Job seekers are using AI to optimize applications for automated screening systems that may not exist, while some employers deploy AI because they receive waves of increasingly similar applications. The result, according to hiring executives and applicants interviewed by Wired, is a loop in which each side’s attempted solution makes the other’s problem worse.
The cycle is sustained partly by uncertainty. Applicant-tracking systems differ widely, and the features an employer buys or activates determine whether AI ranks candidates at all. Yet the belief that a machine controls access to human reviewers is enough to change how people write and submit applications.
Applicants optimize for an uncertain machine
Data scientist Jodi Beggs tested a service that compares application materials with job descriptions. Her score changed because her résumé ran to two pages, because a middle initial appeared inconsistently and because she replaced the word “percent” with a symbol. Feedback that sensitive can encourage applicants to write for a presumed ranking system rather than for the person doing the hiring.
Services such as Jobscan operate on the premise that automated systems rank candidates and leave only the top 10 to 20 percent for review. Automated ranking does happen in some organizations, but Wired found that others continue to evaluate every application manually. Greenhouse CEO Daniel Chait said no two applicant-tracking systems are identical and their behavior can change quickly.
That makes paid optimization a risky proposition. A recruiter cited by Wired received interviews and an offer despite very poor Jobscan scores. The example does not prove such services never help, but it undermines the claim that a particular score reliably predicts whether a real employer will respond.
Employers confront uniform applications
A weak hiring market intensifies the pressure. Candidates spend substantial time applying and often hear nothing, while scams and ghost jobs make genuine openings harder to identify. Employers, meanwhile, report receiving hundreds of applications that look alike and may turn to automated ranking to create quick differentiation.
Practices vary by organizational culture rather than simply company size. Toshiba human resources executive Kim Jones said people review every application at her company. She accepts AI-assisted polishing but said it does not help candidates pass the tracking system. She has noticed unwanted AI use during interviews, where pauses, typing and unusually verbose answers can reveal outside assistance.
Remote software company Doist tested AI rankings against completed hiring decisions. Its team fed old job descriptions and applications into a system to see whether the people actually interviewed and hired would appear on the automated shortlist. There was some overlap, but in two tests the successful employees—both performing well after about six months—were absent from the shortlist.
More AI reinforces the hiring loop
Design professional James Jacobsen used AI to polish applications without improving his results. He then built a more useful personal tracking system that searched listings, scored roles against his preferences and remembered why he had rejected repeated postings. The approach saved search time and produced some interest, but no offer.
Jacobsen also used AI feedback for a six-hour portfolio revision and received a prospective employer’s call two days later, though that opportunity went no further. His experience illustrates both the practical value and the limit of automation: it can organize effort, but the source provides no evidence that it can manufacture demand or guarantee a hiring decision.
Chait calls the wider pattern an AI doom loop. Applicants generate and optimize more material, employers face more uniform volume, and automated filtering then motivates candidates to optimize even harder. Both groups lose trust while believing they must add another layer of AI to keep up.
The reporting does not show that every hiring tool fails or every AI-assisted application is ineffective. It shows that opaque and inconsistent systems encourage defensive behavior on both sides. When nobody knows whether a human or model will read the next document, applications become less personal, filtering becomes more aggressive and effort increases without repairing the basic connection between a qualified candidate and an actual job.
