There is a pattern that repeats in almost every selection interview. Ask a management team to describe its own culture and you get policy language back: values, ambitions, three key words. Change the question to what it is like to work here and the jargon disappears. What takes its place is an example, usually a specific moment and with that you can do something.
The same reversal works in the conversation with a candidate. Tell me about yourself produces a rehearsed summary of the CV. Ask instead about the moment it went wrong, who was in the room at the time and what the person did first, and the answer can no longer be prepared. The difference does not lie in the sharpness of the question. It lies in its direction: you are not asking for a view of oneself, you are asking about behaviour.
The paper round has lost much of its signal
That reversal is not new. What is changing is how much now rests on it. In our own practice we see something shifting this year in the file that arrives before the first interview. CVs and covering letters are prepared with AI and it shows in the text: more polished, more balanced and less different from one another. In the interview itself we see far less of it.
We draw no moral conclusion from this. A candidate who has a language model read along with a letter is doing what anyone does who prepares seriously. The consequence is practical: the written round sorts less than it used to. A CV used to tell you how someone orders their own career: where the emphasis falls, what is left out, which word keeps returning. That signal has weakened. The weight of the assessment therefore shifts to the interview and in most organisations that is precisely the part no one has worked on.
The best-researched instrument you have
Which is striking, because the interview is not the weakest instrument in selection. It is the strongest. In a large-scale revision of decades of validity research, Sackett, Zhang, Berry and Lievens corrected in 2022 a systematic overestimation that had been in the literature since the nineteen-nineties. After correction, the structured interview arrives at a mean operational validity of .42 (the highest of all methods examined), above job knowledge tests (.40), work samples (.33) and cognitive ability tests (.31).
Two caveats belong with that and the authors make them themselves. These are correlations, not success rates: no selection instrument predicts performance reliably and anyone claiming otherwise is selling something. And with the structured interview in particular, the spread around that mean is wide. How you design, conduct and score the interview determines whether you come near that .42 or stay well below it. The instrument is strong, the execution decides.
What makes an interview structured
The structure does not lie in the questions alone. It lies in four agreements you make before the first conversation. You fix the questions and put the same ones to every candidate. You decide in advance which competencies you are testing and what counts as a weak, an adequate and a strong answer on each. You agree who asks what, so that the panel does not cover the same ground four times over. And you score immediately afterwards, separately, before the panel confers: otherwise the most confident voice wins the debrief rather than the sharpest observation.
Within that structure, STARR is the workhorse: situation, task, action, result, reflection. The value sits in the last two. Many panels stop at the action and that is exactly where the rehearsed story ends. Keep asking until the answer becomes specific. What happened afterwards? What would you do differently now and why Watch the pronoun while you are at it: a candidate who consistently says ‘we’ about a result and ‘I’ about a decision is telling you something without meaning to.
A question we always ask
Alongside the behavioural questions, we put the same question in almost every interview: suppose you published an autobiography, what would the title be and why?
It is not a behavioural question and is not intended as one. It asks not for a memory but for a formulation. The candidate has to order their own career in a single line and then account for that choice, and the second part does the work. The title itself is usually pleasant enough, the explanation shows where someone places themselves in their own story. Someone who describes themselves as the one who got things moving is saying something different from someone who sees themselves as the one who stayed when others left. We have yet to meet a candidate who could prepare for this question.
From the other side of the table
A selection interview falls into three parts and candidates almost always prepare for the middle one. That is the part where the substance comes up: what someone has done, how they approached it, what it produced. It is also the part on which the least is decided.
The opening sets the tone. In the first minutes the question on the table is why you are here and the answer to that is rarely a summary of a career. It is a reason. Anyone who cannot give that reason in one sentence leaves the panel searching for the rest of the interview for something that is not there.
The close is wasted most often. An interview that ends with the announcement that there are no further questions ends without direction. An interview that ends by asking what worries the panel most about the next two years ends in a conversation. What a candidate asks often says more about their level than what they answer.
What the regulation now asks of you
There is a reason to put this in order now that has nothing to do with quality. AI systems used in recruitment and selection: screening CVs, ranking candidates, scoring interviews; fall under the European AI Act in the high-risk category. The core obligations for that category were moved, under the digital omnibus agreement, from 2 August 2026 to 2 December 2027. The transparency obligations in Article 50 were not postponed and have applied since August of this year.
In practice that means two things. If you use AI in assessing candidates, the candidate should know. And the deferral of the heavier obligations is not a reprieve but a transition period: an organisation that must be able to show in 2027 how a judgement came about cannot reconstruct that afterwards in a process that was never documented. A structured interview with recorded questions and scores is therefore not only the most reliable instrument, but also the most defensible. This is the position as at September 2026: this file moves quickly and it is wise to have the current text checked before relying on it.
Closer to home
The arithmetic gets harder the smaller the pool. For South Kennemerland and the IJmond, the Dutch public employment service UWV published an overview this summer of more than 240 occupations with favourable prospects, with engineering, construction, energy and installation high on the list. Tightness in our region is easing, but it is not disappearing and job growth remains sensitive to sustained high energy prices.
For employers in the North Sea Canal area that has a concrete consequence. An appointment missed here is paid for twice: first the appointment itself, then the months before a second serious candidate is at the table. In a broad market a lost candidate is a setback. In a tight regional market it is half a year.
In closing
The behavioural interview requires no new instruments and no additional round. It requires that you set down, before the conversation, what you want to know and how you will recognise a good answer. That costs some time of preparation and it yields the only part of the file that was not drafted somewhere else.
So the question is not whether you ask good questions. The question is whether your panel agrees on what a good answer would be and whether it agreed on that before the first candidate walked in.