Do Candidates Actually Like AI Interviews? What the Data Says

Candidates don't have one opinion about AI interviews, because "AI interview" isn't one thing. The survey data shows reaction swings hard on four observable design choices: whether AI was disclosed before the invitation, how much effort the format demands, whether a named human is accountable for the decision, and how fast the candidate hears back. Get those right and an AI screen can improve how people see your company. Get them wrong and you lose candidates before they answer a single question.

The data on whether candidates like AI interviews

The most useful recent numbers come from Greenhouse's candidate AI interview research, a survey of 2,950 active job seekers across five countries, including 1,200 US respondents. Read it as stated experience and perception, not as a measure of whether any system was actually fair.

Among US respondents, 63% said they had been through an AI interview. Of those who had been evaluated by AI, 70% said it was not clearly disclosed beforehand, and 21% only found out after the interview had started. So most candidates who got screened by AI were never told in advance. The survey doesn't establish that surprise is the main reason people walk away, but non-disclosure is the cheapest thing on the list to fix.

The withdrawal numbers are the part that should get a recruiter's attention. 38% said they had dropped out of a hiring process because it included an AI interview, and another 12% said they would. But only 19% wanted less AI in hiring overall. Those two numbers only fit together if candidates are rejecting specific implementations rather than the category.

What they asked for was procedural, not ideological. 46% wanted the option to request a human interview, 44% wanted disclosure upfront, 39% wanted an explanation of what the AI measures, and 38% wanted confirmation that a human reviews the AI's evaluation. None of that requires abandoning automation.

The withdrawal triggers rank in a specific order. Pre-recorded video scored by AI with no human present was the largest listed trigger at 33%, followed by no disclosure of how AI would be used at 27%, AI monitoring during the interview at 26%, and a required AI-led interview with no alternative at 26%. Format burden edges out disclosure, and both matter.

And a well-run AI interview isn't neutral. It improved employer perception for 38% of respondents. A poor one worsened perception for 34%. Roughly symmetrical upside and downside, decided by execution.

Four different things get called the same name

Lumping these together is why so much advice on this topic is useless.

Asynchronous AI voice interviews ask candidates to answer set questions aloud, on their own time, with no camera and no live call. Low burden, no scheduling, no appearance anxiety.

One-way video interviews ask for recorded video answers. Higher burden: candidates need decent light, a tidy background, a quiet room, and the confidence to watch themselves talk. This is the format tied to the largest listed withdrawal trigger in the data.

Live conversational AI puts the candidate in a real-time exchange with a bot. Novelty helps some candidates and unsettles others, and there is no chance to compose an answer.

AI-only decision-making is the one candidates genuinely reject. Pew Research Center's work on AI in hiring found 71% of Americans opposed an AI system making a final hiring decision, with 7% in favor and 22% unsure. That's a general attitude toward decision authority, not a verdict on any interview format, and it's the clearest signal in the whole body of research: people accept AI in the process, not at the end of it.

There's academic support for the mechanism too. A study published in Humanities and Social Sciences Communications found that highly automated AI-enabled interviews reduced perceived procedural justice, organizational attractiveness, and intention to apply. The stated reason matters more than the effect size: applicants react badly when a process looks incapable of recognizing them as individuals or explaining how their answers were scored. That's perception and application intention in a study setting, not a real-world drop-off rate you can copy into a forecast.

AI interview acceptance risk matrix

Use this to audit an existing process. Each row is a choice you either already made or made by default.

The pattern across the rows: candidates trade effort for flexibility fairly willingly. What they won't trade is knowing what's happening and who is accountable.

The silence problem is bigger than the format problem

Here's the number that should reorder your priorities. Among candidates who completed an AI interview, 38% never heard back and another 13% were still waiting when the survey ran. That's 51% who had not received an outcome at the time of the survey. 28% advanced, and 13% got a formal rejection.

Think about what that does. You asked someone to talk to a machine for ten minutes, and then you disappeared. Every suspicion they had about being processed by a system that doesn't care gets confirmed, and they tell people. No amount of thoughtful question design survives that.

Ghosting after an AI interview is worse than ghosting after a resume submission, because effort creates expectation. If your team can't commit to responding to everyone who completes a screen, automating the screen will make your reputation worse, not better. Automated rejection emails triggered by stage exit are a two-hour setup. There is no good excuse here.

This is also where automation genuinely helps the candidate side. If AI review compresses your screening turnaround from nine days to two, that speed is the candidate experience, and it's the part of the pitch you can actually keep. Most of the other candidate experience fundamentals still apply unchanged.

A candidate invitation template that prevents most withdrawals

This covers all four things candidates asked for in the survey data. Adapt the specifics, keep the structure.

Subject: Your next step for [Role] — a 10-minute AI voice interview

Hi [Name],

Thanks for applying to [Role] at [Company]. Your next step is a short

first-round interview conducted by an AI voice interviewer.

What to expect:

- 6 questions, about 10 minutes total, audio only

- No camera, no live call, no scheduling

- Do it from your phone or laptop whenever suits you, before [date]

- Nothing else on your screen or device is recorded

What we evaluate:

Your answers on [specific area 1], [specific area 2], and [specific

area 3]. The AI produces a written summary and a score against those

areas. It does not make the decision.

Who reviews it:

I do. I read every summary alongside your application before deciding

who moves forward. [Recruiter name], [title], [email].

If this format doesn't work for you:

Reply to this email and we'll book a phone call with me instead. That

applies to accessibility needs, accommodation requests, or simply a

preference for a live conversation. Your answer doesn't affect how we

assess you.

When you'll hear back:

By [specific date]. Either way, you'll get a reply from me.

[Link]

Two details do most of the work. The named recruiter, because "a human reviews the AI evaluation" means nothing without a person attached to it. And the last line, because a date you actually hit is the difference between a good and bad experience.

If you offer a human alternative, do not ask candidates to justify the request. The moment a request needs justification, most people stop making it, and you've built an opt-out that doesn't function.

Edge cases that decide whether this works

The averages hide most of the useful signal. These are the situations where AI screening either clearly wins or clearly backfires.

The hourly applicant finishing a shift at 11pm. This is the strongest case for an AI job interview. They can't take a recruiter call during work hours, they've been burned by phone tag before, and a ten-minute voice screen from the parking lot is a genuine convenience. In high-volume retail, warehouse, and support hiring, async formats often see better completion than scheduled calls, largely because there's nothing to schedule. Check your own numbers against completion rates by screening channel before assuming it holds for your funnel.

The senior passive candidate you approached. Wrong tool. Someone who wasn't looking, who you contacted, and who agreed to a conversation reads an AI screening invitation as a status insult. They're evaluating you as much as you're evaluating them, and you just told them a bot handles that. Have the hiring manager call them.

The accommodation request. Speech differences, hearing impairment, anxiety disorders, and non-native accents all interact with voice interviews in ways your vendor's demo doesn't cover. You need a route to a human that a candidate can take without disclosing a diagnosis, and you need someone on the team who owns that request when it arrives. Don't let it land in an unmonitored inbox.

The candidate surprised by the camera. If the invitation says "interview" and the link opens a camera, you've lost people who were mid-way to a yes. Same for anything that records the screen or tracks eye movement. Whatever monitoring exists must be named in the invitation, in plain words, before the click.

The candidate who wants to know why they were rejected. AI-generated summaries make this easier than most recruiters expect, because you have structured notes on every candidate instead of half-remembered call impressions. Being able to say which competency was thin is a real advantage of structured AI candidate screening, and it's worth using rather than hiding behind a template rejection.

A go/no-go rule for AI interviews

Run any role through this before turning on automation.

Go when all four hold: the screen is short and structured, speed or scheduling flexibility genuinely helps this candidate pool, you have named a human who reviews every AI evaluation before a decision, and you can commit to an outcome date for everyone who completes it. High-volume, high-applicant-count, entry to mid-level roles almost always clear this bar.

No-go, or offer a human route by default, when any of these apply: the role is executive or relationship-heavy, the candidate is passive and you initiated contact, the pool skews toward accommodation-sensitive needs, the hire is trust-critical enough that your process is part of the sell, or your team can't guarantee a reply to every completed screen.

The pass/fail question underneath all of it is accountability. If you can't name the person who reads the AI's output and owns the decision, you're not running an assisted process, you're running an automated one, and that's the version candidates reject. The boundary between what an AI interviewer does and where humans decide should be written down before launch, not improvised after the first complaint.

For teams that want the flexibility without the camera burden, asynchronous audio-only screening sits in the lowest-friction corner of the matrix. Kira AI works this way: candidates answer role-specific questions aloud on their own time from any device, no live call and no camera, and recruiters review AI summaries and structured scorecards while keeping the advance or reject decision on their side of the line. The format removes the two things candidates complain about most, scheduling and being on video, without removing the human from the decision.

Key Takeaways

• Candidates reject specific AI interview implementations, not AI in hiring. Only 19% of surveyed US job seekers wanted less AI overall, yet 38% had withdrawn from a process because of an AI interview.
• Non-disclosure is the cheapest problem to fix. 70% of candidates evaluated by AI said it wasn't clearly disclosed before the interview, and 27% named non-disclosure as a reason they withdrew.
• Format burden ranks the options. Async voice sits lowest, pre-recorded video scored with no human present was the largest listed withdrawal trigger at 33%, and AI-only decision authority draws 71% opposition in general public polling.
• Silence after completion causes more damage than the format ever will. Among candidates who completed an AI interview, 51% had not received an outcome at the time of the survey.
• Fix the invitation before you fix the questions: disclose AI, state the time and format, say what's evaluated, name the human who reviews it, offer a no-questions-asked human alternative, and commit to a reply date.
• Skip AI screening for executive, passive-sourced, accommodation-sensitive, or trust-critical hires, and skip it entirely if you can't promise every candidate an answer.