Letting an AI voice agent conduct job interviews, instead of a human recruiter, led to more hires who stayed on the job. In a large field experiment, applicants interviewed by AI were 12% more likely to get an offer, even though humans still made every hiring decision.
The study is by economists Brian Jabarian, then at the University of Chicago, and Luca Henkel of Erasmus University Rotterdam. It was first posted as a working paper on SSRN on August 18, 2025.
Seventy thousand real applicants, three kinds of interview
The researchers partnered with PSG Global Solutions, a recruitment outsourcing firm and subsidiary of Teleperformance, which hires for entry-level customer service jobs. About 70,000 applicants who passed initial screening were randomly assigned to one of three conditions: an interview with a human recruiter, an interview with an AI voice agent, or a free choice between the two.
Both humans and the AI followed the same interview guidelines on topics, order and sample questions. After the interview, every applicant took a standardized test of language and analytical skills. Human recruiters then reviewed the interview and the test and decided whether to make an offer.
Professional recruiters had predicted that AI interviews would do worse. They did better.
More offers, more starts, longer stays
Across all applicants, AI-led interviews raised job offers by 12%, job starts by 18% and the share of people still employed after 30 days by 17%. In the revised version of the paper, the offer rate was 8.70% in the human condition and 9.73% in the AI condition.
The authors explored three explanations. First, interview transcripts showed that AI interviews drew out more hiring-relevant information from applicants. The revised paper attributes this to what the authors call controlled variance: the AI stuck more closely to the topic order and wording of the guidelines while still asking follow-up questions tailored to each applicant.
Second, recruiters gave higher interview scores to applicants the AI had interviewed. But when judging those applicants, recruiters leaned more heavily on the standardized test results.
Third, applicants did not appear to mind. They accepted offers at similar rates and rated the interview and recruiter quality similarly in a customer experience survey. The one notable gap in the revised paper is that AI interviews were rated as less natural.
Most people chose the machine
When offered the choice, 78% of applicants picked the AI interviewer. That choice was not random: applicants with lower test scores were more likely to choose AI.
The authors conclude that AI can match human recruiters at conducting interviews without hurting applicant satisfaction or the firm’s operations. In the revised paper, they also report no drop in the productivity of workers hired through AI interviews, measured by call handling time, customer satisfaction and quality scores.
The design matters for reading the result. The AI only ran the conversation; people still made the call. In this setting, swapping the interviewer, not the decision maker, was enough to raise hiring and retention.
Study Details:
- Title: Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews
- Authors: Brian Jabarian, Luca Henkel
- Journal: University of Chicago Working Paper (SSRN)
- Publication Date: August 18, 2025
- DOI: 10.2139/ssrn.5395709