38% of Your Candidates Walk Away When AI Runs the Interview

In the past twelve months, 63% of candidates in the United States experienced an interview conducted by artificial intelligence. However, 38% of them dropped out of the process specifically because of this. These figures come from the same study, yet most companies only focus on the first statistic.

How many candidates abandon a process due to an AI interview?

Nearly four out of ten. Specifically, 38% of U.S. candidates withdrew from a selection process because it included an AI-conducted interview, and another 12% stated they would do so if it were mandatory (Greenhouse, 2026 Candidate AI Interview Report, based on 2,950 active job seekers in the United States, United Kingdom, Ireland, Germany, and Australia).

Exposure to AI interviews has grown rapidly: 63% of U.S. candidates went through an AI interview in the past year, a twelve-point increase from six months earlier. This means the format is now prevalent, and so is the rejection.

It’s important to clarify what we mean by “AI interview.” It encompasses asynchronous video (where candidates record responses alone, without anyone on the other side, and a model scores them), real-time chat or voice bot interviews, and automatic monitoring during the process. These are very different experiences, and they don’t all cause the same level of discomfort.

Which format is truly driving candidates away?

Pre-recorded video without a human present. According to the same Greenhouse report (2026), here are the reasons candidates gave for abandoning:

Reason for Abandonment	What It Automates	% Mentioning It
Pre-recorded video scored by AI, no human present	Initial recruiter call	33%
Lack of clarity on AI usage	Transparency issue	27%
Automatic monitoring during the process	Human supervision	26%
Mandatory AI-conducted interview (chat or voice)	Screening interview	26%

Notice the second row. It’s not a technology problem; it’s a notification issue. 70% of candidates who went through an AI interview say they weren’t clearly informed, and 20% only found out when the interview started.

Another telling statistic: only 19% of candidates want less AI in processes. 21% want the same level of AI but with more transparency, and 22% want more AI, with visible human oversight. The issue isn’t automation itself; it’s silent automation.

Why doesn’t time-saving show up in the final results?

Because time-saving is measured, but candidate drop-off isn’t.

Talent teams see the visible half: those using AI report saving about 20% of their workweek, roughly an entire day per person (LinkedIn, survey of over 1,000 talent professionals, 2026). This figure makes it into the monthly report without question.

The other half doesn’t appear in any report because there’s no record of the candidate who opened the link, saw “record your response in 15 minutes,” and closed the tab. In the recruitment funnel (the journey from application to signing), this person never appears as a loss; they appear as someone who was never there.

Friction due to length was measured before AI: applications under five minutes are completed 12.47% of the time, while those over fifteen minutes are completed 3.61% of the time (Appcast benchmark, 2025). A twenty-minute asynchronous interview is, in terms of friction, a lengthy application placed right when the candidate hasn’t yet invested emotionally in your company.

There’s also a selection bias that worsens the situation. In the processes we manage at Connecting People for tech roles in LATAM, the first to drop out is never the profile that needs the job; it’s the senior candidate with three other ongoing conversations and no reason to record themselves in front of a camera at eleven at night. Automation filters by tolerance to friction, not by talent. It’s a real filter, but it measures what no one cares to measure.

How much does it cost your employer brand when a candidate discovers AI on their own?

More than it seems, and it lasts longer than the process.

A Gartner survey of 2,918 candidates (first quarter of 2025) found that only 26% trust AI to evaluate them fairly, 32% fear a model will discard their application, and 25% trust the employer less when they learn AI is evaluating them. That last figure is costly: it doesn’t describe an opinion about the tool; it describes a discount on your company.

Then there’s the end of the process, where things really break down. 51% of candidates who completed an AI interview never received a result, and 38% never received any response at all (Greenhouse, 2026). Asking someone for twenty minutes of video and then not responding isn’t a technical system failure; it’s a process decision someone made, or didn’t make, leaving it as is.

What distinguishes a successful AI process from one that drives candidates away?

Four things, none of which involve turning off automation:

1. Notify beforehand, not during. Clearly state in the invitation which part of the process involves AI and what will be done with that information. This is what the 27% who abandon due to lack of notification are asking for.
2. Explain what it measures. 39% of candidates want to know what the model evaluates. A single line suffices.
3. Show the human element. 38% want confirmation that a person reviews the output before any decision, and 46% want the option to request a human interview instead of an automated one. Offering this option costs little and retains the profiles you most want.
4. Close the loop. Respond to everyone who completed the interview, even if it’s a no. Currently, half receive nothing.

None of this is a philosophical stance on AI. It’s recognizing that only 21% of candidates believe employers are using it responsibly and transparently, and in a market where good engineers have choices, that perception is part of the cost of hiring them.

Automating screening addressed a real bottleneck: no one wants to manually sift through 800 resumes. But it was implemented as if the candidate were merely an input in the process rather than the other party in a negotiation. As long as the savings are visible on a dashboard and the drop-off isn’t recorded anywhere, the numbers will continue to look good in reports but bad for the team.