The classic hiring interview, a smart person riffing questions and trusting their gut, is one of the weakest predictors of performance we use. The research has said so for decades. Google ran the experiment on itself and landed in the same place. The fix isn't complicated, just uncomfortable, because it means giving up the part of interviewing that feels the most insightful.
Key Takeaways
- A widely-cited meta-analysis put structured interviews at .51 validity versus .38 unstructured (Schmidt, Oh & Shaffer's review).
- A later re-analysis narrowed the gap, so don't over-index on exact figures; the case for structure holds either way.
- Google found four structured interviews enough to predict a hire with 86% confidence (Google re:Work).
- Standardize the questions and the scoring, or you're mostly measuring rapport.
What the Research Says
The widely-cited meta-analysis by Schmidt and Hunter put structured interviews at roughly .51 validity for predicting performance, against about .38 for unstructured. A later re-analysis by Schmidt, Oh and Shaffer, which reports both numbers, revised interview validities and found the gap smaller than the classic figures suggest. Don't over-index on the precise decimals. The practical point survives either way: structure makes the interview more consistent and fairer, and pairing it with a work sample pushes composite validity past .60.
Google's own hiring research backs it from the practitioner side. In its re:Work guidance, Google reports that four structured interviews were enough to predict whether to hire someone with 86% confidence, and that structured formats both predict better and reduce demographic bias. Google's rubrics are what researchers call behaviorally-anchored rating scales: for each question, a written description of what a poor, borderline, solid, and outstanding answer looks like.
Why Unstructured Interviews Fail
An unstructured interview mostly measures whether the interviewer likes the candidate. Different questions for each person, no consistent scoring, a heavy dose of first impressions, and you get a process that rewards confidence and rapport over ability. It feels insightful in the room. On paper it's close to noise.
| Unstructured | Structured | |
|---|---|---|
| Questions | Improvised, vary per candidate | Same, job-relevant, per candidate |
| Scoring | Gut feel | Defined rubric |
| Predicts performance | Weakly | Meaningfully better |
| Bias | Higher | Lower |
A Concrete Version
Two candidates, same role. In the unstructured version, Candidate A is warm, tells a good story, and reminds the interviewer of a past star hire, so they walk out a "yes." Candidate B is quieter, gives sharper technical answers, but doesn't click socially, so it's a "meh." Now run the structured version: both get the same five job-relevant problems, scored against the same rubric. B outscores A on four of five. Same two people. The structure surfaced the signal that rapport was drowning out.
The Honest Counterpoint
Structure has a real cost, which is why so few teams actually do it: the rubrics are hard to build. You have to write the questions, define what good and bad answers look like, test them, then get interviewers to stick to the script instead of freelancing. Google is upfront that this is the main deterrent. It's genuine work. But it's front-loaded work you do once and reuse on every candidate, versus re-rolling the dice on gut feel in every single interview.
What This Means for Hiring
The fix is to standardize: the same job-relevant questions, a scoring rubric, and a real work sample. That's the backbone of how our vetting works, a consistent five-stage process rather than a different conversation each time, and it's a big reason we can verify seniority reliably enough to hit 97% six-month retention. Structure has a second payoff for distributed hiring: a documented, rubric-scored process is portable. Any interviewer, anywhere, runs the same evaluation, which matters when your pipeline spans timezones. See available engineers.
Frequently Asked Questions
Are unstructured interviews really that weak?
The research puts them well below structured ones at predicting performance, and Google found structured formats both predict better and reduce bias. An unstructured interview largely measures rapport.
What makes an interview "structured"?
The same job-relevant questions for every candidate, scored against a defined rubric (Google calls these behaviorally-anchored rating scales), ideally paired with a work sample. Consistency is what creates the signal.
How many interviews do I actually need?
Google's research found four structured interviews enough to predict a hire with 86% confidence. Beyond that you add cost without much added signal.
What's the single best hiring signal?
A work sample or cognitive test paired with a structured interview, which pushes composite validity past .60, higher than any single method alone.
The Bottom Line
If your interviews are improvised, you're mostly measuring likeability. Standardize the questions and the scoring, add a work sample, cap it around four rounds, and you turn the interview into a real signal, which is exactly what rigorous vetting is built to do.
Roberto Espinoza is CEO of Ruzora, which helps US startups hire pre-vetted senior LATAM engineers in 72 hours. See available engineers.
