How to Write Screening Questions That Actually Predict Job Performance (2026)
Most screening questions don't screen for anything. "Why do you want to work here?" "What's your greatest weakness?" "Tell me about yourself." Candidates rehearse them, the polished ones sail through, and three months later you're wondering how someone who interviewed so well struggles with the actual work. The problem isn't the candidates. It's that the questions reward fluency, not the thing you're hiring for.
The research on this is unusually settled. Across three decades of meta-analysis, how you ask and score interview questions drives most of the predictive power of the entire hiring process. Structured interviews — same questions, same rubric, same dimensions for every candidate — sit at a predictive validity of roughly .42 to .51, versus about .38 for unstructured "let's just chat" conversations. That's nearly twice as predictive of on-the-job performance. And it's not about spending more time: a focused 30-minute structured interview with anchored rating scales outperforms a rambling 90-minute conversation. The structure does the work, not the clock.
So the highest-leverage thing you can do isn't buying a new tool or running another round. It's writing four or five questions that are actually tied to performance, and scoring them the same way every time. Here's how.
Start from the work, not from a question bank
Before you write a single question, write down the two or three things that actually separate a strong performer in this role from a weak one. Not the job description — the real differentiators. For a support hire it might be "stays calm and clear with an angry customer." For a salesperson, "qualifies hard instead of chasing every lead." For an analyst, "notices when a number doesn't make sense."
Each screening question should map to exactly one of those differentiators. If a question doesn't connect to something that predicts performance, cut it — it's costing you signal and time. This single discipline, defining "good" before you source rather than after the first ten applicants, removes most of the noise from a funnel.
Ask about behavior, not hypotheticals
"What would you do if a customer was upset?" tells you whether someone can describe good customer service. It doesn't tell you whether they do it. Hypothetical questions test imagination; past-behavior questions test track record, and track record is the better predictor.
Rewrite every hypothetical into its behavioral twin:
- Instead of "How would you handle a missed deadline?" → "Tell me about a specific time you missed a deadline. What happened, and what did you do?"
- Instead of "Are you detail-oriented?" → "Walk me through the last time you caught a mistake before it shipped. How did you catch it?"
- Instead of "How do you handle pressure?" → "Describe the most stressful week you've had at work this year. What did you actually do day to day?"
The behavioral version is harder to fake because it demands specifics. Strong candidates reach for a real example in seconds. Weak ones drift back to generalities — and that contrast is itself signal.
Score with an anchored rubric, before you forget
A question only predicts performance if you score the answer consistently. Decide in advance what a 1, 3, and 5 look like for each question, in one sentence each. For "caught a mistake before it shipped": a 5 describes a specific instance, the mechanism they used to catch it, and what they changed afterward; a 1 says "I'm just naturally careful." Anchored scales are what separate a structured interview from a friendly chat that happens to use the same questions.
Score immediately after each answer, not at the end of the day. Memory compresses, the charismatic candidate gets a halo, and the quiet-but-precise one fades. Same questions, same anchors, scored on the spot — that's the whole method.
Keep the screen short — and asynchronous
You don't need ten questions. Four well-chosen, performance-linked questions outpredict a sprawling interrogation, and they're far less of a burden on everyone. The mistake most teams make is spending their best questions in a live phone screen — a low-information, high-coordination step where a recruiter and candidate burn a scheduled half-hour plus all the back-and-forth to book it.
There's a better order of operations: ask your two or three sharpest behavioral questions up front, asynchronously, before anyone spends calendar time. Have candidates record short answers on their own schedule, and review them in a batch when it suits you. You hear tone, energy, and how clearly someone thinks on their feet — signal a resume can't carry — without scheduling a single call. This is exactly what Flash Interview is built for: you set one question, candidates record a 50-second answer, and you watch when you're ready. The structured, behavioral question does the predicting; async recording removes the coordination tax so you can apply it to everyone in the pipeline, not just the few you had time to call.
A starter set you can adapt
Replace the bracketed differentiator with your own, keep them behavioral, and anchor each on a 1–5 scale:
- "Tell me about a time you had to [core skill] under a real constraint — a deadline, a tough customer, missing information. What did you do?"
- "Describe a decision you made in your last role that you'd defend even though someone disagreed with it. Why?"
- "Walk me through something you got wrong at work and what changed afterward."
- "What's a problem in [domain] you've solved that you're genuinely proud of? Be specific about your part."
Four questions, same for every candidate, scored the same way. That's not a longer process — it's a sharper one. The teams that hire well in 2026 aren't asking more questions or running more rounds. They've just stopped asking questions that don't predict anything.
Flash Interview lets you ask one screening question and collect 50-second video answers from every candidate — structured, asynchronous, no scheduling. €25 per role, one-time. See how it works.