Research update · 27 September 2026
AI interviews raised offers in one experiment. That does not mean more jobs.
In a Philippine customer-service recruitment experiment, applicants assigned to an AI voice interview were 12% more likely to receive an offer than applicants assigned to a human interview. Human recruiters still judged the interviews and made the decisions. This is an interview-stage effect, not a count of jobs created.
What was measured
Jabarian and Henkel studied 70,884 applications for 48 entry-level customer-service postings at one recruiting firm from March to June 2025. Of these, 67,056 eligible applications were randomized among AI interviewer, human interviewer and choice arms. The firm handled applications across 19 Philippine cities. The revised September 2026 working paper compares offers, starts, retention and hired-worker productivity.
The comparison
Applicants assigned to AI-led interviews were 12% more likely to get an offer than those assigned to human-led interviews. Job starts and short-term retention also rose. The authors report no decline in measured productivity among hires. The randomized design supports a causal interpretation for changing the interviewer in this firm; it does not test replacing human hiring decisions.
Where the result stops
The experiment does not measure AI-written applications, U.S. hiring, wages, layoffs or economy-wide employment. More offers at a recruiter could change who gets a job without increasing the number of jobs. One high-volume customer-service process may not transfer to other occupations or employers.
Why it changes the evidence map
Yesterday's theoretical paper described a possible barrier when AI makes application materials less informative. This experiment changes a later step: who conducts the interview. The two studies are not a head-to-head contradiction. Together they show why “AI hurts hiring” and “AI helps hiring” are both too broad without naming the hiring stage and population. U.S. exposure-based entry warnings and Yale's broad U.S. counter-signal remain separate.