How ‘trace hiring’ can reclaim human authenticity in the age of AI

Applying for a job used to require genuine effort – a natural barrier that automatically filtered out casual candidates. Today, AI has completely destroyed that barrier, leaving hiring teams drowning in a flood of low-effort applications.

The numbers tell the story. In late 2024, applications submitted on LinkedIn rose 45.5%, even as job postings on the platform fell 10.6%. That gap marks a shift driven by AI writing tools. A 2025 NACE survey found that students using AI in their job search submitted about 60 applications on average, double the overall average of 30. Applying used to take real effort, and that effort filtered out casual applicants on its own.

The operational toll is staggering. Greenhouse surveyed 4,100 recruiters and hiring managers across the United States, United Kingdom, Ireland and Germany for its 2025 AI in Hiring Report. It found that 34% of recruiters now spend up to half their working week filtering spam applications, and 91% have caught candidates not being honest during the process.

All of this leaves organizations facing a critical question: how do you identify genuine talent when the application itself no longer proves effort?

A bad hire drags down team output and accelerates turnover. Less visible is the cost of a hiring system built to reward the wrong thing. AI-powered applicant tracking systems stepped in to handle the new volume, and adoption nearly doubled in one year, from 26% of organizations in 2024 to 43% in 2025.

But these tools are trained on historical hiring data, and so risk inheriting historical bias. A University of Washington study of three large language models on more than 500 resume comparisons found white-associated names were favoured 85.1% of the time against 8.6% for Black-associated names. The tools surface candidates who resemble people the company has already hired.

Yet, machine learning models have a fundamental blind spot. The skills that matter most today involve working with AI, because employees must learn to think alongside these systems. A study of 5,179 customer support agents by MIT and Stanford found that access to a generative AI tool raised productivity by 14% on average and by 34% for novice workers.

A small group of organizations has started to respond. Daniel Chait, chief executive of Greenhouse, says HR leaders are deliberately adding friction back into hiring through longer applications, more human screening and skills assessments, a trend the Society for Human Resource Management has also confirmed. Some companies have dropped the text-based application entirely, asking candidates for a one-minute video, an image hidden inside a job posting, or simply an in-person conversation. A candidate willing to appear on camera and follow an unusual instruction has already shown more intent than a resume ever could. Friction alone does not reveal capability, though. Most companies still lack a process that makes a candidate’s strengths visible under real conditions.

Termites leave pheromone traces that coordinate collective work without central direction, the same logic behind Wikipedia edits and GitHub commits. Applied to hiring, this means designing interactions that leave authentic traces of how a candidate thinks, adapts and builds. I call this approach “trace hiring”, and it runs across three stages.

First, candidates respond by video or conversation to a real problem the organization is working on. Candidates who lean entirely on AI produce generic, similar sounding answers. Candidates who engage seriously produce something distinct and move forward.

Second, selected candidates work in small groups on a new problem, given on the day, with one hour to reach a tangible output. Laptops, AI tools and pen and paper are all allowed. Evaluators watch the process rather than the output, noting who drives the conversation, who adapts when an idea fails and who pulls a disengaged teammate back in.

Third, finalists return individually to answer one question. What did you learn from that exercise? Most describe the task or their team. A smaller number describe where their own thinking surprised them, and that capacity for honest self-assessment is hard to fake because it lives in memory. A Korn Ferry study of more than 6,000 employee assessments across 486 publicly traded companies found that companies with higher stock returns also had employees with greater self-awareness.

Trace Hiring spreads the cost of screening fairly. Anyone can advance regardless of degree, network or a polished resume, and the process rewards people who think originally and engage honestly, qualities that rarely show up on an application form.

Companies have spent decades treating academic credentials, brand name employers and years of experience as proxies for quality. These signals are now actively misleading. When Professor Chad Van Iddekinge analyzed more than 80 workplace studies spanning 60 years, he found that prior work experience has almost no link to how employees perform in a new role. And as the World Economic Forum’s Future of Jobs Report 2025 highlights, the other major proxy – the college degree – is rapidly vanishing from job descriptions, as companies shift towards skills-based hiring.

That correlation may be thinner with the rise of AI. Researchers Riedl and Bogert of Northeastern University studied five years of data from 52,000 people on an online chess platform and found that higher skilled players engaged with AI more productively, widening their existing advantage, while broader access caused thinking to converge around the same systems. The candidates worth hiring are the ones who use AI the best.

Organizations that hire well this decade will watch what a candidate does when given a real problem, a group of collaborators and a tight deadline. What someone can do alongside others matters more than what they claim to have done alone. The organizations that win this labour market will be the ones that commit to transparent, effortful human assessment, not the ones that screen the most candidates algorithmically.

AI did not create this problem. It accelerated a reckoning that was already overdue.