AI Interview Tools: What the Research Actually Shows — and It's Not Pretty

Thirty percent of large enterprises now use AI to score job interviews. HireVue alone processes 40 million of them a year. The promise is better accuracy, less bias, more consistency. But when you look at the independent research — not the vendor pitch decks — a very different picture emerges. In this episode, we break down what the science actually says about AI interview assessment accuracy, why the bias problem may be bigger than your vendor is telling you, and what the growing regulatory pressure means for employers who are legally on the hook for their tools' outcomes. The predictive validity numbers are sobering: independent studies consistently place AI interview tools at validity coefficients of .20 to .35 — on par with unstructured human interviews, not the structured behavioral interviews they're marketed as replacing. A 2023 University of Melbourne study found bias against non-native English speakers across all six commercial tools tested. And 38 percent of candidates say they'd decline a job offer from a company that used undisclosed AI screening — a real conversion problem in a tight talent market. We also look at what good AI screening looks like: structured, content-focused, transcript-only analysis with human oversight. That's the approach behind OVI's Milo screening agent, which uses audio-only chats and keeps recruiters in control of hiring decisions. Five questions to ask any vendor before you buy, and why employer liability is the issue that should be driving your compliance posture right now.