The State of Interview Intelligence

The pattern hasn't changed in a century: a senior leader has a theory about what predicts performance, other employers borrow the questions because the company is successful, and checking whether successful candidates became successful employees comes later, if at all.

Structured interviews have ranked at or near the top of selection-validity research for decades (most recently Sackett, Zhang, Berry and Lievens, Journal of Applied Psychology, 2022), yet an unstructured conversation can leave an interviewer feeling so certain that checking the evidence seems unnecessary.

Capturing the conversation used to be the technical obstacle. With 85 percent of employers now using interview intelligence tooling, the harder question is what adoption delivers. Asked what their organization is investing in AI to achieve, employers put efficiency and speed first at 51 percent and better decision quality second at 45 percent. These responses describe broader AI investment priorities, rather than the demonstrated results of interview intelligence. The more revealing number is that only 23 percent report using interview intelligence to improve interview quality or reduce bias.

Employers say they want better decision quality (45 percent), yet only 23 percent use interview intelligence to improve interview quality or reduce bias.

Interview intelligence is three layers, and most buyers own one: 1) Capture, 2) Analysis and 3) Decision support.

Layer 1 · Capture

Capture covers recording, transcription, structured notes, consent and retention; most organizations have it, and many treat it as the destination.

Layer 2 · Analysis

Analysis examines question quality, rubric coverage and agreement between interviewers, and the maturity model places 14 percent of organizations at Level 4 or above, the tiers associated with that work.

Layer 3 · Decision support

Decision support brings comparable evidence into debriefs and connects interview judgments with later performance; 5 percent of employers reach Level 5, where that becomes possible.

AI adoption alone can't move an employer to Level 4; it must also meet the structure, tooling and documentation requirements. Employers receive the highest level for which they pass every gate.

Half the market has purchased tech no one uses. Sixty-three percent of employers reach at least Level 3, and roughly four in five of that group go no further. That is one of the study's most consequential findings. The tools are bought; what's missing is someone to examine the output with the authority to act on it.

Employers describe a process one level above the one they run. Eighty-eight percent say interviewers always or usually use the same core questions and a defined rubric, but only 36 percent say always, the requirement for Level 4. A transcript can pass for analysis the same way a scorecard in a shared folder passes for consistent evaluation. Confusing the presence of a record with the use of evidence helps explain why administrative improvements have outpaced analytical ones.

Crossing into Level 4 takes three commitments a software purchase can't make: a named owner for interview quality, a recurring review by people with the authority to change the process, and executive tolerance for findings that reflect poorly on senior interviewers.

Reaching Level 5 means connecting interview evaluations with post-hire performance records, and agreeing to examine what that connection reveals can be harder than the engineering.

Calibration is the least developed dimension. A recording tool can supply evidence, but someone still has to connect it to outcomes and examine the result.

Consistency, closure, transparency and proportionality likewise require decisions about how the process will run.

• Ghosting: 16 percent of leaders failed to give an interviewed candidate a final answer in the past twelve months, compared with 86 percent of laggards.
• Feedback: every leader says rejected finalists always or usually receive substantive feedback; two-thirds of laggards say the same.
• Documentation: leaders rate their ability to produce complete records at 8.89 out of 10, compared with 7.46 among laggards.

The procurement illusion. Fifty-nine percent of laggards have bought interview intelligence platforms, and their reported ghosting rate is still 86 percent. Software can make follow-through easier, but it can't make the commitment. The stronger distinction is how consistently organizations manage the work after the interview.

Where the leaders are. Leaders appear in every headcount tier, concentrated among mid-sized organizations with enough hiring volume for useful analysis. Large enterprises cluster at Level 3, with sophisticated capture tools, substantial archives and trouble connecting them to post-hire performance.

What leaders are held accountable for. Leaders rate their predictive accuracy at 8.86 out of 10 against 7.77 among laggards, which establishes greater confidence rather than greater accuracy.

Less intuitively, 67 percent of laggards say HR is formally accountable for hiring outcomes and quality of hire, compared with 50 percent of leaders and only 6 of the 24 employers at Level 4.

Holding HR accountable for quality of hire sounds rigorous, but without the analysis to show which interview judgments predicted performance, it's accountability for a number nobody can explain. The leaders have built that analysis. For them, a bad hire is a question the data can answer; for everyone else, it's a verdict.

Self-assessment isn’t self-awareness. Every tier rates itself well. Leaders return an NPS of +72 and laggards +21, a 51-point spread that shows the cohorts see their own capabilities very differently. Both still sit far above the candidate sample's +3. That isn't a statistical discrepancy; it's a structural blind spot, and maturity narrows it without closing it. Self-assessment tells you how employers see their process. It has to be read alongside what candidates actually experience.

The remaining sections examine that distance through the five Efficacy Index dimensions, comparing employer and candidate responses. We begin with the widest gap.

Employers returned 52.5 percent promoters and 4.3 percent detractors (11 people out of 257), while candidates returned 36.8 percent promoters and 33.9 percent detractors (59 out of 174).

One plausible explanation is that employers rate the process they designed, while candidates rate its delivery by whichever recruiter and hiring manager they happen to get. It's easy to admire the blueprint when you're rarely asked to walk through the building.

The gap isn’t rejection talking. Asked what makes a hiring process good, candidates ranked getting the job last, at 6 percent, far behind fairness and communication. And if a bad score were simply a missed offer, detractors would dominate the candidate sample; instead, promoters (36.8 percent) slightly outnumber them.

Employers rate their interviews 8.3 out of 10. 95% of all employer respondents rated their ability to accurately forecast future performance based on interview data. Candidates, asked the mirrored question, averaged 6.93.

The published estimate of around .42 for structured interviews describes measured prediction; the self-rating describes how confident employers feel.

That confidence persists because interviewers get so little feedback: someone who meets a hundred candidates and recommends twenty sees the good hires as proof the interview worked, while the eighty rejected candidates disappear from view.

Without validation, an organization risks confusing the confidence of its interviewers with the accuracy of their judgments, then assigning its most consequential decisions accordingly.

The true cost of the confidence gap. 56 percent of candidates say a poor interview experience would discourage them from reapplying, and 35 percent say it would discourage them from buying from the company.

For consumer-facing brands interviewing thousands of people a year, that's a commercial risk, not just a recruiting one. Recruiting produces the data, but marketing has a stake in it, and keeping interview feedback inside HR limits what the wider business can learn from it.

Sales teams used to keep about as much (a 1-5 rating and the phrase "good fit, strong need") until conversational intelligence gave them a reason to mine those calls around 2018. Those calls revealed competitive threats, pricing pressure, product gaps and churn risk that no single rep’s notes could have shown.

Candidates offer a similar real-time, informed read on how the labor market sees the company, and over the next thirty-six months the employers that use it could gain a meaningful analytical (and competitive) advantage.

The organizations that checked learned things nobody else knows. Only a handful of the leaders we interviewed had compared interview evaluations with post-hire outcomes.

A director of talent acquisition at a financial services firm in Des Moines walked us through it:

Popularity had been doing work that predictive evidence should have done. An engagement survey, exit interview or hiring-manager satisfaction score wouldn’t have answered the same question. The insight required connecting two datasets that usually sit in different systems, under different owners.

Six things the transcript may already be telling you. The applications below are hypotheses drawn from the qualitative interviews, not findings tested by this survey.

On adverse impact, ACM FAccT research found that asynchronous video screening itself reduced perceived fairness for candidates with disabilities and from underrepresented backgrounds, whether a human or an algorithm did the evaluating.

Nobody has the data to measure quality of hire. Forty-nine percent of employers list it among their top workforce priorities, and 60 percent say HR is held accountable for outcomes like it. The most common measure in our interviews was a manager satisfaction survey at ninety days, and one third acknowledged they don't measure it at all.

LinkedIn's ongoing research consistently ranks it the most important talent metric and the hardest to capture. Our research suggests why: outcome data sits in the HRIS, but the scores and reasoning behind the original decision often were never captured. Recording the conversation doesn’t automatically preserve the reasoning that turned it into a hiring decision.

Why no one uses interview data. Putting interview data to work takes one person responsible for interview quality, interview and performance data standardized in a single system, and leaders willing to change the process in response to what they find. These commitments support the move from Level 3 to Level 4.

The Des Moines director offered a useful warning: "Nobody warns you that the reward for finally measuring something is a set of conversations you would rather not have."

“I evaluate my own hiring decisions using a metric I personally control. Nobody calls that out in a calibration meeting.” - Director of TA, Global Technology Company

Candidates seek evaluation clarity, not process flexibility. Asked which interview format is fairest, only 25 percent of candidates chose a purely conversational interview; 52 percent preferred a hybrid and 20 percent a fully structured format.

Their clearer frustration is not knowing how the decision gets made: candidates rate their understanding of the evaluation criteria at 6.79 out of 10, the lowest score in the study.

A government talent leader highlighted this governance gap, noting nineteen years of pristine, fully compliant records that were “never once compared against how any of those people performed.”

Documentation competence reflects compliance, not quality. Employers rate their readiness to produce complete interview documentation in a legal dispute at 8.28 out of 10, and it tends to be strongest where an external requirement makes it unavoidable. An organization can pass a documentation check without ever checking whether its judgments predict performance.

Three rounds is an artifact, not a process. 62 percent of employers run exactly three interview rounds for professional roles, and nearly six in ten say three is ideal; 44 percent of job seekers think two is enough to judge fit.

A talent lead at a UK consulting practice explained her team's fourth stage: "That additional leadership check was instituted because an executive made a bad hire seven years ago, and decided adding their personal blessing was the solution. That's literally the entire backstory." Interview stages accumulate like organizational scar tissue.

When questions and scoring change between candidates, an interviewer can't readily tell ordinary differences in answers from signs of assistance, and errors go uncorrected in both directions.

What candidates admit depends on how you ask. Asked directly about live AI assistance during a video interview, 4 of 174 candidates said yes. Asked differently later in the same survey, 17 reported covert AI use and 20 said someone had helped them answer in real time. Question wording changes what people disclose, and anyone interpreting candidate AI statistics, including ours, needs to account for that.

Nobody documents where interview prep ends and the real interview begins. Employers ranked four uses of candidate AI by acceptability: mock interviews first (mean rank 1.91), then preparing answers to practice questions (2.24), assistance during live interviews (2.62) and use on take-home assignments (3.23).

Employers generally viewed preparation more favorably than assistance during the assessment itself. Few had put that preparation-versus-participation line in writing, so two candidates using the same tools can get different treatment from different interviewers.

The Des Moines director's team published theirs on the careers page: "Use whatever you want to prepare. Do not let software answer live questions for you."

Disclosure is becoming law, and what you measure matters. Illinois' rules on AI in hiring are already in force, and New York City's Local Law 144 has required bias audits and candidate notices for automated hiring tools since 2023. Connecticut's notification requirements take effect in October 2027, Colorado's enforcement begins in January 2027, and Ontario pairs its 45-day rule with AI-related requirements.

Recording a human conversation and having AI conduct or evaluate one are different things. Research published in PNAS by Goergen, de Bellis and Klesse (2025), across 13,342 participants, found that people who expected AI assessment emphasized analytical qualities and downplayed intuition because they believed the algorithm would reward it. That finding concerns perceived AI evaluation, not recorded human interviews. The fixes aren't glamorous: disclose how the tools are used before substantive questioning begins, explain what will and won't be analyzed, and make sure actual use matches the explanation.

Candidate misrepresentation is a pervasive market concern, with 83 percent of job seekers reporting widespread embellishment and approximately 50 percent encountering phantom job requisitions or fraudulent recruiters.

Fraud gets caught downstream, not in the interview. Talent leaders most often described discovering misrepresentation during background checks, reference verification and post-offer compliance work, despite their confidence in spotting it live.

A global head of talent at a Dublin technology firm described two verified proxy incidents in engineering:

"Both were caught during onboarding compliance rather than in live rounds, which tells you where the actual risk control lives, and it is definitely not where we thought."

Treating the interview as a real-time lie detector funds a capability nobody has demonstrated; the structured record earns its keep later, when risk teams can revisit what was said.

Organizations are starting to confuse preparation with fraud. As the Dublin technology leader observed: "Hiring managers have started treating polish as suspicious."

Penalizing preparation can disadvantage candidates who use free materials to learn an unfamiliar process, along with multilingual applicants whose practiced answers are more fluent than spontaneous ones.

Structured scorecards can make consistent feedback more practical. When interviewers record evidence against the rubric, software can help pull those findings together, giving recruiters a clearer basis for explaining the decision.

Sixty-eight percent of employers say that in the past twelve months they failed to give a final decision to someone who had completed at least one live interview, and 57 percent of candidates say it happened to them in the past eighteen months. In iHire's October 2025 survey of 1,024 job seekers, 53 percent said an employer had ghosted them during their search.

A talent acquisition partner at a university in Ann Arbor explained how it happens:

"The committee decides in April and nobody is assigned to tell the other four finalists, and then it is June, and telling them in June feels worse than not telling them, so nobody does."

Ghosting usually reveals a gap in ownership, and a clearer workflow closes it by assigning the handoff and checking that the candidate actually got an answer. A requisition can be closed in the system while the candidate is still waiting by the phone.

The Sacramento HR director, whose organization never ghosts yet never provides specific feedback, explains that ‘Every rejected finalist receives a letter that tells them nothing at all, on time, forever.’

Since 1 January 2026, Ontario employers with 25 or more staff have been required to tell every candidate interviewed for a publicly advertised role whether a hiring decision has been made, within 45 days of the last interview. That turns notification into a legal obligation, rather than an afterthought.

A rejection doesn't erase a good interview experience. Candidates named fairness and communication above everything else in judging a hiring process; getting the job ranked last, at 6 percent.

That challenges the convenient assumption that negative feedback is simply rejection talking. The fixes are cheap relative to the cost of getting them wrong: explain the criteria at the start, give useful feedback at the end and assign responsibility for both. The leader cohort reports doing all three at roughly twice the rate of everyone else.

The objection to feedback is usually legal exposure. But the exposure people fear arises when recruiters are expected to explain a hiring decision without knowing what informed it, leaving them to fill gaps in the evidence with assumptions.

The Toronto lead again: “Everyone citing legal risk is describing a perceived risk, not an incident. I have been asking for years whether anybody can name an actual lawsuit that began with honest feedback delivered kindly.”