Structured interviews: how to build one that actually predicts performance
Structured interviewing is the best-evidenced improvement available to most hiring processes, and the most commonly claimed but rarely implemented. The difference is not the question list — it is what happens to the answers.
- A structured interview means the same job-relevant questions, the same order, and the same anchored scale for every candidate.
- Structure improves predictive validity mainly by reducing the variance between interviewers, not by making any one interviewer smarter.
- The rubric, not the question list, is the part that does the work — questions without anchored ratings are still a chat.
- Score each competency independently and immediately; a single overall impression contaminates everything downstream.
- The four structure-killers are drift, halo, unequal probing and post-hoc justification.
What actually makes an interview structured
Most teams that say they run structured interviews mean they have a shared question list. That is a necessary condition and nowhere near a sufficient one. Structure is a property of the whole assessment, and it has five components.
- 01Job-relevant content. Questions derive from an analysis of the role, not from what the last interviewer found interesting.
- 02Consistency of delivery. Every candidate gets the same questions, in the same order, with follow-up probing governed by rules rather than curiosity.
- 03An anchored rating scale. Each competency has defined behavioural anchors describing what a 1, a 3 and a 5 look like for this role.
- 04Independent scoring. Each competency is rated on its own evidence before any overall impression is formed.
- 05Evidence capture. The rating is recorded alongside what the candidate said that justified it.
Drop any one and the structure leaks. Drop the rating anchors specifically, and you have a consistent conversation being assessed inconsistently, which delivers a fraction of the benefit while costing all of the effort.
Why structure works, and what it does not fix
The research consensus across decades of selection psychology is consistent on the direction: structured interviews substantially outperform unstructured ones at predicting job performance, and the gap is one of the larger effects in the field.
The mechanism is worth understanding because it tells you where to spend effort. Structure mostly works by reducing variance — between interviewers, between candidates, and between one interviewer’s Monday and their Friday. It constrains the assessment to job-relevant evidence and denies the interviewer room to reward familiarity, articulacy or shared background.
What it does not fix
- A bad job analysis. Structure applied to the wrong competencies measures the wrong things very consistently.
- Rater leniency. If everyone scores 4s, structure has not helped; anchors have to describe observable behaviour, not adjectives.
- Pipeline composition. A structured interview assesses who applied; it does not change who applied.
- Interviewer capacity. Structure makes each interview better, not faster — which is precisely the constraint that makes teams abandon it under volume.
Building the rubric
Start from the role and work outwards. Six to eight competencies is the practical range: fewer and you cannot discriminate between candidates, more and raters stop attending to the differences.
- 01List the outcomes the role has to produce in its first year, in concrete terms.
- 02For each outcome, name the behaviour that produces it. That behaviour, generalised, is a competency.
- 03Discard anything you cannot observe in an interview. "Attention to detail" is observable through a worked example; "passion" is not observable at all.
- 04Write behavioural anchors for each competency at a minimum of three points on the scale, describing what a candidate would actually say or do.
- 05Write two to three questions per competency that give a candidate a genuine opportunity to demonstrate it.
- 06Define the probing rules: what a rater does when an answer is thin, and what they must not do.
An anchor is a description, not an adjective
| Rating | Weak anchor (avoid) | Usable anchor |
|---|---|---|
| 1 | Poor communication | Answers required repeated clarification; left out information the listener needed; did not check understanding |
| 3 | Adequate communication | Answered the question asked, in a followable order, with enough context for a listener outside the team |
| 5 | Excellent communication | Adapted the level of detail to the listener, surfaced the key point first, and checked understanding before moving on |
The four mistakes that unstructure a structured interview
1. Drift
By the fortieth interview the question set has quietly changed. A rater has decided one question is not useful, added a favourite, and reordered the rest. Nobody decided this; it accumulated. Version the question set and audit delivery against it.
2. Halo
One strong answer early sets an impression that colours every subsequent rating. The fix is procedural, not attitudinal: score each competency immediately after its section, and do not permit revision of an earlier score once a later section has begun.
3. Unequal probing
The most consequential and least noticed. Raters probe candidates they find promising and let thin answers pass from candidates they do not. The candidate then scores lower on evidence the rater declined to seek. Define probing rules — how many follow-ups, on what trigger — and apply them identically.
4. Post-hoc justification
The rater decides, then writes evidence supporting the decision. The record looks impeccable and proves nothing. Capture evidence before the score, and require the evidence to be a quotation rather than a characterisation.
Holding structure at volume
Structure survives ten candidates and collapses at four hundred. Every organisation discovers this at the same point: the process is sound, the volume arrives, and the first thing sacrificed is the part that makes the process worth having.
The three ways teams resolve it, in ascending order of how well they work:
- 01Screen harder on CVs first. Preserves interviewer capacity by rejecting most people on the least predictive evidence available — the opposite of the intended effect.
- 02Shorten the structured interview. Preserves the form and loses the content; four competencies assessed in ten minutes discriminate poorly.
- 03Automate the delivery and scoring, keep the structure intact, and spend human interviewer time only on candidates who have already demonstrated evidence. This is what FirstPanel does — the same rubric, the same anchored scale, the same evidence requirement, applied to every applicant rather than to the survivors of a CV sift.