated more than 27 million “hidden workers” in the U.S. alone, and similar proportions in the U.K. and Germany. These are people who want work (or more hours) but keep getting filtered out by rules like “no gaps in employment” or “must have a degree.”

Think about who has a gap on their résumé: caregivers, veterans, people recovering from illness, immigrants rebuilding careers, people who were laid off in a downturn. The report lists all of them. A rigid filter doesn’t see a story. It sees a missing date.

Old biases, now automated#

The most famous warning came from Amazon. In 2018, Reuters reported that the company had scrapped an experimental AI recruiting engine. It was trained on ten years of résumés, mostly from men, and it taught itself that men were preferable. It penalized résumés that included the word “women’s,” as in “women’s chess club captain,” and downgraded graduates of two all-women’s colleges. Amazon said recruiters never relied solely on its rankings.

You might hope newer AI fixed this. The evidence says otherwise.

A fair caveat: humans are biased too. A huge field experiment sent more than 83,000 fake applications to 108 of the largest U.S. employers. Distinctively Black names cut the chance of being contacted by 2.1 percentage points. So the honest comparison isn’t “biased AI versus fair humans.” It’s that AI learns from our past decisions and then applies them at a scale, speed and secrecy no single recruiter could match.

Age and disability: the quiet exclusions#

Some of the most serious challenges involve older workers.

In Mobley v. Workday, a job seeker alleges that Workday’s AI applicant-recommendation system discriminated by race, age and disability. He says he applied to more than 100 jobs at companies using Workday and was turned down every time. In May 2025, a federal judge in California preliminarily certified a nationwide age-discrimination collective of applicants aged 40 and over. In arguing against it, Workday itself told the court that “1.1 billion applications were rejected using Workday” in the relevant period. (That figure counts all rejections on the platform, not only AI-driven ones.) In June 2026, the court let a disability claim proceed and dismissed a race claim brought by another plaintiff. Workday denies the allegations. As of late October 2026, the case is still in pretrial litigation and no court has found Workday liable.

There is already a settled case. In 2023, the tutoring company iTutorGroup paid $365,000 to settle an EEOC lawsuit. The agency alleged the company programmed its application software to automatically reject women aged 55 or older and men aged 60 or older, turning away more than 200 qualified applicants. That was a simple coded rule, not sophisticated AI, which is exactly the point. Automation makes it effortless to exclude people at scale.

Disability and accent raise their own problems. A 2025 University of Melbourne study interviewed Australian HR professionals about AI hiring. As the Guardian reported, one vendor disclosed a transcription error rate under 10% for U.S. English speakers, but 12% to 22% for non-native speakers with accents. Participants worried that candidates whose disability affects their speech could be mis-transcribed and then scored poorly. Recruiters often couldn’t explain the tool’s decisions, so they couldn’t give candidates any feedback.

Video analysis has had its own reckoning. In January 2021, the vendor HireVue announced it had stopped analyzing applicants’ facial expressions, and its CEO said the feature “wasn’t worth the concern.” Emotion scientist Lisa Feldman Barrett told Wired that it is “a bad idea to make psychological inferences, and therefore determine people’s outcomes, based on facial data alone.”

Why employers use these tools anyway#

It would be unfair to paint employers as villains. Many are drowning.

In June 2025, The New York Times reported that LinkedIn was processing about 11,000 applications a minute, up 45% in a year. Job seekers now use AI to tailor résumés in seconds, and some pay for bots that apply to jobs for them. One HR consultant received more than 1,200 applications for a single remote role.

Research shows this flood is doing real damage to the signals employers rely on. A 2025 study of Freelancer.com by economists at Dartmouth and Princeton found that in their sample of coding jobs, the average number of applications per posting more than doubled after ChatGPT arrived, from about 29 to about 63. Before AI writing tools, a carefully tailored proposal predicted a worker who would do good work, and employers rewarded it. After, that signal largely stopped working. The authors’ model suggests that when writing can no longer signal effort, the market becomes less meritocratic: top-ability workers are hired 19% less often and bottom-ability workers 14% more often.

So we have an arms race. Candidates use AI to get past AI, employers add more AI to cope, and the humans on both ends get lost. That’s my read, anyway, and I think it’s the heart of the problem.

To be fair, AI can also help when it’s used well. In a large field experiment with about 70,000 applicants for real jobs, people randomly assigned to structured AI voice interviews were 12% more likely to get job offers than those interviewed by human recruiters. Those hires also started and stayed at higher rates. Crucially, humans still made the hiring decisions. The AI gathered information consistently; it didn’t serve as judge.

Rules are arriving, slowly and unevenly#

Around the world, lawmakers have started to respond, but the patchwork is thin.

  • European Union: The EU AI Act lists AI used “to analyse and filter job applications, and to evaluate candidates” as high-risk, which brings requirements for risk management, human oversight and transparency. But under a deal EU lawmakers reached in May 2026, those rules for employment AI won’t apply until 2 December 2027.
  • New York City: Local Law 144 took effect in July 2023 as the first law anywhere to require bias audits of hiring algorithms. A Cornell-led study checked 391 employers and found only 18 posted audits and 13 posted the required notices. Because employers largely decide for themselves whether the law covers them, the researchers called this “null compliance.” In December 2025, a New York State Comptroller audit found the city’s enforcement fell short. Of 12 test calls to the city’s 311 line to report a problem, only three reached the right agency.
  • Illinois: Since 2020, the Artificial Intelligence Video Interview Act has required employers to notify applicants, explain how the AI works, and get consent before using AI to analyze video interviews. Since January 1, 2026, the state’s Human Rights Act has also barred employers from using AI that has a discriminatory effect, or using zip codes as a proxy for protected classes, and requires them to tell workers when AI is used.
  • U.S. federal government: In early 2025, the EEOC removed its technical guidance on AI and discrimination, including its guidance on AI tools screening out people with disabilities. An April 2025 executive order told federal agencies to deprioritize enforcing “disparate impact” liability. That’s the legal theory most suited to catching a neutral-looking algorithm that harms a group. The underlying laws, like Title VII, the ADA and the Age Discrimination in Employment Act, still apply.

Trust is the real casualty#

People notice all of this. In a 2023 Pew Research Center survey, 66% of U.S. adults said they would not want to apply for a job with an employer that uses AI to help make hiring decisions. On final hiring decisions they were even clearer: 71% opposed letting AI make the call, and just 7% favored it. Among those who wouldn’t apply, the most common reason was losing the human side of hiring. One man in his 40s described the keyword problem bluntly: “I’m not a person, just a series of keywords and if I don’t fit the exact hiring model I’m immediately discarded.”

There’s a hopeful note, too. Respondents leaned toward thinking AI would beat humans at one thing: treating all applicants the same way (47% said it would do better, 15% worse). People aren’t against consistency. They’re against being judged by a black box with no human in the loop.

In my view, the damage goes beyond any single rejection. When people believe a hidden algorithm decides their future, they stop believing that skill and effort count. Some give up. That is part of what the Hidden Workers report means by people “missing from the workforce”: roughly 9 million Americans who aren’t job hunting but would work under the right circumstances. Others game the system with keyword stuffing and AI-written applications, which makes the flood worse and the filters harsher. A labor market that can’t recognize talent ends up wasting it.

The bigger picture: what we lose#

Step back from the lawsuits, statutes and surveys, and here’s what worries me most. Again, this is my opinion.

We waste human potential. Work is how most people build a life, support a family and find purpose. A system that quietly discards millions of capable people because of a gap, an accent, a missing keyword or a birth year wastes talent the world badly needs.

Bias stops being visible. A biased hiring manager can be confronted. A biased model is hidden inside a vendor contract, applied to millions of applications, and defended as “objective.” The Melbourne researchers found that even recruiters often couldn’t say why a candidate was rejected.

Matching gets worse, not better. Hiring is supposed to connect people with the right roles. When applications stop signaling anything and filters reward keyword-gaming over ability, both sides lose. The employer misses a great hire, and the candidate misses a job they’d have been great at.

What job seekers can do#

None of this is your fault, but a few things help:

  1. Write for both the machine and the human. Use a simple, single-column format with standard headings, and use the job posting’s own words for skills you genuinely have.
  2. Explain gaps briefly. A one-line note (“Caregiving, 2023-2024”) is better than an unexplained blank.
  3. Go around the front door. Referrals, direct messages to hiring managers, alumni networks and industry events still reach humans.
  4. Apply with care, not at volume. Research suggests that mass, AI-generated applications are drowning out the very signals that help strong candidates stand out.
  5. Know your rights and ask. Depending on where you live, you may be entitled to notice, an explanation, consent or an accommodation. You can always ask whether AI is being used and request a human review or an accessible alternative.

What employers should do#

  1. Keep a human in charge of every rejection. Use AI to organize information, not to deliver verdicts. The strongest positive evidence above came from setups where people made the decisions.
  2. Kill the knock-out rules. Review automatic filters for degrees, employment gaps and exact keyword matches. The Hidden Workers research shows these rules screen out people who could do the job.
  3. Test for adverse impact yourself, and keep testing. Don’t rely on a vendor’s promise. Audit by age, disability, race, sex and accent, publish the results, and act on them.
  4. Be transparent. Tell candidates when and how AI is used, offer an alternative process, and make accommodations easy to request.
  5. Fix the volume problem upstream. Write sharper job descriptions, use short work-sample tasks and structured interviews, and invest in referrals. These cut noise without silently discarding people.

The bottom line#

Automation in hiring isn’t going away, and it shouldn’t. Done well, it can make hiring more consistent and even fairer. But today’s typical setup asks software to say “no” millions of times a day with little oversight, little transparency and, in many places, little accountability. That is a choice, not an inevitability. We can choose tools that widen the door instead of narrowing it, and keep a human hand on the handle.

Sources#

Filed under Future of WorkAI HiringJob SearchFairnessHidden Workers