Candidates interviewed by AI voice agents received 12% more job offers, were 18% more likely to actually start the job, and stayed 17% longer than those interviewed by a human recruiter for the same role. That is the headline finding from a peer-reviewed field experiment covering more than 70,000 real job applicants, published by economists Brian Jabarian of University of Chicago Booth and Luca Henkel of Erasmus University Rotterdam, and independently covered by Bloomberg and McKinsey.
This article covers two things: what the research actually found and why it surprised the people closest to the process, and what it changes about how you should prepare for an interview that may not have a human on the other end of it.
A Result Almost Nobody Expected, Including the Experts
Before the results came in, the researchers asked working recruiters what they thought would happen if AI conducted the interviews instead of people. The consensus was clear and confident: only 15% thought AI could even match human performance, let alone beat it. Recruiters were not being dismissive for no reason. Conducting a good interview is widely seen as a deeply human skill, built on rapport, intuition, and the ability to read a room.
The study was designed carefully enough to make the result hard to dismiss. More than 70,000 applicants across 48 different job postings for Fortune 500 companies were randomly assigned to one of three groups: interviewed by a human recruiter, interviewed by an AI voice agent, or given a free choice between the two. Critically, the interview questions were standardised across every group, and in every case, the final hiring decision was still made by a human recruiter reviewing transcripts and recordings. The AI was never deciding who got hired. It was only conducting the conversation.
Why the AI Interviews Actually Performed Better
The explanation the researchers found is not that the AI was a better judge of character. It is that AI-led interviews extracted more usable information for the human making the final call. Analysing the transcripts, AI-conducted interviews covered an average of 6.8 distinct topics relevant to the role, compared with 5.5 in human-led interviews. Because the AI ran a more structured but still responsive conversation, with less variance from one candidate to the next, recruiters reviewing the resulting transcripts had more consistent, comparable material to assess, and they scored the AI-led interviews higher as a result.
One additional finding is worth sitting with on its own. Gender discrimination in hiring outcomes dropped by nearly half in the AI-led interviews compared with the human-led ones. The structured, standardised nature of the conversation appears to have reduced the inconsistency that lets unconscious bias creep into less structured human interviews.
What Candidates Actually Thought of It
The part of this research that contradicts the usual assumption most directly is how candidates themselves responded. When given a genuine choice between an AI interviewer and a human one, 78% chose the AI, most commonly citing convenience and lower anxiety. Candidates rated their experience in the AI interviews about as positively as those in human-led ones, with no meaningful wave of complaints about talking to a machine instead of a person.
There were real limitations worth being honest about. About 5% of applicants stopped an AI interview partway through, and 7% ran into a technical glitch that added stress to the process. And among candidates who actively chose the AI option when given the choice, average standardised test scores were somewhat lower than among those who chose a human interviewer, suggesting the preference for AI may correlate, at least partly, with candidates seeking a process they find less intimidating.
What This Means If You Are the One Being Interviewed
The instinct many job seekers have toward AI in hiring is defensive: figure out how to get past it, the same way you would approach an ATS keyword filter. This research suggests that posture is the wrong one specifically for AI-led interviews, because the mechanism producing better outcomes is not a trick to be beaten. It is structure and consistency, and you can prepare directly for both.
An AI-led interview asks a standardised set of role-relevant questions and is built to extract specific, comparable information rather than to be charmed or impressed. The candidates who perform best in that structure are the ones who answer with specific, concrete evidence rather than broad, general claims. If a structured question asks about a time you handled conflict, the answer that performs well is the one with a real situation, a real decision, and a real outcome attached to it, not a generalised description of your conflict-resolution philosophy. The more topics you can speak to with genuine, specific detail, the more usable information the system has to pass along to the human who ultimately decides.
This means the most useful preparation is not guessing what a machine wants to hear. It is making sure your own career evidence, the specific projects, decisions, and outcomes that actually happened, is organised and ready to be retrieved in detail under interview conditions, structured or not.
How Candoorai Prepares You for This
Candoorai's interview preparation module is built around the same principle this research validates: specific, structured evidence consistently outperforms generalised self-description, regardless of whether the interviewer on the other end is a person or a system.
Drawing on your experience, the platform helps you build out a documented bank of real examples from your career, mapped against the specific role and company you are preparing for, so that whatever format the interview takes, screening call, technical round, AI-led conversation, or final panel, you are answering with the same grounded, specific material rather than improvising a different story each time. The Insider Intel module adds the context of how a specific company and team actually operate, so your examples land as relevant rather than generic.
Whether your next interview is conducted by a person or a voice agent, the underlying skill is the same: specific evidence, clearly organised, ready when you need it. Start free at candoora.io.