Skillfishing in the Age of AI

Executive Summary AI has worsened the hiring problem it was meant to solve , making misrepresentation easier, screening less reliable and the damage done by poor hiring decisions harder to detect.

The résumé is no longer a reliable indicator of talent. Fewer than 4 in 10 employers trust it as a top signal.

Adding a human in the loop does not fix the issue: Recruiters mirror algorithmic choices ~90% of the time .

Most bad hires are detected only after the damage is done , as no performance baseline is set at hiring.

Restoring discipline at 3 checkpoints - at the time of offer, 30 days and 90 days later - can help remedy the situation.

. The past five years saw two of HR's most celebrated advances arrive in quick succession: a shift to skills-first hiring, which promises to surface talent on merit rather than credentials; and the deployment of AI screening tools, which promises to make that process faster, fairer, and more consistent. Both have proven to be genuine improvements. But both, in combination, have also created a vulnerability that most organisations have not yet recognised, let alone addressed. That vulnerability is skillfishing : the growing phenomenon of candidates whose apparent skills, as presented through AI-generated resumes, AI-coached interviews and rapid micro-credentials, do not reflect their actual on-the-job capabilities. Skillfishing is not a new form of dishonesty. What is new is the industrial scale at which AI has made it possible, the structural reasons why AI screening tools are poorly positioned to catch it and the measurement gaps inside organisations that allow it to go undetected long after a bad hire has been made.

This paper argues that AI has not merely failed to prevent skillfishing. It has, in several important ways, actively enabled it. It traces a four-stage loop through which AI inflates candidate signals, AI screening misreads those signals, the wrong hire enters the organisation and weak post-hire measurement buries the damage.

The Problem The Hiring Arms Race No One Is Winning

Something has gone wrong in hiring. Despite years of investment in AI screening tools, talent intelligence platforms and skills-first recruitment redesign, the process is not getting better; it is only getting harder for everyone. According to SHRM 's 2025 benchmarking survey, both cost-per-hire and time-to-hire have increased over the past three years, the same period that saw accelerated AI adoption across the recruitment lifecycle. More applications are generating less signal, not more.

The reason is an arms race. Employers deployed AI to screen candidates faster. Candidates, knowing AI reads their applications first, deployed AI to optimise those applications for the very tools evaluating them. The result is a closed loop in which both sides use AI to communicate with each other, and authentic human capability has become harder than ever to see through the noise. At the centre of that loop is skillfishin, a phenomenon HR has always known, but which now operates at a new order of magnitude.

Defining the Moment

Three forces have converged to make this the defining hiring challenge of 2026. The first is urgency : 94% of CHROs identify AI capability as their most pressing talent need, yet only 35% feel prepared to assess it. That gap between demand and readiness is exactly where skillfishing thrives. The second is a shift to skills-first hiring which, however well-intentioned, removed the credential proxies that once provided a verification floor without replacing them with anything equally reliable. The third force is the most consequential. The tools candidates use to present themselves have outpaced the tools employers use to evaluate them. That asymmetry is neither dishonesty nor a technological failure, but a structural capability gap. This is the heart of the skillfishing trap.

Stage One of The Loop How AI Inflates the Skills Signal

The scale of AI use in job applications is now large enough to be treated as a structural feature of the hiring market. Around 65% of candidates use AI at some point in the application process, according to Career Group Companies' 2025 Market Trend Report. Real-time AI whisper tools and answer-generating apps during live interviews have become prevalent enough that 57% of hiring managers now say they should never be permitted, according to TopResume 's survey of 600 hiring managers. This has become the new baseline.

The most visible symptom is a collapse of the resume as a trust signal. In 2026, according to the Willo Hiring Trends Report , fewer than 4 in 10 employers rate resumes among the most reliable indicators of talent. According to Monster's AI Resume Trends Report , the percentage of resumes that include at least one AI-related term jumped from 3.7% in 2023 to 12.8% in 2025. Behind those claims, AI depth varies wildly. The signal that once existed, that putting a skill on a resume carried some implicit accountability, has been diluted to near-irrelevance. When most candidates use AI to optimise their applications for AI screening systems , homogenisation degrades the hiring signal for everyone. Matching tools trained to find differentiation cannot function when every application looks the same.

Stage Two of The Loop Why AI Screening Cannot Catch It

The instinctive response to gaming is counter-gaming: deploy better AI to detect the AI-generated applications. However, this is the wrong solution. Predictive screening models learn from historical hires, which means they learn to prefer candidates who look like past employees, not candidates who have the capabilities required for future roles. When candidates use AI to reverse-engineer the keywords a model was trained to look for, the model cannot distinguish genuine skill from constructed presentation; it was never designed to.

The human reviewer does not reliably correct for this either. Research from the University of Washington found that recruiters that use AI tools with in-built biases mirror the AI's choices up to 90% of the time, even when they had the information to decide independently. This automation bias means that adding a human to the loop does not fix the problem if that human is systematically deferring to the tool. The oversight is procedural, not substantive. A 2024 Harvard Business School and Accenture study found that 88% of employers acknowledge their screening tools reject qualified candidates. Most continue using them anyway, because the volume pressures are real and the errors are largely invisible.

Stage Three and Four of The Loop The Damage and Why No One Notices

When a skillfisher makes it through and into an organisation, the damage begins quietly. Team velocity slows, expectations go unmet and colleagues absorb the additional workload. The conventional estimate of a bad hire costing the firm 1-3x of their annual salary was established before AI-assisted misrepresentation existed at scale and before organisations were hiring urgently for skills as difficult to evaluate as AI capability. Both factors inflate the true cost substantially and neither appears cleanly in a budget line.

The deeper problem is measurement. SHRM 's research found that nearly one in four organisations has no mechanism to measure the ROI of their AI-related hires. Without a performance baseline established at the point of hire, the connection between a hiring decision and its downstream outcomes is almost impossible to make. Bad hires attributed to team fit or role ambiguity are rarely traced back to the hiring process that produced them. The loop closes on itself. The skillfisher goes undetected and the next hire is made with the same broken signal.

Three Things CHROs Should Act On The loop described in this paper is not inevitable. It is, though, the predictable result of deploying AI at the front of the hiring funnel without the verification and feedback infrastructure that gives it meaning. Breaking it requires three deliberate intervention points where human-anchored assessment overrides AI signal:

Before the offer. Work samples, live scenario assessments and structured reference conversations should be non-negotiable gates for all senior and specialist roles. The purpose is not distrust but to collect evidence AI cannot fabricate: capability demonstrated in real time, in a context where presentation tools are irrelevant.

The first 30 days. A skills baseline established at onboarding, even a structured conversation with the line manager documented and tied to hiring criteria, creates the ‘before’ state without which no meaningful performance comparison can be made. It also surfaces genuine gaps early enough to address through development rather than termination. This is the highest-value, lowest-cost intervention window in the entire hiring cycle.

The 90-day review. This should be used not just as a performance assessment but as a hiring quality audit, asking where the candidate's actual performance diverges from the signals that got them hired. Done well, this becomes the mechanism through which the loop is broken. Over time, organisations that track these divergences will learn which signals in their context are predictive and which have been inflated. That learning cannot come from AI. It has to be built deliberately, by HR leadership, one hire at a time.

The call to action is not for firms to remove AI from hiring, but anchor it with human-validated verification at the moments that matter most.