A logistics company advertises a junior data analyst role. Within four days it has 1,400 applications. Many of the cover letters use the same phrases in the same order. The recruiter, juggling two other roles, switches on the applicant tracking system’s AI ranking and adds a one-way video interview scored by software.
Three weeks later the shortlist looks strong on paper. In the first live interview, the top-ranked candidate pauses for several seconds before every answer and appears to be reading. Two referred candidates, both currently employed, withdrew when the automated interview invitation arrived. The other 1,300 applicants have heard nothing. Each step made sense. The result is a process neither side trusts.
Hiring has always been a contest of impressions. What has changed in the past two years is that machines now do much of the presenting and much of the judging. Candidates use AI to write CVs and apply in bulk. Employers use AI to filter and, increasingly, to interview. A small but real number of “candidates” are not who they claim to be. This guide looks at why that arms race destroys the information hiring depends on, why it erodes trust in both directions and what one employer can do about it.
Cheap signals stop carrying information.
When a polished application costs nothing to produce, polish no longer tells you who is committed or capable.
Suspicion is contagious.
Controls aimed at the few cheats are felt by the many honest candidates, and the best of them have the most options to leave.
Restore signal and trust together.
Move effort to the stages where it reveals ability, tell candidates how AI is used, and close the loop with everyone.
What is actually happening?
Recent surveys point the same way, though most come from vendors or industry bodies and are best read as indicators.
In the UK, the Institute of Student Employers’ 2026 recruitment survey, summarised by Wonkhe in October, found that 66% of employers had identified unauthorised use of AI tools during interviews and 92% had seen suspected or identified cheating at some point in the process. Most of those same employers now use AI for their first filtering stages.
In the US, iHire’s 2026 State of Online Recruiting report, reported by HR Dive on 1 October, found that 23.6% of employers had received at least one fake applicant in the past year and 55.7% named too many unqualified applications as their main problem. On the other side, 39.3% of job seekers said fake or “ghost” jobs were their main problem, 57.5% were frustrated by applying and never hearing back, and 44% viewed employers who use AI in hiring negatively.
Greenhouse’s 2026 Candidate AI Interview Report, a survey of 2,950 job seekers in five countries, found that 63% of US respondents had already experienced an AI interview, 70% of those said AI use had not been clearly disclosed beforehand, and 38% had withdrawn from a hiring process because it included one. CNBC reported in September that some candidates now blacklist such employers.
The viral stories make the same point. In October a US recruiter’s account of exposing a “deepfake” candidate, by asking him to wave a hand across his face until the live filter glitched, spread widely. Hiring has become a place where each side half expects the other to be faking.
Why does more volume mean less information?
The economist Michael Spence’s signalling theory, published in 1973, explains the core problem. Employers cannot see ability directly, so they rely on signals, which are useful only if they are cheaper for strong candidates to send than for weak ones. A well-researched, tailored cover letter used to be such a signal: it took time, and the time it took said something about interest and care.
Generative AI has cut the cost of that signal to nearly zero for everyone. A tailored letter, a fluent CV or a keyword match now says little about the person behind it.
In a four-country survey of 2,900 job seekers and 1,236 recruiters and hiring managers, 74% of US job seekers said they used AI in their search and 49% said they had submitted more applications than a year earlier. Only 21% of recruiters were very confident that their own AI systems were not rejecting qualified candidates.
Two predictable responses follow. Candidates, seeing low response rates, apply to more jobs, which lowers response rates further. Employers, overwhelmed, filter harder and automate more, which candidates then try to game. Each choice is rational individually and damaging collectively: an arms race, with more effort on both sides and less information at the end.
There is also a stranger effect. A survey released in October, widely reported under the headline that six in ten job seekers are rethinking how they write to avoid sounding like AI, found many had already rewritten applications for that reason. Candidates are trying to prove they are human; employers are trying to detect the same thing. Neither is spending that effort on whether this person can do this job.
People tend to judge fluent, well-organised text as more credible and its author as more capable. That shortcut was never reliable, and it is now actively misleading. A screener that rewards polish is selecting for access to good tools, not for the ability you are hiring.
Why does trust collapse in both directions?
Trust in hiring is reciprocal. Each side decides how much honesty and effort to offer by guessing what the other side is doing.
From the candidate’s side, it feels like shouting into silence. They apply many times and hear nothing, suspect some adverts were never live, and may then be invited to talk to software nobody mentioned. Research on applicant reactions, much of it built on Stephen Gilliland’s 1993 model of procedural justice in selection, finds that candidates judge a process fair when they can show what they can do, communication runs both ways, the rules are explained and feedback is timely. One-way, undisclosed AI interviews fail several of those tests at once. Greenhouse’s November 2025 survey found that only 8% of candidates believed AI made hiring fairer.
From the employer’s side, the experience is of being flooded and deceived. Recruiters see identical letters, candidates reading answers from a second screen and, occasionally, someone else’s face or credentials. The natural response is more control.
Here the psychology turns against the employer. Experiments by Armin Falk and Michael Kosfeld (2006) found that when one party visibly restricted another’s choices, many people responded by doing less than they otherwise would. The control signalled distrust, and distrust was returned. In hiring, the cheat barely notices the extra checks. The honest candidate feels insulted by them.
The candidates most willing to leave an unpleasant process are those with alternatives: people already in work, in demand and confident in their skills. A survey reported by TechRadar in November found employed job seekers were noticeably more likely than unemployed ones to have abandoned a process over an AI interview. A process that drives these people away still filters, but for the wrong thing.
Ghosting as a norm, not a lapse
Ghosting is better understood as a norm each side has learned from the other than as individual bad manners. When candidates expect silence, they feel less obliged to withdraw politely; 51.6% of iHire’s employers cited ghosting or unresponsive candidates as a challenge. When employers receive 1,400 applications, replying to each feels impossible, and the silence teaches the next round of candidates to expect nothing.
Candidates treated as interchangeable behave as if they are, accepting interviews they will not attend and offers they may not keep. See our guides to rejecting candidates without losing them and why good candidates say no to offers.
What managers can do
No single employer can end the arms race. Each can step out of it for its own vacancies, and gain an advantage by doing so.
Move effort to where it reveals ability. If the written application no longer separates candidates, stop using it as the main filter. Use short work samples, role-specific screening questions or a brief live conversation. Work samples and structured interviews were among the better predictors of performance long before AI; see why structured interviews beat gut feeling. They are also harder to outsource.
Disclose how AI is used, before it is used. Tell candidates which stages are automated, what is assessed and whether a person reviews the result. Greenhouse found 46% of candidates wanted the option of a human interview; offering one costs little.
Tell candidates what AI use you accept. Say plainly whether drafting or checking an application with AI is acceptable, and what is not, such as live help in an interview. Clear rules turn a guessing game into an agreement.
Verify proportionately and explain why. Identity checks are reasonable, especially for remote roles with access to systems or money. Explain that they protect honest candidates from fraudulent competition.
Close every loop. An automated but honest message is far better than silence. Set a date by which every applicant will hear something, and keep it.
Advertise only real jobs. Talent-pool adverts and roles kept open “in case” are a large part of why candidates distrust job listings. If an advert is speculative, say so.
Do not try to detect AI-written applications by style alone. Detectors are unreliable and tend to flag people writing in a second language. Judge content against the job. Automated tools in hiring are also subject to discrimination and data protection law in most countries, so check what applies before switching one on.
For your next vacancy, track how many people withdraw at each stage and how many applicants never hear anything. If good candidates leave at the automated step, the tool is costing you more than it saves.
Back to the analyst role
For its next round, the logistics company asked three specific questions about its routing data, 150 words each, replaced the recorded interview with a 20-minute call for a longer shortlist, and told every applicant on day one when they would hear back.
Applications fell to about 300. The recruiter read them herself in an afternoon, because generic answers were easy to spot. The person hired said it was the first advert in weeks that seemed written by someone who wanted to hear from her.
Discover more from Kurums | Business Intelligence
Subscribe to get the latest posts sent to your email.