38% of US candidates have walked away from AI interviews. Here's what the data actually says.

The stat travels fast, usually without its source or its context: 38% of US candidates have already withdrawn from a hiring process because it included an AI interview. It comes from Greenhouse's 2026 Candidate AI Interview Report, published April 2026 — a survey of 2,950 job seekers across the US, UK, Ireland, Germany, and Australia.

If you're using AI anywhere in screening — we build an AI interviewer, so that includes our customers — this number deserves a closer read than the headline. Because the same report says something the headline version leaves out.

What the report actually found

• 38% of US candidates have already withdrawn from a hiring process because it included an AI interview — and another 12% say they would if required.
• 63% of candidates have now done an AI interview — up 13 points in six months. This is no longer an edge case; it's the front door of hiring.
• 70% weren't clearly told upfront that AI would evaluate them.
• The top US reason for walking away (33%): a pre-recorded, AI-scored video with no human present. Second: the company didn't disclose how AI would be used (27%).

Read those last three together and the story changes. Candidates aren't boycotting a technology — most of them are encountering it constantly and completing the process. They're walking away from a specific experience: talking at a camera, scored by something undisclosed, with no conversation and no transparency.

The format is the message

A pre-recorded one-way video gives the candidate nothing a conversation gives: no follow-up question when their answer was almost there, no ability to ask what the role actually involves, no evidence anyone is listening. It optimizes entirely for the company's time and asks the candidate to absorb all of the awkwardness. Candidates have noticed — 33% of US walkaways name exactly this.

Non-disclosure compounds it. A candidate who finds out afterward that an algorithm scored them has every reason to assume the worst about how. The 70% non-disclosure figure isn't a compliance footnote; it's most of the trust problem.

What a candidate-respecting AI interview looks like

We'd propose these as the testable design standards — for our product or anyone's:

1. Disclose upfront, in plain language. Before the interview: it's an AI interviewer, here's what it evaluates, here's what happens with the recording. (In Illinois this is law — AIVIA consent requirements tightened in January 2026 — but it should be the default everywhere.)
2. Two-way beats one-way. A real conversation with follow-up questions, where the candidate can also ask things, is categorically different from dictating into a void. It's also better screening: you learn what someone thinks when their first answer gets probed.
3. Show the work. Every interview produces a full transcript the recruiter sees, and the score traces to what the candidate actually said — content-only evaluation, no tone or facial inference.
4. A human owns the decision. The AI's job is the structured first conversation and a scored report against your rubric; the hire/advance decision stays with the recruiter.

That's the design we build to at Vettika. We're deliberately not quoting our own completion or satisfaction numbers here — our measurement instrumentation isn't live yet, and we don't publish experience claims we can't show methodology for. (The same report finds 38% of candidates came away more positive after a well-run AI interview — a different 38%, easy to conflate with the headline one. Done well, this format earns trust; done opaquely, it leaks pipeline.)

What to do with this as a recruiter

• Audit your current screen: would a candidate describe it as a conversation or a deposition?
• Check your disclosure: is the AI's role stated before the candidate commits time, or discoverable afterward?
• Ask any vendor — us included — for transcripts of real interviews and for how their scores are derived. If the answer is a black box, the 38% is your forecast.

All figures from the Greenhouse 2026 Candidate AI Interview Report (April 2026, N=2,950, five countries), verified against the primary source June 11, 2026.