Candidates Are Interviewing at 1 AM. Your Hiring Process Should Notice.

Roughly a quarter of the AI-run job interviews on one recruiting platform now happen between 10 pm and 2 am. The vendors call that accessibility, and they are partly right. It is also a measurement of how little candidates expect from the other end of the process, and it carries a compliance cost most small employers have not priced in.

“Twenty-four percent of Ribbon’s AI interviews take place between 10 pm and 2 am local time, according to data from the more than 500 companies that use the service. For his manufacturing clients, that figure jumps to 35 percent.”

— Kate Taylor, WIRED

Our take

The convenience framing is true as far as it goes. A parent with kids in bed at ten, or a line cook coming off a double, genuinely could not take a 2 pm call. But notice what else is true: nobody schedules a 1 am conversation with a human being. Night scheduling is what people do when there is no other party in the room.

The vendors’ own numbers undercut the vendors’ story. Greenhouse’s 2026 Candidate AI Interview Report surveyed 2,950 active job seekers and found that after an AI interview, 28 percent advanced to a next round, 13 percent got a formal rejection, and 51 percent received no follow-up at all. The pitch is that AI interviews solve the résumé black hole. Their own measurement says the black hole simply moved one step later in the funnel. We disagree with the framing that candidates face a choice between being ghosted and getting an AI screen. On the published evidence, most of them get both.

The number that should matter most to a small employer is the withdrawal rate. In that same research, 38 percent of candidates said they had already pulled out of a hiring process rather than be interviewed by AI, and another 12 percent said they would. For a company sorting ten thousand applicants, losing a third of the pipeline is a rounding error you can absorb. For a Las Vegas shop hiring one office manager off twenty applicants, it is the difference between a hire this month and a re-post next month.

Disclosure is the cheapest fix available and almost nobody is doing it. Greenhouse found 70 percent of candidates were never clearly told upfront that AI would evaluate them, 21 percent only discovered it after the interview started, and 57 percent believe disclosure should be legally required. One sentence in the job post costs you nothing. Not saying it costs you every candidate who finds out the hard way, and they tell people.

Here is the part the reporting skips entirely: jurisdiction follows the candidate, not the employer. California’s FEHA automated decision system regulations took effect October 1, 2025 and require anti-bias testing, notice, and four-year record retention. Illinois’ HB 3773 notice requirement took effect January 1, 2026. New Jersey issued guidance on automated employment decision tools in December 2025. Colorado’s AI Act is scheduled for June 30, 2026. Nevada has no AI-specific hiring statute of its own. That is not the relief it sounds like. If you run a Henderson business and hire a remote bookkeeper who lives in Riverside, you are running a California screen and you own the California obligations.

Which raises the vendor question. If your applicant tracking system bolts on an AI interviewer, you are still the employer making the decision. Ask for the bias audit and the retention policy in writing before you switch it on. If the vendor cannot produce them, you are the one holding the file when someone asks.

Finally, read the night data as a product requirement rather than a novelty. If a third of applicants for hands-on roles want to interview at 11 pm, the answer is not only an AI agent. It is asynchronous scheduling and real evening slots on a real calendar. Southern Nevada’s workforce is disproportionately shift-based. A 9-to-5-only interview calendar is itself a screen, and the people it screens out are the ones who already have jobs.

Used narrowly, these tools earn their keep. Asking every applicant the same four certification questions is a scheduling problem, and automating it is fine. Scoring “enthusiasm” from a webcam is a judgment problem wearing a scheduling problem’s clothes.

What this means for your business

1. Disclose the AI step in the job post, not at the interview. One line: “The first round is a recorded interview scored by software; a person reviews every result.” Then make the second half true.
2. Send a decision to every candidate who completes a screen. Even a templated rejection beats the 51 percent who currently hear nothing. In a market where you hire from referrals, silence is expensive.
3. Open real evening and weekend interview slots. Put two 7 pm blocks a week on a booking link. You will reach employed candidates that daytime-only competitors never see.
4. Get your vendor’s bias audit and data retention policy in writing before you enable the tool. Keep rubrics, scores, and interview recordings for four years if any applicant lives in California.
5. Automate only questions with verifiable answers. Certifications, licenses, equipment experience, shift availability. Keep judgment calls on a human calendar.

Read Kate Taylor’s original reporting at WIRED.

---