Questions that actually tell candidates apart: how to write them for any level of role

There is a test that throws out half the questions in any interview script, and it takes ten seconds per question. Imagine the answer of a candidate you would reject, and put it next to the answer of one you would hire. If the two look alike, the question is not measuring anything: it is occupying a minute.

That is what it means for a question to discriminate, and it is the only thing that earns it a place in a script. Everything else — that it sounds professional, that it fills the form nicely, that everyone else asks it — does not count.

The test does not change with the role. A well-written behavioural item tells people apart just as reliably on a shop floor as in a director search. What does change with the role is what counts as a complete answer, and how much detail you have to insist on before accepting one. That is the second subject of this article, and the one that ruins the most scripts when it goes unexamined.

The question everyone answers yes to

“Do you have experience leading teams?” is the perfect example of a useless item. The desirable answer is obvious, it is binary and it is free: nobody applying for a supervisory role is going to say no. The same goes for “are you a team player?”, “are you organised?”, and the whole family of questions that hand you the right answer inside the wording.

These questions are not harmless. They cost conversation time, which in any interview is the scarce resource, and they produce the illusion that something was measured. A report full of yeses says nothing, but it looks complete.

A knockout requirement is not a competency

They are two different objects, and confusing them is the most common way to waste an interview. A knockout requirement is a verifiable, binary fact: the current licence, availability for the night shift, the right to work, the certification the role requires by regulation. A data point settles it, it takes seconds, and it does not need a conversation.

A competency is something else: it is not verified, it is assessed, and that needs a story and a rubric to place it on a level. The issue on rubrics covers how that part gets written.

The rule that follows is blunt and saves a great deal: nothing that a checkbox can settle should take a minute of conversation. Requirements belong earlier, in the application or in the system the company already runs — which is what syncing with the ATS is for — and only people who meet them reach the interview.

An interview spent confirming requirements produces a long transcript and an empty record: plenty of text, nothing assessable. It is the same symptom a script without follow-ups produces, and for the same reason.

Behavioural versus hypothetical

“What would you do if an important client complained about the service?” measures one very specific thing: how well this person talks about work. It rewards whoever knows the genre, has read about the role, and improvises fluently. It says nothing about how they behave when a client actually complains.

“Tell me about the last complaint you handled from start to finish” asks for something else: an event, with its date, its own action and its ending. It is verifiable in the sense that matters here — not that a previous employer could confirm it, but that the answer contains concrete elements that can be checked against each other and against the rest of the conversation.

Two honest caveats, because the method has limits. First: behavioural questions favour people who narrate well, and that has to be compensated with the follow-up rather than ignored. Second: someone early in their career needs the context opened up — school, a previous job in another field, something they organised outside employment — instead of being asked about “your last role”, a question that already assumes its own answer.

From the weak question to the one that separates

Five rewrites, written as an example for this article. The middle column is the one that matters: it names why the first version separates nobody.

One item, three bars

Nothing so far has depended on the level of the role, and that is not an oversight: the behavioural question that works for a warehouse works for a head of operations, and for the same reason in both cases — it asks for an event with the person’s own action and an ending. What does depend on the level is where the bar sits: what the answer has to contain before it counts as good, and how far to press before treating it as closed.

Getting that wrong produces two classic mistakes, and they are mirror images. A frontline script written to a management bar rejects capable people for failing to articulate a decision their job never asked them to make. A management script written to a frontline bar settles for “I coordinated the team and it went fine”, which is exactly the sentence that separates nobody at that level.

It reads more clearly on a single item. Take the one about the supplier who let you down, and move only the bar.

With sample answers in front of you the difference stops being abstract. These are written for this article, and all three answer the same question.

• Frontline — “The material did not arrive on Tuesday. I called the supplier and got no answer, so I went to the shift lead and we filled the order from the next warehouse over. It shipped a day late.” That is closed: own action and ending are both there, and nothing more is needed.
• Technical — “We switched suppliers.” It sounds settled and it is not. What decides the level is missing: why that one and not the other, what was given up in the switch, and what broke afterwards because of it. The follow-up goes there, not to the outcome, which that sentence already handed over.
• Management — “I renegotiated the contract and we saved money.” Also not closed. At this level a complete answer names who it cost: the department left without the deadline it asked for, the supplier the relationship ended with, the team that worked two weekends. Without that it is a headline, and a headline cannot be assessed against any rubric.

Worth noticing: the bar does not rise by making the conversation longer. It rises by demanding something else inside the same answer. Duration is configured per script, and the maximum is agreed with each implementation; what makes an interview deep is what gets asked and what gets accepted as closed, not how many minutes it took. How that configuration usually lands for each kind of process — high volume, technical profiles, executive searches — is set out in the cases by type of process.

The bar is written down, not carried in someone’s head: next to each item goes what a complete answer contains for this role. If it is not written, the interviewer sets the level, and we are back to the impression structure was meant to replace.

The follow-up is written in advance, not improvised

No first answer arrives with the three parts you need. Usually the person’s own action is missing, or the outcome, or the magnitude, and sometimes all three. So every item is written with two things: what a complete answer contains, and what to ask when each part is missing.

• If the person’s own action is missing: “what did you do, specifically?”. It is the follow-up most often needed, because people narrate in the plural out of politeness.
• If the outcome is missing: “how did that end up?”.
• If the magnitude is missing: “how big was the problem?” — days, units, people, money, whatever fits that operation.
• If the answer is a policy rather than an event: “give me a concrete case where you did that”.
• And where the role calls for it, the one that reaches judgement: “what else could you have done, and why not that?”.

Writing them in advance is not bureaucracy: it is what stops the follow-up from becoming the new source of variance. If one interviewer presses and another does not, the comparison is over, and we are back at exactly the problem structure was meant to solve — the full argument is in the issue on structured interviews.

They also need a ceiling, and the ceiling is what moves with the level: one follow-up per item on a frontline role, two where the decision and its cost have to surface. Without one, a single item eats the interview and the competencies further down never get explored — and a third round of pressing does not yield better evidence, it yields an uncomfortable person.

How many questions fit, and in what order

Fewer than you would like. Each behavioural item with its follow-ups consumes several real minutes, and that sets the ceiling for a round; the arithmetic is in the issue on what fits into one round.

The order is not neutral either, and here the level changes nothing. Nerves and adjusting to the format take the first minute, so the opening question should be the easiest to answer: put the most discriminating one there and you get a bad answer to a good question. The one that separates most belongs second or third, with the person settled and time still ahead.

And it is worth not closing on the hardest one. The last question is the one people walk away with, and an interview is also the impression the company leaves — something worth deciding on purpose, as the issue on candidate experience argues.

What a good question does not fix

It does not fix a badly defined role. If the hiring team and HR do not agree on what is being looked for, excellent questions will produce excellent information about something nobody needs.

And it does not answer, on its own, what was never its to answer. A question serves the round in which it is asked, and a serious process is built from several: the same candidacy travels through them without turning into three different people, and who moves on is decided by someone on your team or by the company’s ATS, never by a rule that advances on its own. That does not mean the instrument cannot decide. It means it decides its own part, on the evidence it actually gathered, and that the hiring decision rests on all the rounds together — how that path holds up is set out on the platform page.

What is worth reviewing now and then, at any level, is who is being left out. A question with real discriminating power also excludes people who would have done the job well: the cut-off is set by the rubric and decided by a person. That is the subject of the issue on bias, which starts exactly where this one ends.