Your Candidates Are Already Being Interviewed by AI. Most of Them Weren’t Told.

In a 2026 survey, 63% of U.S. job seekers said they had already experienced an AI interview.

Seventy percent of those candidates said they were not clearly told upfront that AI would evaluate them.

And 21% said they discovered that fact only after the interview had begun.

Those numbers deserve more attention than another debate about whether AI is "coming for recruiters."

It is already here.

The more consequential question for Talent Acquisition leaders is whether organizations have built the governance, communication and human accountability necessary for candidates to trust what happens next.

Because the emerging evidence suggests something uncomfortable:

The problem may not be that candidates categorically reject AI. The problem is that organizations are deploying it faster than they are explaining it.

And that distinction changes what TA leaders should do about it.

The AI hiring conversation has been asking the wrong question

For several years, much of the recruiting industry's AI debate has revolved around capability.

Can AI screen resumes?

Can it schedule interviews?

Can it source candidates?

Can it summarize conversations?

Can it conduct interviews?

Can it recommend who advances?

Increasingly, the answer to many of those questions is yes.

But capability is not governance.

A system being technically able to perform part of the hiring process tells us very little about whether an organization should use it in that way, how candidates should be informed, who remains accountable for the outcome, what happens when the system gets something wrong, or how its use changes the experience of applying for a job.

That distinction matters because hiring decisions are unusually consequential.

An automated recommendation about a movie changes what someone watches tonight.

An automated recommendation about a candidate can influence someone's livelihood.

Candidates appear to understand that difference.

In Gartner's first-quarter 2025 survey of 2,918 job candidates, only 26% said they trusted AI to evaluate them fairly. Thirty-two percent worried AI could incorrectly fail their application, while 25% said they trusted employers less when AI was used to evaluate their information.

Those are survey perceptions, not proof that AI systems actually make unfair decisions.

That distinction is important.

But candidate perception is itself part of the hiring experience.

An organization can deploy a technically sophisticated system and still create an experience candidates perceive as opaque, unfair or untrustworthy.

And a substantial body of research suggests those reactions are not simply resistance to new technology.

What 53 studies tell us about algorithmic hiring

One of the strongest pieces of evidence published this year does not come from a recruiting technology vendor.

Researchers Josephine Moritz, Larissa Pomrehn, Holger Steinmetz and Marius Wehner published a 2026 meta-analysis in Human Resource Management Review examining reactions to algorithmic decision-making across Human Resources.

The analysis synthesized 365 effect sizes from 73 samples across 53 studies, representing 24,578 participants.

Their conclusion was not that algorithms are inherently bad.

It was more nuanced — and more useful.

Algorithmic decision-making was negatively associated with system-related reactions including justice, fairness, trust and trustworthiness, as well as organization-related reactions including organizational attractiveness and job-pursuit intentions.

Importantly, those effects varied according to how the technology was used.

The extent of the algorithm's role in the decision mattered. So did the nature of the interaction between the person and the system.

That is an important finding for Talent Acquisition.

The relevant design question is therefore not:

Do candidates like AI?

It is:

What authority are we giving AI, at what point in the hiring journey, under what conditions, and with what human accountability?

Those are fundamentally different questions.

And organizations treating every AI use case as interchangeable may be creating risk they cannot see on their recruiting dashboards.

An AI scheduling assistant is not an AI interviewer

Consider two hypothetical implementations.

Company A uses AI to coordinate calendars, identify available interview times and send scheduling options. A recruiter remains responsible for the candidate relationship and intervenes when exceptions occur.

Company B asks candidates to complete an AI-conducted interview. The system evaluates their responses and produces a recommendation that materially influences whether they advance.

Both companies can accurately say:

"We use AI in recruiting."

But the candidate experience, decision consequence and governance requirements are radically different.

That distinction is becoming increasingly important because regulatory frameworks are beginning to recognize it too.

New York City's Automated Employment Decision Tool law, for example, restricts covered employers and employment agencies from using certain automated employment decision tools unless the tool has undergone a bias audit within the preceding year and specified information is made publicly available. Covered candidates residing in the city must also receive advance notice of the tool's use and information about the job qualifications and characteristics it will evaluate.

Illinois' Human Rights Act now prohibits employers from using AI in specified employment contexts when that use has a discriminatory effect based on protected classes, and it establishes notice requirements around covered employer AI use. The relevant amendment took effect January 1, 2026.

Internationally, the EU AI Act explicitly identifies certain AI systems used for recruitment and selection — including systems analyzing and filtering applications or evaluating candidates — as high-risk applications.

The legal requirements differ significantly by jurisdiction and technology, and organizations should obtain appropriate legal advice rather than treating this article as a compliance checklist.

But the direction is difficult to miss.

"We bought a recruiting tool" is increasingly insufficient governance for technology that can influence people's employment opportunities.

The disclosure problem

The Greenhouse findings are especially interesting because they challenge one assumption frequently made about candidate resistance to AI.

Candidates may not simply want employers to stop using it.

They appear to want to understand what it is doing.

Greenhouse's 2026 Candidate AI Interview Report surveyed 2,950 candidates across the United States, United Kingdom, Ireland, Germany and Australia, including 1,200 U.S. respondents. Unless otherwise noted in its release, Greenhouse reports its findings from the U.S. sample.

Among those respondents:

70% said employers had not clearly disclosed upfront that AI would evaluate them.

21% said they learned AI was involved only after the interview began.

38% wanted assurance that a human reviewed the AI's evaluation before a decision was made.

29% wanted evidence that the system had been audited for bias.

Perhaps most revealingly, experiences were not uniformly negative.

Thirty-eight percent said an AI interview left them with a more positive perception of the employer, while 34% reported a more negative one.

That does not prove disclosure causes better candidate outcomes.

The survey does not establish causality.

But it does undermine the simplistic narrative that candidates inherently reject AI interviewing.

Some candidates are clearly capable of walking away impressed.

The question is what separates those experiences from the ones that damage trust.

Transparency helps — but "we use AI" is not enough

Research predating today's generative-AI boom provides another warning.

In a 2021 experimental study involving 124 participants, researchers examined whether providing information about an automated interview improved applicant reactions.

The results were surprisingly complicated.

Simply giving candidates process information did not necessarily improve their reactions and could sometimes worsen them. Providing a justification for why the automated process was being used performed better, although neither intervention increased organizational attractiveness compared with providing no information.

More recent experimental research similarly suggests that different forms of AI transparency can influence perceptions around person-job fit assessments.

The implication is subtle but important.

Transparency should not become another compliance checkbox.

Imagine receiving this message before an interview:

Artificial intelligence may be used in connection with your application.

Technically, the organization disclosed something.

Experientially, the candidate learned almost nothing.

Compare that with communication that answers:

• Where AI appears in the hiring process.

• What it does.

• What information it evaluates.

• How its output is used.

• Whether a human reviews the result.

• Who remains accountable for the hiring decision.

• What a candidate should do if something goes wrong.

That is not merely disclosure.

It is orientation.

And hiring experiences need orientation whenever uncertainty increases.

Candidates are using AI too — heavily

There is another reason TA leaders should resist framing this as employers versus candidates.

Candidates are adopting AI themselves.

Gartner reported that 39% of candidates in its fourth-quarter 2024 survey of 3,290 job candidates had used AI during the application process. Among those users, common uses included generating résumé text, cover-letter text, writing samples and responses to assessment questions.

A smaller 2026 study from Selection Lab involving more than 800 applicants reported that 55% currently used AI while applying for roles, with respondents expecting that use to increase.

Different samples and methodologies produce different estimates, so these percentages should not be treated as directly interchangeable measures of prevalence.

But the direction is clear.

Both sides of hiring are automating.

Candidates use AI to tailor resumes, prepare for interviews and complete applications.

Employers use AI to source, screen, schedule, assess and increasingly interview candidates.

Each side then becomes suspicious that the other side is no longer authentic.

That is how hiring enters an automation trust spiral.

Candidates believe an algorithm may reject them, so they use AI to optimize themselves for the algorithm.

Employers encounter increasingly optimized applications, so they deploy more technology to determine which candidates are authentic.

Candidates encounter more automated defenses, so they deploy better automated tools.

Eventually both sides are using machines to determine whether the human on the other side is real.

That is not science fiction.

It is an operating-design problem.

Fraud is real. That doesn't justify indiscriminate suspicion.

TA teams have legitimate reasons to worry about candidate fraud.

Gartner's second-quarter 2025 survey of 3,000 job candidates found that 6% acknowledged interview fraud, defined by Gartner as posing as someone else or having another person pose as them during an interview.

Separately, a March 2026 peer-reviewed study examined self-reported generative-AI use in pre-hire assessments across large applicant samples.

In one study of 5,675 applicants, fewer than 3% self-reported using generative AI, although as many as 19% reported using generative AI when combined with broader algorithmic resources such as search engines.

Those findings are not contradictory.

They are measuring different behaviors.

Using ChatGPT to improve a resume is not the same thing as having someone impersonate you during an interview.

Using a search engine during an assessment is not automatically fraud.

Generating an entire work sample may or may not violate the employer's rules depending on what candidates were told.

The governance failure occurs when organizations collapse all of those behaviors into one category:

AI = cheating.

If candidates are expected to follow rules about AI use, organizations first need to define those rules.

Otherwise employers risk enforcing standards candidates were never given.

The emerging double standard

This leads to perhaps the most uncomfortable part of the AI hiring conversation.

Organizations increasingly expect candidates to disclose when they use AI.

Candidates often receive considerably less visibility into the organization's use of it.

That asymmetry matters.

If a candidate submits AI-generated answers while pretending they personally wrote them, an employer may reasonably question whether the application represents the candidate's abilities.

But if an employer allows AI to evaluate that candidate without clearly explaining the system's role, the candidate may reasonably question whether the hiring process represents the organization's values.

Both concerns can be legitimate simultaneously.

PathPair's position is not that employers and candidates must use identical rules.

Their responsibilities are different.

The principle should instead be reciprocal transparency:

The more consequential AI becomes to either side's representation or decision-making, the clearer its role should become.

That standard does not require revealing proprietary algorithms.

It does not require publishing fraud controls.

It does not require candidates to provide a forensic accounting of every tool used while editing a resume.

It requires enough transparency for both parties to understand the conditions under which they are participating.

TA leaders need an AI inventory before they need another AI policy

Many organizations begin AI governance by writing a policy.

That may be backwards.

Before deciding what the rules should be, Talent Acquisition should know where AI already exists.

That can be harder than it sounds.

AI may be embedded inside:

ATS functionality.

CRM tools.

Sourcing platforms.

Scheduling systems.

Assessment vendors.

Interview-intelligence platforms.

Fraud-detection products.

Job-ad tools.

Chatbots.

Resume-ranking functionality.

Candidate matching.

Transcription.

Interview summaries.

Email drafting.

Analytics.

And features can change through vendor updates without the organization deliberately implementing a new system.

A useful first exercise is therefore not:

"What AI tools have we purchased?"

It is:

"At which points in the candidate journey can an automated system influence what a candidate sees, what we know about them, how they are evaluated, or what happens to them next?"

Those are very different inventories.

The second reveals the actual hiring experience.

Four questions every AI touchpoint should answer

TA leaders do not need to become machine-learning engineers.

But every consequential AI touchpoint should have an accountable owner who can answer four basic questions.

1. What is the system actually doing?

"AI-powered recruiting" is not a meaningful operational description.

Is the system drafting?

Summarizing?

Matching?

Ranking?

Scoring?

Recommending?

Rejecting?

Conducting an interview?

Detecting possible fraud?

The verb matters because the consequence changes with it.

2. What authority does its output have?

A recommendation a recruiter routinely reviews is different from a score that automatically removes candidates from consideration.

Human review also needs to be substantive.

A person clicking "approve" on hundreds of algorithmic recommendations is technically human involvement.

It may not represent meaningful human judgment.

3. What does the candidate know?

Would a reasonable candidate understand that the technology exists and how it affects their participation?

If the answer is no, ask why.

There may occasionally be legitimate reasons not to disclose particular controls — fraud detection is an obvious example where excessive detail can undermine the control itself.

But secrecy should be intentional and justified, not simply inherited from a vendor implementation.

4. What happens when it fails?

Every system fails eventually.

A candidate may be incorrectly flagged.

A transcript may be wrong.

An accommodation may not work.

A ranking may appear inconsistent with the evidence.

A vendor may change its model.

A candidate may challenge the result.

Who owns the exception?

Who can override the technology?

Who communicates with the candidate?

What evidence is retained?

If nobody knows, the organization does not have an AI system.

It has an AI dependency.

The financial question: What does opaque AI actually cost?

This is where responsible analysis requires restraint.

There is evidence that applicants can react negatively to algorithmic decision-making.

There is survey evidence that candidates sometimes withdraw from AI-enabled hiring processes.

There is evidence that organizational attractiveness and job-pursuit intentions can be affected.

But the available evidence does not support a universal dollar value for "AI candidate experience damage."

Any consultancy claiming that an undisclosed AI interview costs every company some predetermined amount is manufacturing precision.

The financial exposure is organization-specific.

Consider a modeled scenario — not an industry benchmark.

Suppose an employer has 1,000 qualified candidates per year reach a hiring stage involving consequential AI.

Assume, hypothetically, that 5% of otherwise viable candidates withdraw specifically because of how that AI experience is implemented.

That would equal:

1,000 × 5% = 50 candidate withdrawals.

What are those withdrawals worth?

We cannot responsibly answer without additional evidence.

If the roles are easily replenished, financial exposure may be minimal.

If they are difficult-to-fill positions where qualified replacement candidates require additional sourcing, interview time or vacancy days, exposure may be meaningful.

If candidates are also customers, the risk may extend further.

But those consequences must be measured.

The correct business question is therefore not:

"How much money does bad AI cost?"

It is:

"What candidate behaviors are associated with our AI-enabled process, and what business consequences can we actually substantiate?"

That is a harder question.

It is also the one executives can trust.

The bigger risk is designing hiring around distrust

AI has exposed a weakness in modern recruiting that technology alone cannot repair.

Employers increasingly distrust whether applications represent candidates.

Candidates increasingly distrust whether people are actually evaluating their applications.

Employers deploy technology to validate candidates.

Candidates deploy technology to survive automated hiring.

Both sides respond rationally to the system they encounter.

And the system becomes less human as a result.

This is precisely why the phrase "human in the loop" is insufficient.

A human can technically exist somewhere in the workflow while having almost no meaningful relationship with the candidate.

The better question is:

Where does human judgment actually matter?

Scheduling an interview?

Probably not most of the time.

Explaining an unexpected delay?

Often.

Summarizing interviewer notes?

Potentially useful for AI.

Deciding whether a candidate deserves consideration?

Human accountability becomes far more important.

Drafting a rejection?

AI can help.

Owning the fact that a person has been rejected?

That responsibility remains human.

Technology should remove administrative friction so people have more capacity for judgment, communication and care.

If instead automation removes the human interaction while preserving all of the uncertainty, we have not engineered a better hiring experience.

We have simply made an impersonal one faster.

What TA leaders should do now

The immediate response does not need to be "stop using AI."

That would ignore legitimate benefits.

AI can reduce administrative burden. It can help organize information. It can improve accessibility in some contexts. It can help recruiters manage workloads that would otherwise make timely candidate communication nearly impossible.

The evidence does not justify an anti-AI position.

It justifies a governed-AI position.

Start by mapping every AI touchpoint across the candidate journey.

Identify what each system does, what data it uses, what decision or experience it can influence, who owns it and what happens when it fails.

Separate administrative automation from evaluative automation.

The closer a system moves toward judging people or materially influencing employment decisions, the stronger the requirements for validation, oversight, transparency and human accountability should become.

Review candidate communications.

Do not assume vendor disclosure is sufficient simply because a sentence exists somewhere in an application privacy notice.

Ask whether a reasonable candidate actually understands the role AI plays.

Define acceptable candidate AI use too.

If candidates may use AI to prepare but not during an assessment, say that.

If resume assistance is acceptable but generated work samples are not, say that.

If particular tools are permitted, say that.

Ambiguity creates accidental violations.

Finally, measure candidate behavior around these experiences.

Do candidates withdraw disproportionately after automated stages?

Do candidate questions increase?

Are complaints concentrated around particular tools?

Do candidates understand who is making the decision?

Are exceptions escalating?

Do recruiters themselves understand the systems they are expected to explain?

Without measurement, "our candidates seem fine with it" is not evidence.

It is an assumption.

The future of AI hiring may depend less on intelligence than accountability

Artificial intelligence will almost certainly become more capable.

That is not the interesting prediction anymore.

The interesting question is whether hiring organizations will mature quickly enough to govern what those systems can do.

Candidates are already adapting.

Regulators are already paying attention.

Academic research is beginning to tell us how people react when algorithms enter consequential human decisions.

And employers are discovering that automation creates new responsibilities alongside new efficiencies.

The companies that navigate this well probably will not be the ones with the most AI.

They will be the ones that can clearly answer:

Where is AI being used?

Why is it there?

What authority does it have?

What does the candidate know?

Who remains accountable?

And what happens when the technology gets it wrong?

Those are not technology questions.

They are hiring-system questions.

And ultimately, they are trust questions.

PathPair's founding philosophy is that technology can accelerate hiring, but people remain responsible for creating trust.

The evidence emerging around AI hiring makes that distinction increasingly practical rather than philosophical.

AI can assist judgment.

It can organize information.

It can remove repetitive work.

It can even conduct parts of an interview.

But an organization cannot automate accountability.

Someone still has to own the experience.

If this raised questions about your own hiring process

You do not need to know whether your organization has an "AI problem."

You need visibility into what candidates actually experience.

PathPair's complimentary Bridge Assessment is designed to determine whether enough evidence exists to justify a deeper investigation into the hiring experience — including the systems, technology, communication, ownership and operational conditions shaping it. PathPair's operating model explicitly requires evidence before recommendations and allows the answer to be no further engagement recommended when a deeper Diagnostic is not justified.

Because the goal should never be to find problems.

It should be to understand reality.

—

SOURCES & FURTHER READING

Moritz, J.M., Pomrehn, L., Steinmetz, H. & Wehner, M.C. — “A Meta-Analysis on Reactions to Algorithmic Decision-Making in Human Resource Management,” Human Resource Management Review, 2026.
Peer-reviewed synthesis of 53 studies and 24,578 participants.
Read the study

Greenhouse — 2026 Candidate AI Interview Report, May 2026.
Multi-market candidate survey; 2,950 respondents overall, including 1,200 U.S. respondents.
Read the Greenhouse findings and methodology

Gartner — “Gartner Survey Shows Just 26% of Job Applicants Trust AI Will Fairly Evaluate Them,” July 31, 2025.
Reports findings from several Gartner candidate surveys conducted in 2024 and 2025.
Read Gartner's research summary

Robie, C. — “Candidate Generative AI Use in Pre-Hire Employment Assessments: Self-Reported Incidence and the Impact of Warnings,” International Journal of Selection and Assessment, March 2026.
Read the open-access study

Langer et al. — “Highly Automated Interviews: Applicant Reactions and the Organizational Context,” Journal of Managerial Psychology, 2020.
Experimental evidence examining applicant reactions and social presence in automated interviews.
Read the research abstract

Langer et al. — “Spare Me the Details: How the Type of Information About Automated Interviews Influences Applicant Reactions,” International Journal of Selection and Assessment, 2021.
Read the study

New York City Department of Consumer and Worker Protection — Automated Employment Decision Tools / Local Law 144.
NYC AEDT guidance

Illinois General Assembly — Illinois Human Rights Act, Artificial Intelligence provisions. Effective January 1, 2026.
Illinois statutory text

European Union — Artificial Intelligence Act, Annex III / Employment. Recruitment and candidate-evaluation applications are among the employment use cases identified as high-risk under the Act.
EU AI Act employment classification