Can AI Conduct Better Job Interviews Than Humans?

Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews

By Brian Jabarian and Luca Henkel
Winner of the 2025 Thaler-Tversky Award and the NABE E.A. Mannis Prize

What We Studied

We ran one of the largest natural field experiments on AI in the workplace. 70,000 real job applicants were randomly assigned to be interviewed by either an AI voice agent or a human recruiter as part of a live hiring process at a global recruiting firm.

This was not a lab study — these were real interviews, for real jobs, with real consequences for applicants and firms.

What We Found

Applicants interviewed by AI received 12% more job offers than those interviewed by humans.

These gains weren't just on paper. AI-interviewed candidates also had higher job start rates and better retention, with no decline in productivity once hired.

The key mechanism: AI voice agents achieved what we call "controlled variance." Their interviews were more structured and consistent than human interviews, while still being responsive to individual applicants. This meant they collected more hiring-relevant information per interview.

Human recruiters made the final hiring decisions in all cases — they simply had better information to work with when the initial interview was conducted by AI.

Why It Matters

This study provides causal evidence that AI can improve real hiring outcomes at scale — not by replacing human judgment, but by improving the information that feeds into it.

The findings have implications for how firms organize their hiring processes, how workers experience recruitment, and how policymakers think about AI adoption in labor markets.

This research is part of the Artificial Process Outsourcing program, a five-year initiative studying how AI reorganizes workflows and decision-making in firms.

Read the Paper

View on SSRN →

Selected Media Coverage