Felix, my youngest adult son, who navigates an intellectual disability following multiple infant neurosurgeries, is largely monosyllabic in writing but highly conversational in person. Imagine an emotionally intelligent SpongeBob SquarePants in human form—that’s my son.
Earlier this year, Felix used ChatGPT to fill out a warm and eloquent application, write an email, and polish his resumé for a volunteer role at a charity store. He passed the process—a Zoom interview and an online video tutorial, along with a test to prove he was human. His references confirmed he was reliable and good with people, and his police check was squeaky clean.
There was no in-person interview. Despite disclosing his disability designation in the application, no one sat with him before he was hired to notice that he might require accommodations—training, structure, guidance. The hiring system was optimized to process candidates online, bypassing the face-to-face connection and conversation that would have determined fit.
Four days after starting work, Felix was let go. The feedback was kind. But the overoptimized digital hiring process failed to test for human competency for the role.
The Phantom Fit
Felix wasn’t trying to deceive anyone. He had used ChatGPT the way most people today use Google—to support learning and discovery, and to turn his thoughts into a socially acceptable format. His story is a baseline case study of what is missing in most organizations: a hiring manager validating a candidate in person to assess whether the human they’re hiring can do the job.
I call this a “phantom fit”—when an AI-assisted presentation inflates a candidacy beyond actual capability or fit, creating a false-positive impression. It is not fraud. No rules are being broken. Universities, career coaches, and employers themselves actively encourage candidates to use AI to accelerate their work. The tool is legal, widely available, and socially normalized.
This is where Jim Rohn’s old management maxim becomes an AI-era challenge: don’t send your ducks to eagle school, because it fails everyone in the system. What AI is doing for candidates, at scale, is miscasting people— a second-order effect—potentially presenting talent to incompatible environments.
A 2026 Fabric analysis of nearly 20,000 interviews found that more than a third of candidates had used AI to misrepresent their abilities. Among technical roles, that figure rises to 48 percent; independent data from CodeSignal corroborates the trajectory, finding technical assessment misrepresentation of human skills doubled in a year, from 16 percent to 35 percent of attempts. More troubling: the same Fabric analysis found 61 percent of candidates flagged for AI-assisted interview behavior scored above the pass threshold and would have advanced through a standard hiring process undetected. Though, application processes become a moot point when 75 percent of applicants are using AI tools. Don’t we now just need to assume everyone is using AI in their applications and spend time focusing on the interview process?
James Pycock, VP of Product at Albert, a Series A AI company in the Bay Area, has witnessed a more concerning trend. “We’ve interviewed people who are clearly using AI during the interview, literally reading off screen,” he told me. “And we had one candidate who wasn’t human at all—a computer-generated avatar someone had sent. It took our senior engineering interviewer the better part of half an hour to realize this wasn’t actually a person.”
Phantom fits come in many forms: the candidate who aces a technical screen because an LLM answered in real time, then can’t use the tools on the job; the executive whose articulate leadership philosophy was AI-generated, who becomes epistemically fragile when a novel situation arises; and the people trying to compete in a job market with AI-polished applications landing in incompatible environments.
Phantom fits have become an oversized problem: talent overoptimize their abilities in an overoptimized system which simultaneously removes the human experience that determines whether people are fit for the mission of an organization—not just fit for purpose.
‘Bot vs. Bot’
While candidates inflate their presentations, employers will be tempted to double down on AI screening tools, further ranking candidates against job descriptions using algorithms that reward keyword optimization over genuine fit. In addition, job descriptions themselves are increasingly AI-generated, producing requirements that don’t reflect what the role actually needs. This is now “bot vs. bot”: the company’s AI filters candidates, the candidate’s AI games the filters, and at no point does either party develop a reliable picture of the other, slowing down hiring and costing the employer more time.
In my first book Elephants Before Unicorns in 2019, I argued that people leaders needed more emotionally intelligent practitioners capable of assessing—human to human—what would actually benefit an organization’s trajectory for the AI era. The industry invested in the opposite direction. AI adoption in HR tasks climbed to 43 percent in 2025, up from 26 percent the year before. Recruiters were among the first roles cut. And human roles were automated: IBM, for example, replaced roughly 200 HR positions with AI agents.
The outcomes are unfavorable. Cost-per-hire is up 113 percent since 2017. Time-to-hire went up, too. The entire pitch was speed, and speed worsened. In its report, “Recruitment Is Broken,” SHRM concludes that the AI arms race does not benefit either side. With recruiter numbers diminishing, hiring managers are left to carry the relational and assessment load.
In-Person Is the First Step to Trust
The corporate response has been swift. Google, Cisco, and McKinsey have all reintroduced mandatory in-person interviews, specifically to counter AI-assisted fraud. In-person interview rounds rose from 24 percent in 2022 to 38 percent in 2025, specifically in response to AI-assisted cheating in remote interviews. “Remote work and advancements in AI have made it easier than ever for fake candidates to infiltrate the hiring process,” said Scott McGuckin, VP of Global Talent Acquisition at Cisco. Albert is doing the same, said James Pycock. “We’ve just gone back to basics,” he told me.
That said, organizations have known for decades that in-person is not a reliable detector of phantom fits either. A candidate with strong social skills and a plausible backstory will pass a face-to-face interview just as easily as a remote one. Physical presence does not equal confirmed identity, and confirmed identity does not equal genuine capability or fit.
The in-person assessment creates the conditions where the whole person shows up. Where you notice how someone responds to the unexpected. Where the accommodation conversation becomes possible. Where a human being can sense something that an AI system might ignore. More interestingly, during an era in which human collaboration is needed to solve significant climate and societal problems, the in-person interview is an appropriate setting for each party to ask questions of each other and where leaders and talent can demonstrate their problem-solving skills in real time.
The Human Signal
To tackle the problem of phantom fits and institutional miscasting, we need to rebuild the human signal—the relational intelligence that organizations systematically stripped out during downsizing.
Here are five shifts for the C-suite to model for their organization:
1. Change your interview style.
During the in-person interview, avoid relying on accomplishments or productivity outputs. Instead, verify systems-thinking and organizational mission deliverables. Start asking how candidates think when AI scaffolding is removed. Introduce genuine ambiguity across systems. Request failure examples that require lived memory. Stop asking questions a GPT can answer.
2. Innovate the interview structure.
A single interview is a snapshot of a well-prepared and rehearsed performance. Build in project-based assessments and pay candidates for their time to deliver the project. A phantom fit rarely survives the first month. The cost of a bad hire can take up to six months or longer to restore; a structured working interview surfaces it before hiring, saving tens of thousands of dollars and stalled projects.
3. Rebuild the reference conversation to surface miscasts.
Most companies use references as a proof of previous employment. Instead, have hiring managers conduct references via phone or Zoom and change the topic from what the candidate achieved to how they solved challenges when things went wrong. What environments brought out their best, and what didn’t? This may save months or years in miscasts and provide direct shortcuts to help your new team member integrate faster.
4. Train interviewers to probe their unease.
A 2025 Checkr survey of 3,000 managers found that 59 percent had personally suspected a candidate of using AI to misrepresent themselves, yet only 19 percent were confident their process would catch it. While the instinct was present, for most, the mechanism to act on it was not.
When something feels off— the answer is too polished, the specifics too vague—follow your gut. Ask the candidate to walk you through a specific moment—not the outcome, but what they were thinking in the middle of it. Ask what changed, or what went wrong, and then ask what they’d do differently now. Shift from the prepared answer to the unrehearsed one by introducing a constraint they haven’t anticipated: “Tell me the same story but from your manager’s perspective.” Or say: “That’s interesting—tell me more about the part where XYZ happened.” A candidate with genuine experience will go deeper. A phantom fit will loop back to the surface instead of being happy to share the details.
If properly developed by the organization, what can often be interpreted as a negative bias with a gut read can become the beginning of a better question, a better conversation, and the beginning of an excellent collaboration.
5. Have the fit and accommodation conversation explicitly.
With an estimated one in five people globally identifying as neurodivergent—and workplace accommodation requests rising for the second consecutive year—the need to understand what a candidate actually requires to do their best work is no longer optional.
Every organization has an operating reality, or cultural style of working, that most candidates will never have encountered. For instance, Nvidia’s model—no hierarchy, no 1:1 meetings, public group feedback, mission as the only boss, ranked #5 on Fortune’s 100 Best Companies to Work For— is genuinely unlike that of most workplaces. A candidate who can’t operate without structure, who needs private feedback to function, or who hasn’t thought about how they adapt to new environments that “attack the problem” will be a miscast hire waiting to happen.
The conversation is the same whether you’re hiring into a flat AI company or a charity store: what does this person need to succeed here, can they adapt fast enough and can this environment enable them to do their best work?
Build this line of inquiry into every interview. Ask what structure helps them. Ask what overwhelms them. Ask how they’ve adapted to a new culture before, and what that cost them. Ask what they need from a manager and provide mentoring and coach support for the first 100 days. In the past, this line of questioning would have been signaled as a welfare HR check-box but today we need to ask “what conditions bring out your best, and what gets in the way?”
This is how you find out whether you are the right environment for this person—and whether they are right for you. A polished application can survive a Zoom interview. It cannot survive a direct, human conversation about how someone actually works, asked by someone genuinely listening for the answer.
My son’s story needed just one honest conversation through an attentive hiring manager.
The emotionally intelligent leader tasked in hiring talent for this complex time of AI, climate and societal disruption will succeed when the human signal is prioritized.
Albert’s Pycock landed in the same place from a completely different direction. “I think leaders may end up being more human,” he said. “Back to human relational skills.”
