Fake candidates are applicants who misrepresent their skills, experience or identity, increasingly with AI-generated resumes and rehearsed answers. Screening built on self-reported information rewards the best performer rather than the best candidate. Assessments that measure behavior in the moment are harder to fake, because there is no correct answer to rehearse.
That distinction matters more now than it did even a year ago.
AI has made it remarkably easy to improve the application a hiring team sees without necessarily changing the capability of the person behind it. Gartner found that 39% of job candidates were already using AI during the application process in late 2024. More recent research underscores how widespread AI-assisted job searching has become: a Clutch survey of 590 U.S. job seekers found that 80% use AI tools during their job search, while 86% said AI helped them increase the number of applications they submit each week.
Using AI to improve a resume is not inherently dishonest. But it changes what a resume can tell you.
When almost anyone can generate polished application materials, prepare sophisticated interview responses, and optimize how their experience is presented, hiring teams need to ask a more fundamental question:
Which hiring signals can a candidate not fake?
The application layer just broke
For decades, hiring processes have relied heavily on what candidates tell employers about themselves.
A resume reports experience. An application reports qualifications. An interview asks candidates to describe what they have done and what they would do.
All three can contain useful information. But all three depend, to varying degrees, on self-report. Generative AI changes the reliability of that signal because candidates can now tailor resumes, refine their positioning, anticipate interview questions and rehearse polished responses at a scale that wasn’t previously possible.
Using those tools does not automatically make someone a fake candidate. The problem is what happens when a hiring process treats a polished presentation as evidence of capability.
There is also a more serious subset of the problem. Harver’s research suggests that 5% to 15% of applicants may have an “intent to deceive” or “no intent to stay.”
For talent leaders, that creates a widening gap between presentation and potential.
The question is no longer simply whether an application looks convincing. It is whether the hiring process can distinguish what a candidate says about themselves from what they can actually demonstrate.
Traditional screens can reward the best performer
Consider a familiar interview question: “Tell me about a time you had to make a difficult decision with incomplete information.”
A candidate can research that question before an interview. They can ask an AI tool to generate an ideal answer. They can refine it against the job description. They can rehearse it until it sounds spontaneous.
What exactly are you measuring when they deliver it well?
Possibly judgment. But you may also be measuring preparation, coaching, access to technology or simply the ability to perform convincingly in an interview.
That has always been a limitation of self-reported hiring information. Generative AI amplifies it.
The strongest hiring signal, therefore, may not be another question about what a candidate says they can do.
It may be observing what they actually do.
Detection matters. But detection is an arms race.
Organizations still need protections against impersonation, unauthorized assistance and other forms of hiring fraud. Harver has built anti-fraud and proctoring capabilities specifically for that purpose.
But fraud detection and candidate measurement answer two different questions.
Detection asks: Is something suspicious happening?
Measurement asks: What can this person actually demonstrate?
Employers need both, but they should not confuse them.
Every new method for detecting AI assistance creates an incentive to find a new way around it. That is the nature of an arms race. You can strengthen identity verification, monitor suspicious behavior and identify anomalies, and those controls have value.
For a deeper look at those safeguards, see Harver’s Raising the Bar on Hiring Integrity: Harver’s Anti-Fraud & Proctoring Capabilities.
The larger opportunity is to reduce how much the hiring decision depends on information that can be generated or rehearsed in the first place.
That is a measurement problem.
Move from what candidates say to what they can demonstrate
The shift away from self-report starts by building more of the hiring decision around objective evidence of what a candidate knows, how they think, how they behave, and what they can actually do.
Harver’s predictive assessments are designed to do exactly that.
Instead of relying primarily on resumes and interview responses, employers can evaluate candidates across the characteristics that matter for success in a specific role. Harver offers more than 450 assessment options covering work behaviors, cognitive abilities, job knowledge and skills, language proficiency, realistic job previews, simulations, work preferences and other job-relevant characteristics.
Consider the difference in the signal.
A candidate can write that they have excellent customer service skills. A simulation can show how they respond to a realistic customer interaction.
A candidate can claim proficiency in a language or job-specific skill. An assessment can measure it.
A candidate can describe themselves as a strong problem solver. A cognitive assessment can provide evidence of relevant abilities.
That doesn’t make assessments immune to every form of candidate fraud. It changes what the employer is asking the candidate to provide. Instead of relying solely on a candidate’s account of their capabilities, the hiring process creates opportunities to demonstrate them.
Harver then matches assessment results against the characteristics associated with success in a particular role, helping employers identify fit using consistent, job-relevant evidence rather than presentation alone.
That distinction becomes increasingly valuable as AI makes resumes, applications and interview preparation easier to optimize. The more easily a candidate can perfect what they say, the more important it becomes to measure what they can demonstrate.
And some forms of assessment take that principle one step further by measuring behavior as it happens.
The signal that is harder to fake: behavior in the moment
Harver’s Game-Based Assessments, powered by pymetrics, take a fundamentally different approach.
Rather than asking candidates to describe themselves, the assessments use a series of interactive experiences to observe behavior as it occurs.
The assessments can evaluate attributes such as attention, decision-making, risk tolerance, effort, learning and focus. Each assessment generates thousands of behavioral data points.
There is no ideal story to memorize. There isn’t a list of interview questions for an AI assistant to answer on the candidate’s behalf. And candidates aren’t simply being asked to rate themselves on traits they know an employer values.
They have to respond.
That difference is important.
If I ask you whether you are comfortable taking calculated risks, you can infer what answer I might want.
If I observe your decisions while you are navigating uncertainty, I have a different kind of evidence.
One is a claim.
The other is behavior.
Real behavior gives us a different view of candidates
Game-Based Assessments are not useful because games are inherently impossible to manipulate. No responsible assessment scientist should claim that any hiring method is completely “uncheatable.”
They are useful because the underlying measurement makes traditional forms of rehearsal less effective.
The candidate is producing a pattern of behavioral information across multiple interactions rather than supplying one obviously desirable response.
Harver describes its Game-Based Assessments as capturing thousands of data points while measuring non-verbal, behavioral attributes. The platform has matching data from more than 100 million assessments, and Harver reports a 98% completion rate for its gamified assessment experience.
That gives employers something increasingly valuable in an AI-mediated hiring environment: evidence that does not begin and end with the candidate’s description of themselves.
It can also surface candidates a resume may undervalue.
A person may lack a familiar job title, recognizable employer or traditional credential while demonstrating the attention, learning, decision-making or other attributes associated with success in the role.
When applications become easier to manufacture, evidence of potential becomes more, not less, important.
References can be rehearsed, too
The same principle applies beyond assessments. If the goal is to make hiring decisions based on evidence rather than presentation, reference checking deserves the same scrutiny.
A reference may sound credible on the phone, but that doesn’t necessarily mean the person is who the candidate says they are. A family member could pose as a former manager. A candidate could attempt to complete their own reference check. And a friend who knows the right things to say can reinforce a work history that was already exaggerated elsewhere in the hiring process.
That makes reference checking another potential point of vulnerability when the process relies primarily on trust and self-reported information.
Technology can add another layer of verification. Checkster, Harver’s Reference Checking solution, uses a proprietary algorithm to assess the legitimacy of references and flag potentially fraudulent activity, including candidates passing family members off as former managers or attempting to complete their own reference checks.
The solution also collects structured feedback from multiple references and brings that information together in a single report, giving hiring teams another source of evidence to compare against what they have already learned about a candidate.
That doesn’t mean every discrepancy proves fraud, just as an unusual assessment result doesn’t automatically mean someone cheated. It means hiring teams have another signal telling them when something deserves a closer look.
That’s the larger lesson for hiring in an era of generated applications: don’t rely on a single signal, and don’t assume that a convincing signal is necessarily a verified one. Combine demonstrated behavior with structured reference checking and appropriate integrity controls to build a more complete picture of the person behind the application.
Stop asking only whether you can catch the fake
Hiring integrity matters. Identity matters. Fraud prevention matters.
But if organizations focus exclusively on catching candidates who cheat, they risk missing the more consequential question: What evidence are we trusting in the first place?
A resume can be optimized.
An application can be generated.
An interview answer can be rehearsed.
A reference can be fabricated.
Those signals aren’t worthless. But their limitations are becoming harder to ignore.
Behavior demonstrated in the moment gives hiring teams another form of evidence, one based less on how candidates describe themselves and more on how they actually respond.
The hiring organizations that adapt won’t simply become better at detecting suspicious applications. They will become less dependent on easily manipulated signals.
Which leaves one question every CHRO and talent acquisition leader should be asking:
Can your hiring process tell a real candidate from a good performance of one?
Find out what your process would tell you about a candidate who rehearsed every answer. Explore Harver’s Game-Based Assessments.
Frequently asked questions
Can candidates cheat on a game-based assessment?
No assessment should be described as completely impossible to cheat. The advantage of a game-based behavioral assessment is that it reduces reliance on questions with an obvious “correct” answer. Instead of simply asking candidates to report desirable qualities, the assessment captures patterns of behavior across interactive tasks. That makes conventional preparation and rehearsed responses less useful.
Are AI-written job applications a problem for hiring?
AI assistance itself is not necessarily a problem. Candidates can legitimately use AI to improve writing, organize experience or prepare for interviews. The hiring problem arises when employers treat polished self-reported information as strong evidence of underlying capability. As AI makes that information easier to optimize at scale, organizations need additional objective signals.
What is the difference between detecting cheating and designing it out?
Cheating detection looks for suspicious behavior after a candidate encounters an assessment or hiring step. Designing for integrity reduces the value of cheating by measuring things that are difficult to supply through a rehearsed response. Employers still need fraud detection and identity controls, but assessment design can reduce dependence on signals that are easily manipulated.
How do you tell a real candidate from a rehearsed one?
Do not rely on a single signal. Combine appropriate identity and integrity controls with scientifically validated assessments that capture demonstrated skills, abilities and behavior. The goal is not to decide whether a candidate “sounds authentic.” It is to gather objective evidence that supports or challenges what the candidate says about themselves.
