What 2,800 Job Seekers Tell Us About Trusting AI Interview Screens

A candidate clicks into an interview link and finds no human on the other end, just a voice asking the next question. Maybe they are relieved it is not another resume form. Maybe they are bracing for a script that will not adjust no matter what they say.

Both reactions are fair. Hiring is stuck in what Greenhouse calls an AI doom loop, and the data on who is stuck in it with whom is clearer than the recruiters running these screens tend to assume.

What the survey found

In October 2026, Greenhouse published findings from its 2026 Future of Hiring Candidate Report, based on a survey of more than 2,800 job seekers across the U.S., U.K., Ireland, Germany and Australia. The report was written by Tim Song and published on the Greenhouse blog on October 20, 2026.

The headline numbers challenge the assumption that candidates grit their teeth through AI screens. Nine in ten U.S. candidates say they are open to an AI interview. Only 7 percent would refuse one outright. Candidates are 2.7 times more likely to take an AI interview than to withdraw from a process where no human interviewer is available.

Nearly half, 48 percent, have already taken an AI interview in the past year. Of those, 70 percent say it let them demonstrate their skills well.

The friction sits in the execution. Forty-nine percent of U.S. candidates want to know a human will review their responses. Forty percent want to know how they will be evaluated. Eighty-eight percent want at least one legal requirement that makes employers upfront and accountable about AI in interviews, with advance notice the leading ask at 55 percent.

Sharawn Tipton, Chief People Officer at Greenhouse, framed the problem in the report’s accompanying blog post: candidates need a system that actually works and gives them a fair shot, and bolting bad AI on top only hurts them and the employers doing it.

The rejection problem underneath

One statistic from the same report explains why the design question matters so much. Eighty-one percent of candidates across the five markets Greenhouse surveyed were rejected from a job in the past year without speaking to anyone at the company, human or AI.

Among U.S. candidates specifically, 96 percent wish they could show employers something their application simply does not capture. More than half, 52 percent, point specifically to their ability to learn and adapt.

That is the gap an AI screen could close, if it were built to close it. A well-designed conversational screen can ask a candidate to talk through a real problem, follow up on a vague answer, and surface what a resume cannot. A poorly designed one asks the same question regardless of what the candidate said, scores tone or confidence instead of content, and tells the candidate nothing about how they will be evaluated.

The difference between those two outcomes is a design choice, and the framework for understanding it has been sitting in the selection research literature for three decades.

What the selection literature predicted

In 1993, Stephen W. Gilliland published “The Perceived Fairness of Selection Systems: An Organizational Justice Perspective” in the Academy of Management Review. The paper has been cited more than 2,500 times since, and it laid out the procedural rules that shape whether candidates see a selection process as fair.

Gilliland’s argument was that candidates judge a selection system by the process, not by whether they got the job. They look at whether the system gave them a chance to show what they can do, whether the questions were job-relevant, whether the system treated them with respect, whether they could ask questions and get answers, whether the outcome was explained, and whether the process was consistent across candidates.

Those are not abstractions. They map almost directly onto what the Greenhouse report found candidates asking for in 2026. Human review of responses is an accountability mechanism. Advance notice about how they will be evaluated is a transparency rule. Both are procedural fairness rules Gilliland named thirty years before the first AI screen shipped.

The selection research literature has spent thirty years showing that procedural fairness predicts whether candidates recommend an employer, accept an offer, and stay in the pipeline. The Greenhouse data suggests the same rules apply to AI screens, because an AI screen is a selection procedure, and candidates react to it the same way they react to any other.

What changes when you treat the screen as a selection procedure

The practical implication is that the design questions for an AI screen are the same ones the selection literature has been answering since Gilliland. A well-designed screen gives the candidate a chance to show job-relevant skills rather than asking everyone the same surface-level questions. It follows up on vague answers instead of moving on regardless. It makes clear that a human will review the responses, so the candidate is not talking into a black box. And it stays consistent across candidates, rather than drifting based on accent, tone or pacing.

Greenhouse built its own AI screening product, Voice AI, around those constraints. The blog post describing the research is also a product page, and the design choices it names are straight from the fairness playbook: dynamic follow-ups, full transcripts, human review on every decision, and a stated policy of not scoring tone, accent or confidence.

That is one vendor’s approach, and one vendor’s research. The broader point is that the fairness rules are not proprietary. Any employer running an AI screen can apply them, and the selection research says the ones who do will see better candidate reactions, higher acceptance rates, and less reputational damage from a process that is, for most candidates, their first substantive interaction with the company.

The cost of getting it wrong

The Greenhouse report makes one more point that ties the design question to employer brand. A poorly designed AI screen is often a candidate’s first interaction with a company, and the research on procedural justice says candidates generalize from that interaction to the whole employer. A screen that feels arbitrary, opaque or disrespectful costs more than the candidate in front of it. It costs the next ten candidates who hear about it.

The survey data says the opening is there. Most candidates are willing to take an AI interview. The ones who have taken one mostly say it worked. The ones who have not are asking for specific, reasonable design choices before they will trust the process.

The recruiters who treat those design choices as a checklist from a thirty-year-old framework are the ones most likely to close that gap. The ones who treat AI screening as a technology procurement problem are the ones most likely to widen it.

Sources: Greenhouse 2026 Future of Hiring Candidate Report, described in Tim Song, “Good conversational AI interview screening, according to candidates,” Greenhouse Blog, October 20, 2026 (greenhouse.com/blog). Gilliland, S. W. (1993), “The Perceived Fairness of Selection Systems: An Organizational Justice Perspective,” Academy of Management Review, 18(4), 694.