The invitation says forty-five minutes with the hiring manager and someone from the team. It does not say what will happen in those forty-five minutes, which is the part you would actually like to prepare for. So you prepare for everything: your story, your questions, the gap in 2023, the thing on your CV you hope nobody asks about.
Two quite different exercises go by the name interview, and they reward almost opposite kinds of preparation. The research separates them by a wide margin. Real interviews sit somewhere between the two, and roughly where is usually visible within the first few questions.
Two exercises, one word
For decades cognitive ability tests were treated as the stand-out predictor of job performance. In a 2022 paper in the Journal of Applied Psychology, Paul Sackett, Charlene Zhang, Christopher Berry and Filip Lievens argued that the field had been correcting those estimates wrongly.1 The technical claim is about range restriction: when you study people who were hired, you are studying a narrowed slice of the applicants, and the standard adjustments for that had been overdone. Most of the revised estimates came out lower, typically by .10 to .20.1
The interesting part is not that everything shrank. It is that things shrank by different amounts, so the order changed. The four authors restated the table the following year, drawing out what employers should do about it. Against the Schmidt and Hunter estimates they replaced, cognitive ability tests fell from .51 to .31 and work sample tests from .54 to .33, which the authors describe as ".20 lower for cognitive ability" and ".21 lower for work sample tests". Job knowledge tests fell from .48 to .40. The structured interview fell from .51 to .42, a drop of less than half the size.2
That left it on top. "Structured interviews emerged as the predictor with the highest mean validity", as the authors put it.2
The unstructured interview came in at .19, down from .38.2
That number is the one worth sitting with. These coefficients are correlations between how someone scored on a selection method and how well they later did the job, across many hires. They are not probabilities about any individual. But the gap between .42 and .19 is the difference between a method that carries real information about who will do the work well and one that carries very little. The unstructured interview reported a standard deviation of .16 around that mean, which puts the bottom of its 80% credibility band at roughly zero.2 That band is an estimate of how much validity varies across the settings the underlying studies covered, not a register of employers, and it does not identify anyone in particular. What it does say is that the spread reaches down to about no predictive value at all.
What structure actually means
Structure is not a mood, and it is not the same thing as a formal or unfriendly interviewer. It has a definition, and it is old enough to be well tested.
Campion, Palmer and Campion defined it in 1997 as "any enhancement of the interview that is intended to increase psychometric properties by increasing standardization or otherwise assisting the interviewer in determining what questions to ask or how to evaluate responses".3 Their typology — described in a later review as the most comprehensive available — sets out 15 distinct components grouped under two dimensions.3
The content dimension covers what happens in the conversation: basing questions on a job analysis, asking the same questions of each applicant, limiting prompting, follow-up and elaboration, using better types of question, using longer interviews or more questions, controlling ancillary information, and not taking questions from the applicant until the end.3
The evaluation dimension covers what happens to your answers: rating each answer or using multiple scales, using anchored rating scales, taking notes, using multiple interviewers, using the same interviewers across all applicants, not discussing applicants between interviews, training the interviewers, and combining the ratings statistically rather than by judgement.3
Fifteen components, and structure is a matter of degree rather than a switch. Most real interviews sit somewhere in the middle. What matters for you is that roughly half of those components are visible from your side of the table.
Reading how much structure is in the room
No single sign settles it, and that is worth saying before listing any. An interviewer working without a framework can still read from prepared questions and still take notes. A carefully built structured interview can allow standardised follow-ups, and most use some of the 15 components rather than all of them. The typology describes what structure is made of; it does not claim that any one component is necessary, or that any combination is sufficient, and it was written for the people designing interviews rather than the people sitting them. What you are reading from your chair is a quantity, not a category.
What you can do is count. The questions arrive in the same shape each time, often "tell me about a time when", and sound written rather than thought of. There is a pause and a pen after each answer, rather than a note taken when something happens to catch someone's interest. The follow-up you were counting on, the "say more about that" that would have let you land the point, does not come, or comes in the same words it came in for the candidate before you. More than one person is present and they take turns rather than converse. You ask a question of your own early and are told there will be time at the end. Any one of those on its own means little, and they do not carry equal weight. The more of them you notice, the more standardisation you are dealing with, and that is as precise as this gets from the candidate's side.
The other end reads differently. Your interviewer has your CV open and is looking at it for the first time. The conversation follows whatever you said thirty seconds ago. Nobody is writing anything down. It is often the more enjoyable hour, and it is the one the research values at .19.
The more structure, the more coverage pays
Four things follow from structure, each in proportion to how much of it you turn out to be in. Most are freeing rather than intimidating, and the first is worth doing whatever you are walking into, because specific evidence is never the wrong thing to have brought.
You cannot steer it, so stop planning a narrative. In a conversation you can guide the other person towards your strongest material. Against a fixed question set you cannot, and the effort spent constructing an arc is wasted. What pays instead is coverage: for each requirement named in the advert, one specific episode you could describe in two minutes, with what the situation was, what you personally did, and how it came out.
One weak answer need not be fatal. Rating each answer separately is one of the fifteen components, not a guarantee, and plenty of structured interviews score at the level of a competency or give one overall judgement instead. Where per-answer rating is in use it is designed to stop a single impression colouring everything else, which cuts against you when you open brilliantly and in your favour when the third question catches you cold. You will not know which scheme you are being marked under, so the habit that pays under either is the same: answer, let it go, and give the next question your attention rather than the last one.
Nobody is going to dig you out. This is the practical consequence of limited prompting, and it is where good candidates lose marks. If you describe a situation and never say what you did, an unstructured interviewer would ask. A structured one may not be permitted to. Give the outcome without waiting to be asked for it, and be explicit about which parts were yours rather than your team's.
Rapport is not the currency. It is not worthless, since you are still talking to people, but it is not what is being written down. Warmth that costs you specifics is a bad trade in this room.
Coverage is also the one thing that genuinely benefits from rehearsal out loud rather than in your head, because the failure mode is not forgetting the example. It is taking ninety seconds to reach the point. If you would rather practise that against the actual posting than against a generic list, JobCraftly will run a practice round from the job description you are interviewing for and give you feedback on the answers you actually gave.
Sometimes you can go further than guessing at the criteria, because they are published. The UK Civil Service prints its nine behaviours — seeing the big picture, changing and improving, making effective decisions, leadership, communicating and influencing, working together, developing self and others, managing a quality service, delivering at pace — on gov.uk, each with worked examples of what it looks like at each grade.4 Its candidate guidance tells applicants to "think about examples you can give of times when you have shown the behaviours outlined in the job advert".5 The advert names the behaviours; the framework describes them; both are public before you apply.
That is one employer, and a very large public one, and most private employers publish nothing comparable. It is still worth thirty seconds checking whether yours does. When an organisation has written down what it assesses, reading it beats inferring it.
Where the structure is missing, the burden shifts to you
The temptation is to conclude that a loose interview matters less. It does not. It still decides, and it decides on a basis that, averaged across the studies, tracks later job performance weakly. That is the whole of what the .19 says. It does not say what the rating was made on instead, and the meta-analysis does not go looking — the interviewer's mood, their private theory of who does well, and the direction the conversation happened to take are plausible candidates rather than findings, and it is a mistake to hand yourself an explanation the evidence has not supplied.
One thing helps, and it is the same thing: supply the structure yourself. When a question is vague, answer it with a specific episode anyway. Concrete evidence is what survives a conversation nobody wrote down, and it is the part of your preparation that does not care which kind of interview you drew.
Self-presentation is worth a word, with its limits attached. The 2014 review meta-analysed impression management inside structured interviews and found that how much candidates promoted themselves correlated .26 with the rating they received, non-verbal tactics .18, and other-focused tactics such as praising the interviewer or the organisation .13.3 Every one of those numbers comes from structured interviews. The review did not compare the two formats, so nothing here measures what the same behaviour does in a loose conversation, and none of it establishes that anyone was fooled — a candidate with more to promote will promote more. What it does show is that these effects persisted in the setting built to suppress them. Reading that as a reason they matter at least as much where nothing is suppressing them is an inference, and it deserves no more weight than that.
What a validity coefficient is not
None of this describes your interview. A meta-analytic estimate describes the average across many studies and many hires; it says nothing about whether a particular employer built their structured interview competently, and a badly built one is just a stilted conversation with a form attached.
The authors are direct about the uncertainty. Structured interviews average .42 but with a standard deviation of .19, giving an 80% credibility interval from .18 to .66, which they suggest reading as ".42, plus or minus .24" rather than as a single number.2 The best predictors on their revised list are the ones "specific to individual jobs", built by analysing the actual work.2 That is a claim about design effort, not about interview format, and it is the reason a structured interview at one employer can be worth so much more than the same label at another.
So the practical value of all this is smaller than the numbers suggest, and more useful. You will not know which employer you have drawn, and you will not know for certain which exercise you are in until you are some way into it. You will have a fair idea by the third question — enough to stop planning a story for an interview that may have no room for one, and to bring the evidence that works either way.
