By Lee Flanagan
✨ AI Summary:
- 38 percent of American candidates withdraw from hiring processes to avoid AI interviews—a trust gap that expanded scheduling hours cannot fix.
- AI interview footage often goes unwatched by humans, leaving decisions based solely on variable, opaque rubrics that candidates cannot understand or challenge.
- Offering 24/7 scheduling flexibility for AI screening optimizes availability, not accountability; it moves the credibility problem to a later hour.
- Hiring leaders must assign human accountability for AI scores and communicate clearly to candidates how and why they are being evaluated.
Tim Millard tidied his living room at 9:30pm, set up his laptop, and waited for a question to appear on screen. He had thirty seconds to think. Then a ding, and two minutes to record his answer to a camera that would never talk back. “It felt so out of this world,” he told WIRED. He got an automated rejection email two months later. The interview email had left the timing open-ended, and Millard figured he might as well get it done as soon as possible. “What else was I going to do at 9:30 at night?”
That scheduling detail is the whole story the flexibility framing leaves out. The rise of the 1am job interview is being sold as convenience. Look at the numbers behind it and you find a candidate pool that is not delighted by AI interviews. It is opting out of them.
What the 10pm to 2am window actually measures
Ribbon, the voice-AI recruitment platform, reports that 24 percent of the AI interviews run on its service happen between 10pm and 2am local time, rising to 35 percent among its manufacturing clients. Greenhouse says 15 to 20 percent of candidates interviewed by its voice agent now book overnight slots. Ribbon cofounder Arsham Ghahramani frames this as access: parents who cannot spare thirty minutes before their kids are in bed, hourly workers, people whose current job eats every daylight hour. “This is a genuinely new option for people,” he says.
Ribbon and Greenhouse both sell the software generating these numbers, so their framing deserves scrutiny rather than repetition. Access is a real thing to measure and a real problem to solve, but it answers a different question than the one candidates are actually raising. None of the candidates WIRED interviewed complain that AI interviews are scheduled at inconvenient hours. The complaint is not about time slots. It is about wanting a human being on the other end of the conversation, or at least clarity about what is doing the judging.
The withdrawal numbers nobody is leading with
Greenhouse’s own May survey found that nearly two-thirds of job applicants have now been interviewed by an AI agent, up 13 points in six months. That is fast adoption by any standard. The same survey found 38 percent of American candidates had withdrawn from a hiring process specifically to avoid an AI interview, and another 12 percent said they would drop out if one were required. Read those two findings side by side and adoption looks less like acceptance and more like exposure outpacing tolerance.
A scheduling fix cannot close that gap. Widening the hours a candidate can access an AI interview does nothing to address why more than a third are walking away from the format itself. Our view is straightforward: you do not solve a trust deficit by making the untrusted thing more available. You solve it by making the thing itself legible.
The black hole candidates are actually describing
Greenhouse’s Ophir Samson says he does not think AI should replace human interviewers. He also pitches the AI interview itself as a résumé add-on, a way to surface strong candidates who would otherwise vanish into an applicant pool competing on AI-written CVs. “It’s a shitty time to be a candidate,” he says. “Your résumé goes into a black hole. You’re competing with thousands of candidates that are all using AI to write their résumés. So, good candidates can’t stand out.”
He is not wrong about the black hole. What his own data undercuts is the first claim. Footage from these interviews is passed to the recruiter along with the AI-generated score, but WIRED reports that the footage routinely goes unwatched by humans. Rubrics vary by role: a welder is assessed against technical certifications, a sales candidate is scored on an “enthusiasm” metric, the example Ghahramani himself gives. A system built on an unwatched tape and a variable rubric starts to look exactly like the black hole Samson describes, just with an AI badge on the door instead of an ATS logo.
Availability without accountability is the wrong trade
In our work with hiring teams, the instinct when candidate volume overwhelms a process is to reach for more automation and more scheduling flexibility. That instinct treats screening as a throughput problem, and it is not one. Widening the hours a candidate can be evaluated by an unwatched, variably-scored system does not make the system more defensible. It just makes it more available. If a rubric cannot survive a candidate asking “who is checking this?”, adding a 1am slot does not fix that. It just moves the anxious silence to a later hour.
Hiring leaders do not need to abandon AI screening. They need to keep a human accountable for what the score means, and give candidates a way to find out how it was reached. Millard’s closing line to WIRED is worth sitting with: “It feels like if you are going to scream, you’re going to scream into the void. And nobody is going to hear you.” Ask yourself whether your process could survive that same complaint. Every hour added to the interview window without answering it is another hour spent optimizing the wrong variable.
Original reporting: WIRED.
Frequently asked questions
What actually happens during one of these AI interviews?
Per WIRED’s reporting, a candidate answers timed questions on camera, or through voice AI software, with no human interviewer present, and an AI system scores the responses against a role-specific rubric. Both the score and the footage are typically passed to the recruiter, but the footage often goes unwatched, so the score ends up carrying the decision.
Does scheduling AI interviews for 10pm to 2am actually widen access, as Ribbon claims?
Ribbon’s cofounder frames the window as helpful for parents and hourly workers who cannot interview during the day, and Ribbon’s own usage data, 24 percent of interviews overall and 35 percent for manufacturing clients, shows real uptake in that slot. That addresses when candidates can interview, not the separate trust problem the same reporting documents.
Why are so many candidates withdrawing from AI interview processes?
Greenhouse’s May survey found 38 percent of American candidates had withdrawn rather than face an AI interview, with another 12 percent saying they would drop out if one were required. The reporting ties this to a lack of clarity around how scoring works and what happens to the footage, rather than to interview timing.
If interview footage often goes unwatched, what is the hiring decision actually based on?
Largely on the AI-generated score, since footage is passed to the recruiter but frequently not reviewed by a human. That gap between what gets scored and what gets checked is the real credibility risk, not the hour the interview happened.
Should hiring teams stop offering overnight AI interview slots?
No. Scheduling flexibility is not the problem here. What needs fixing is making sure a human stays accountable for how scores are generated and that candidates are told clearly how they are being judged, whatever time the interview happens.