AI Interviewer Efficacy by Role: Where AI Interviews Actually Work

AI interviewer efficacy isn't one question. It's six.

The largest field experiment on the technology randomly assigned roughly 70,000 applicants and found AI-interviewed candidates were 12% more likely to get an offer, 18% more likely to start, and 17% more likely to still be there at 120 days. Meanwhile, a survey of 2,950 active job seekers found 38% had walked away from a process because it required an AI interview. Both numbers are real. They collide only if you treat "AI interview" as one thing across every role.

Those studies measured different populations and different implementations. The field experiment ran on high-volume, entry-level customer service hiring. The abandonment survey pooled every candidate across every role and every botched rollout. Whether AI interviewing helps or hurts depends on the archetype you're hiring for and how you roll it out.

We covered the aggregate case in Do AI Interviewers Actually Work? What the Research Says. This piece breaks it down by role: six archetypes, the evidence behind each, and a plain verdict on where AI interviews earn their keep.

How to read the evidence on AI interviews

Not all evidence carries the same weight. Here's the hierarchy we use, strongest first.

Peer-reviewed field experiments sit at the top. The Jabarian and Henkel study ("Voice AI in Firms," SSRN working paper, 2025) is the strongest available: a randomized design across 70,000 applicants, 48 positions, and 43 firms. The caveat most coverage skips: those roles were high-volume, entry-level customer service positions run through a recruitment-process outsourcing firm, per the Chicago Booth Center for Applied AI writeup. The gains are real, but they're population-specific.

Candidate surveys measure sentiment, not outcomes. The 2026 Candidate AI Interview Report surveyed 2,950 job seekers (1,200 US-based) and tells you how candidates feel: abandonment rates, trust levels, disclosure gaps. It can't tell you whether the hire was actually better.

Vendor-reported metrics come last. Completion rates and time-to-hire cuts from suppliers are useful directional data, but they're marketing until someone else replicates them. We label every vendor figure below accordingly.

Executive takeaway: Match the claim to its evidence tier before you act on it. A vendor completion rate and a randomized retention gain are not the same kind of fact.

AI interviewer efficacy by role: the verdict table

For each archetype below, read the evidence strength first (it tells you how much to trust the verdict), then the failure mode (the thing that breaks the deployment if you ignore it).

• Hourly / frontline (high-volume) — Evidence: Strong (peer-reviewed RCT + vendor case data). Outcome: More offers, more starts, higher early retention; large completion and time-to-hire gains. Failure mode: Impersonal design causing drop-off. Verdict: Deploy.
• Corporate knowledge worker (first round) — Evidence: Moderate (RCT-adjacent + structure research). Outcome: Consistent, comparable first-round signal. Failure mode: Non-disclosure eroding trust. Verdict: Deploy with disclosure.
• Sales SDR (high-volume) — Evidence: Moderate (analogous to frontline structure). Outcome: Communication and structure assessable at scale. Failure mode: Over-indexing on script polish. Verdict: Deploy.
• Healthcare clinical (volume screening) — Evidence: Moderate (credential pre-verification). Outcome: Faster credential and eligibility screening. Failure mode: Automating clinical judgment. Verdict: Deploy for screening only.
• Engineering — Evidence: Contested. Outcome: Skills signal works; candidate trust wobbles. Failure mode: AI-on-AI gaming arms race. Verdict: Pilot, with safeguards.
• Senior AE / executive — Evidence: Weak to none. Outcome: No credible outcome evidence. Failure mode: Withdrawal of scarce, high-value candidates. Verdict: Don't.

Executive takeaway: Deploy where the evidence is strong, disclose where it's moderate, and stay human where it's weak. The verdicts aren't hedges; they track the data tier.

Where the evidence for AI interviews is strong

The favorable case clusters in high-volume, structured, communication-assessable roles. That's exactly the population the strongest study measured.

Hourly and frontline hiring

This is the clearest win. The Jabarian and Henkel field experiment on entry-level customer service at volume produced 12% / 18% / 17% gains in offers, starts, and 120-day retention. Why? AI ran a more structured interview, covered more required topics with more consistent phrasing, and gave human evaluators a cleaner signal to act on.

Vendor-reported data tells the same story. Chipotle cut time-to-hire by roughly 75% and lifted application completion from 50% to 85% while shortening application-to-start from 12 days to four. In high-volume airline hiring, one vendor reports a 93% interview completion rate through a structured chat interview (vendor-reported). Directional, not definitive, but it lines up with the randomized result.

Corporate knowledge worker, first round

For first-round corporate screens, the structure case is solid but the trust case is shaky. A structured AI interview gives you comparable signal across candidates, which is the whole point: consistent triage before a human goes deep. The failure mode isn't accuracy. It's disclosure. 70% of candidates weren't told AI would evaluate them, and 34% came away with a more negative view of the employer. Tell people AI is in the room, and the first-round case holds.

Sales SDR

SDR hiring looks a lot like frontline hiring: high volume, high turnover, and a core competency (structured communication) that AI can assess directly. The evidence is analogous rather than direct, but the mechanism transfers cleanly. One thing to watch: don't over-reward polished script delivery over actual reasoning.

Healthcare clinical, volume screening

For high-volume clinical roles, AI earns its place at credential and eligibility pre-verification. Clinical judgment stays human. We go deeper on this vertical in AI in Healthcare Recruitment: 4 Use Cases.

Executive takeaway: High-volume role plus communication or eligibility competency? Deploy with disclosure and a human final call.

Where the evidence for AI interviews is contested or weak

The favorable verdicts above are credible precisely because the unfavorable ones below exist.

Engineering: an unsolved arms race

Skills assessment through AI works in principle, but engineering hiring has a problem nobody has solved yet: candidates use AI to beat AI interviews. When both sides run language models, the interview measures prompt fluency as much as engineering judgment. Add candidate skepticism from a labor pool with options, and you've got an arms race with no stable equilibrium. If you pilot, pair it with live human follow-up and anti-gaming safeguards, and don't treat the signal as clean.

Senior AE: small pools, high withdrawal risk

Senior AE hiring inverts the frontline math. The pools are small, the evaluation hinges on relationship nuance that structured interviews flatten, and these candidates have the leverage to walk. The 38% abandonment figure from the 2026 Candidate AI Interview Report concentrates here. One strong senior AE quietly dropping out costs you more than any efficiency gain.

Executive: an evidence-free zone

No credible outcome evidence exists for AI-interviewing executives. Jabarian's team flagged that voice AI is less suitable for roles centered on crisis management, creative problem-solving, and thought leadership, per the Chicago Booth Center for Applied AI. Those are the defining competencies of executive work. Don't.

Executive takeaway: Scarce pool, judgment-heavy competency, no offsetting evidence. Keep it human.

Why the same role produces opposite results

Implementation quality resolves most of the contradiction. You can point the same AI interviewer at the same role and get the field experiment's gains or the survey's abandonment, depending on four design choices.

Disclosure. The 70% non-disclosure rate documented in the 2026 Candidate AI Interview Report is the category's biggest own-goal. When you don't tell candidates, AI reads as deception. When you do, it reads as process.

Choice. When applicants were offered a choice, 78% chose the AI interviewer. Agency flips the experience from something done to candidates into something they opted into.

Failure handling. Even in the favorable experiment, 3.2% of candidates quit from AI aversion and 8% from a system failure. System failure drove more than twice the drop-off of aversion. A clean human fallback path is where most of your avoidable loss lives.

Human review. A guaranteed human decision at the end keeps the screen defensible and keeps hiring managers from bolting on extra interviews they don't trust.

One more finding worth noting: in the same experiment, AI interviewing nearly halved gender discrimination in interview scoring. When you instrument it well, AI compresses bias instead of scaling it. If your rollout has stalled despite good tooling, the fix is almost always in these four variables. See AI interviewer best practices.

Executive takeaway: Same tool, same role, opposite outcomes. These four design variables are what separate the field experiment's gains from the survey's abandonment.

Deploy where the evidence lives

Deploy AI interviews for hourly and frontline hiring, first-round corporate screens (with disclosure), high-volume SDR pipelines, and clinical volume screening. Pilot cautiously in engineering. Keep senior AE and executive hiring human.

Humanly is built for the archetypes where the evidence is strongest: high-volume hourly and first-round corporate. Disclosure and a guaranteed human handoff are defaults. You get chat, SMS, and voice screening, native ATS integrations, and a built-in bias-audit layer so the fairness gains are measured, not assumed.

See how Humanly deploys AI interviews where the evidence supports them.

FAQs

Does AI interviewer efficacy depend on the type of role?

Yes. The strongest evidence, a 70,000-applicant field experiment, comes from high-volume, entry-level customer service hiring, where AI-interviewed candidates saw 12% more offers and 17% higher 120-day retention. For judgment-heavy roles like senior AEs and executives, there's no comparable outcome data, and withdrawal risk is high. Efficacy tracks the archetype, not the technology.

Where do AI interviews work best?

High-volume, structured, communication-assessable roles: hourly and frontline positions, first-round corporate screens, SDR pipelines, and clinical volume screening. In these roles, consistent signal at scale beats bespoke human triage, and both the field-experiment and vendor-reported evidence point the same way.

Why do so many candidates abandon AI interviews?

The 2026 Candidate AI Interview Report found 38% of US job seekers abandoned a process because it required an AI interview, and 70% weren't told AI would evaluate them. Walk-aways concentrate among candidates with options. Disclosure, format choice, and reliable systems with a human fallback sharply reduce the drop-off.

Are vendor-reported AI interview statistics trustworthy?

Useful directional signals, but weight them below peer-reviewed research. Chipotle's reported 75% time-to-hire reduction and one vendor's 93% completion rate align with the randomized evidence but haven't been independently replicated. Always check whether a statistic is peer-reviewed, survey-based, or vendor-reported before acting on it.

Should executives be interviewed by AI?

No. There's no credible outcome evidence for AI interviews at the executive level, and the researchers behind the strongest pro-AI study note that voice AI is less suitable for crisis management, creative problem-solving, and thought leadership. Keep executive and senior sales evaluation human.