AI interviews improved hiring outcomes in a 70,000-applicant trial, researcher explains

Job applicants interviewed by AI were more likely to receive offers and start work in a 70,000-person field experiment. In a November 13 Chicago Booth Review interview, Carnegie Mellon researcher Brian Jabarian explained the distinction behind those results: AI conducted the conversations, but human recruiters still decided whom to hire.

Jabarian and coauthor Luca Henkel studied applicants seeking real customer-service jobs through a recruiting firm. Candidates were randomly assigned to interviews with either human recruiters or AI voice agents.

The interview offers an explanation of earlier research, not a newly released November experiment. Their paper, Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews, was first submitted to arXiv on July 30 and revised September 11. It reports that AI-interviewed applicants were 12% more likely to receive offers—a relative increase, not a 12-percentage-point gain.

More offers, without lower measured productivity

The researchers followed what happened beyond the interview: offers, job starts and continued employment. Their paper reports higher starts and worker retention among AI-interviewed applicants, with no decline in productivity among those hired.

That distinction is central to Jabarian’s interpretation. More offers alone could mean an interview makes candidates look stronger without identifying better matches. He pointed to the absence of a productivity decline as evidence against that explanation, arguing that the AI interviews helped reveal skills recruiters valued rather than simply inflating applicants’ apparent suitability.

Jabarian described better retention through four months, but also said separation after 30 days did not differ. How those measures relate is unclear in the interview.

The same topics, more consistent conversations

Both kinds of interviewer were required to cover 14 topics. According to Jabarian, the AI agents covered topics more consistently and showed less variation in follow-up questions and conversational flow. He identified fatigue as one reason a human recruiter’s execution could vary across interviews, even when the assigned task stayed the same.

Consistency did not mean identical conversations. The paper describes AI interviews as more structured while remaining responsive to individual applicants. Its transcript analysis associates that combination with collecting more hiring-relevant information.

The overall takeaway of this research is not that you should replace all the time humans by AI.

Brian Jabarian, speaking to Chicago Booth Review

Candidates knew when they were speaking with AI. Human recruiters then reviewed interview transcripts and audio before making final hiring decisions. Those features define the workflow tested: disclosed automated interviewing followed by human evaluation, not an undisclosed system selecting employees on its own.

Jabarian’s recommendation is to consider the organization of work as a whole, rather than deciding task by task whether AI should replace a person.