Interview questions that predict performance: 7 to ask, and how to score them

Search "interview questions that predict performance" and you'll get a hundred lists. Almost all of this is opinion and has no science behind it.

The questions that actually predict how someone does the job aren't secret, and they aren't clever. And they share one common trait: they force the candidate to show real evidence instead of just speaking well.

I've watched a lot of interviews. The most reliable tell is whether you capture the answer and score it the same way for everyone. Get that wrong and the best question in the world predicts nothing.

So here are seven questions worth asking, what good looks like for each, and the part most teams skip that makes any of them work.

Why most "best questions" lists are useless

Most question lists rank on how interesting the question sounds rather than whether the answer tells you anything about the job. "If you were a tree, what kind would you be" is memorable and predicts nothing.

The research has been settled for a long time. Structured, evidence-based interviewing, where everyone is asked comparable questions and scored against a rubric, is one of the most predictive hiring methods we have. It leaves unstructured "let's just chat" interviews far behind.

So the bar for a good question is simple: will the candidate's answer reliably predict how they'll perform at work? The questions that clear this bar all do the same thing: they make the candidate produce evidence.

The 7 interview questions that predict performance

These seven come from the patterns we see across 5.5 million captured conversations, picked on one test: does the question force evidence? Don't treat them as a script. The exact wording matters less than the follow-ups, which are where the evidence actually shows up. Pull from our interview questions library for role-specific versions.

1. Walk me through a project you owned end to end. What did you personally do?

This separates the people who did the work from the mere bystanders. What good looks like: specific first-person decisions and actions, with a clear "I did this."

2. Tell me about a decision you made without enough information.

Judgment under uncertainty is most of real work. Candidates rarely get complete information on the job either, so how they reason when the data is thin tells you how they'll operate in week three. What good looks like: a clear reasoning process and the tradeoff they accepted. A lucky outcome tells you nothing.

3. Describe a time your work failed or fell short. What did you change after?

Ownership and learning predict growth better than a flawless track record. What good looks like: they name their own part and a concrete change. And they don't blame others for the circumstances.

4. How would you approach this specific problem from the role in your first 90 days?

A situational question tied to the actual job. What good looks like: they ask clarifying questions and reason about your context instead of reciting a textbook answer.

5. Show me how you would actually do the core task.

A short work sample is the single most predictive thing you can do in an interview. What good looks like: they think out loud and handle the edge case as well as the happy path.

6. Teach me something you know well that I probably don't.

Depth and communication at once, and very hard to fake. What good looks like: they pitch the explanation to your level and check that you followed.

7. What would you need from us to do your best work here?

Self-awareness and fit, and it makes the interview two-way. What good looks like: a specific, honest answer about how they work. "A good culture" is a dodge.

The part that makes any question predictive

Here's the uncomfortable truth about every list like this one. The question is only half of it, and it's the easy half.

The same great question predicts nothing if one interviewer scores a "3" generously, another scores it harshly, and a third fills the scorecard from memory two days later. The signal you worked to get is lost the moment the answers are remembered instead of recorded.

This is not a rare failure. Across 5.5 million interviews, only 31% ever get a scorecard at all, and 42% of candidates who advance had no submitted scorecard when the decision was made.¹ When the evidence is missing, the question was never the thing that decided.

This is the work Metaview does. It captures every answer, writes the notes against the competency each question is testing, and drafts the scorecard from what the candidate actually said. Metaview captures the evidence, and recruiters and hiring managers add the judgment, so the same question gets scored the same way, candidate after candidate. And when the score is actually recorded, it carries weight: candidates rated a strong yes go on to advance at 47.4%, against 6.2% for a no.¹

The questions and the method have to match. If the rubric is vague, even a great question gets scored on gut feel.

How to put it to work

You don't need all seven questions in one interview. Pick the four to six that map to the competencies the role actually needs, ideally set in the intake call, and spread them across the loop.

• Give each interviewer one or two competencies to own, so they go deep instead of trading broad impressions
• Score against the same rubric, during or right after the interview, never from memory
• Recalibrate the bar now and then with regular calibration, and pair the questions with competency-based interviewing
• Review how each question is scoring across the team in Reports, and cut the ones that never separate candidates

The best interviewers have always worked this way. They ask for evidence and they score it the same way every time. The questions on this list just give you a place to start, and capturing the answers is what makes them predict anything at all.

¹ Source: Metaview's corpus of 5.5 million captured conversations (5.2 million candidate interviews), 2026. Scorecard-coverage, advancement, and early-versus-final agreement figures are drawn from platform data.