The best interview questions to ask candidates are structured, behaviorally anchored, and tied to specific job competencies. They ask for concrete past actions ("Tell me about a time you...") rather than hypotheticals or self-ratings, and they are scored against a rubric. Structured interviews predict job performance roughly twice as well as unstructured ones, according to decades of meta-analysis (Schmidt & Hunter, 1998). That single design choice, not question cleverness, drives most of the accuracy.
How structured questions extract better signal
An interview is a measurement instrument. Every instrument has reliability (does it give the same reading on repeated use?) and validity (does it measure what you think it measures?). Unstructured conversation scores low on both because each interviewer probes differently, weighs answers differently, and is swayed by first impressions. Structure fixes the instrument: same questions, same order, same scoring anchors across all candidates for a role.
Behavioral questions work because past behavior is the most accessible predictor of future behavior. When you ask for a specific episode, you force the candidate to retrieve a real memory. That memory carries detail density, motive, and outcome that invented answers rarely sustain. Situational questions ("What would you do if...") test judgment and work well for roles with little prior experience, but they measure reasoning, not demonstrated action.
One level deeper: the mechanism is signal-to-noise. A resume gives you claims. A structured behavioral question gives you a sample of the candidate's actual reasoning under mild retrieval load. The scoring rubric converts that sample into a comparable number. Without the rubric, you have recreated the unstructured interview with extra steps.
Quantified context: what the evidence shows
The Schmidt and Hunter (1998) meta-analysis of 85 years of hiring research ranked selection methods by predictive validity for job performance. Structured interviews sat near the top, with validity coefficients around 0.51, versus roughly 0.38 for unstructured interviews. Work-sample tests scored similarly high. The gap matters: doubling predictive validity roughly halves the rate of bad hires relative to chance.
Other numbers worth holding. Typical interviewers reach a confident impression within the first few minutes, then spend the rest of the interview confirming it (a confirmation-bias pattern documented in interview research). Interview-to-offer ratios in competitive roles often exceed 5:1, meaning most questions get tested repeatedly and should be treated as calibrated instruments, not one-offs. Panel interviews with independent scoring before discussion reduce groupthink; discussion-first panels tend to converge on the loudest voice.
These are directional figures from a stable research literature, not precision estimates. Treat them as evidence that structure dominates question style, and that scoring discipline dominates both.
The question types compared
| Question type | What it measures | Predictive strength | Best used for | Main failure mode |
|---|---|---|---|---|
| Behavioral ("Tell me about a time...") | Demonstrated past action | High | Experienced hires | Vague answers with no specifics |
| Situational ("What would you do if...") | Judgment and reasoning | Moderate to high | Early-career, novel roles | Rehearsed, textbook answers |
| Work sample / job simulation | Actual task performance | High | Skills-heavy roles | Time and setup cost |
| Self-rating ("Rate your skill...") | Self-perception | Low | Rarely useful alone | Inflation and Dunning-Kruger effects |
| Hypothetical brainteaser | Puzzle solving under pressure | Low | Almost nothing job-relevant | Rewards trivia, alienates candidates |
The table's practical message: pair behavioral and situational questions, add a work sample where the job allows it, and drop self-ratings and brainteasers from scored use.
Edge cases and the counterintuitive caveat
The counterintuitive finding: charismatic interviewers who "connect" with candidates often produce worse predictions. Rapport feels like insight but is largely a fluency signal. The candidate who answers smoothly is not necessarily the candidate who performs. Structured scoring deliberately removes that warmth from the decision.
Other edge cases. Senior roles need fewer behavioral questions and more strategic judgment prompts, because the candidate's track record is already documented. Internal candidates need questions about the new scope, not a rehash of visible performance. Remote interviews lose nonverbal context, so increase question count and lean harder on written follow-ups. Panel interviews without independent scoring degrade toward the unstructured baseline.
Red flags in a candidate's answers, worth noting in the rubric: No specific example after two follow-up probes, only generalizations. Blames every prior failure on others with zero self-reflection. Claims credit for team outcomes with no description of personal action. Describes the same story under two different competencies. Refuses follow-up detail on scope, numbers, or timeline.
Practical actions: building the question set
Start from the job's top four to six competencies, derived from a real task analysis rather than a wish list. For each, write one behavioral question and one scoring anchor per level (weak, adequate, strong), describing what evidence qualifies. Then run the same set for every candidate and have interviewers score independently before any group discussion.
A workable sequence for a 45-minute interview: Two behavioral questions on core competencies, five minutes of follow-ups each. One situational question on a realistic job scenario. Ten minutes for candidate questions and role logistics, unscored. Five minutes for independent scoring against the rubric before you leave the room.
Give each scored answer a number on a defined scale, such as 1 to 4, with written behavioral anchors. Aggregate scores across interviewers after independent scoring. Where scores diverge by two or more points, discuss the specific evidence, not impressions. Keep questions stable across a hiring cycle so results are comparable; rotate them only between cycles, and log which questions produced high-variance scores.
Follow-up probes are where most signal is won: "What was your specific role?", "What did you decide, and what happened?", "What would you do differently?" A candidate who cannot supply scope, action, and outcome after three probes is giving you data about the answer's completeness, not just its content.
Limitations and what interviews cannot do
Interviews measure a candidate's ability to describe and reason about work, which overlaps imperfectly with doing the work. Work samples and job simulations capture performance more directly where feasible. Interview scores also drift over a hiring cycle as interviewers calibrate lazily, so periodic anchor refreshers help. Small sample sizes (three to five interviewers) mean single scores carry wide uncertainty; average them and resist reading meaning into one-point differences.
No question set eliminates bias. Structured scoring reduces it rather than removing it, and legal or regulatory constraints on what may be asked vary by jurisdiction, so keep every question job-related and apply it uniformly. Interpretation of validity figures should stay cautious: they come from aggregated research, not from your specific role, and local validation is the only real proof for your hiring process.
This article is general educational information about interview design, not legal, employment-law, or HR compliance advice. Hiring and candidate-assessment rules vary by jurisdiction and role. Consult a qualified local employment professional before adopting any selection process.