The Job Interview: Ask Structured Questions, Choose Better

Why good conversations often select poorly

An applicant arrives on time, seems likable, talks fluently about previous jobs, and asks clever questions. After an hour the owner has a good feeling and makes an offer. Six months later it turns out that the pleasant manner in the interview revealed little about how reliably someone follows up on quotes or how carefully they document.

This is not an isolated case but a pattern. Unstructured interviews mainly measure how well someone conducts a conversation. For some positions that matters; for most it is marginal. Add to this thinking errors everyone knows: the first impression colors everything that follows, and people similar to us seem more competent. The article entrepreneurs' thinking errors describes why confirmation bias is especially hard to notice in personnel decisions.

This guide therefore answers one question: How does a small business conduct a job interview that predicts later performance as well as possible without turning it into an exam? To answer it, I first clarify what a structured interview is, present a mental model from personnel assessment, summarize the findings on accuracy, and then build an interview guide step by step. The legal limits come at the end.

What a structured job interview is

Levashina, Hartwell, Morgeson, and Campion (2014) summarized two decades of research on structured interviews. Structure there essentially means two things. First, content: the questions are derived from an analysis of the position, and all applicants get the same ones. Second, evaluation: each answer is rated against standards set in advance, preferably by more than one person, and only combined into an overall judgment at the end.

An unstructured interview is the opposite: questions arise from the flow of conversation, each applicant experiences a different interview, and the judgment emerges as an overall impression. There are many gradations in between. For a small business it does not have to be the full program. The same questions for everyone and a simple rating scale already deliver most of the benefit.

The mental model: sign or sample

Since a 1968 paper by Wernimont and Campbell, personnel assessment has distinguished two strategies for evaluating applicants. The sign approach looks for characteristics assumed to be related to performance: personality, intelligence, demeanor. The sample approach, by contrast, brings a piece of the later work into the selection process: a work sample, a case task, questions about how someone actually handled a typical situation.

Sackett, Zhang, Berry, and Lievens (2023) draw exactly this conclusion from their reassessment of the research. The most accurate methods are those tailored to a specific position: structured interviews, job knowledge tests, work samples, and empirically keyed biographical questionnaires. General measures of ability and personality perform worse. Questions of the type “Tell me about a time when …” are, for the authors, a form of sample, because they ask about past behavior rather than self-descriptions.

An everyday analogy makes the difference clear. Anyone who wants to hire a cook can ask whether he enjoys cooking and infer his skill from his enthusiasm. That is a sign. He can also have him cook a dish. That is a sample. In the interview itself this is not always possible, but the questions can be asked in a way that comes as close as possible to a sample.

What research shows about accuracy

For a long time, a review by Schmidt and Hunter (1998) was considered the standard. Sackett, Zhang, Berry, and Lievens (2022) showed that earlier meta-analyses systematically overcorrected for restricted samples, and recalculated the values. In their 2023 follow-up paper they set the old and new values side by side. The measure is validity, that is, how closely the result of a method is related to later job performance, on a scale from 0 to 1.

Three findings matter for owners. First, the structured interview is at the top at 0.42, the unstructured interview at 0.19 less than half as high. Under the old values, the two were closer together at 0.51 and 0.38. Second, years of job experience, at 0.07, say almost nothing about later performance. Third, the values vary widely. For structured interviews, the interval containing 80 percent of the values ranges from 0.18 to 0.66. A poorly built structured interview can therefore be weaker than a good unstructured one.

The values are not undisputed. Other researchers consider the new corrections too cautious, especially for cognitive ability tests. This dispute does not change the ranking between structured and unstructured interviews. Wingate, Bourdage, and Steel (2025) showed in a further meta-analysis of 37 studies that interviews predict both core task performance and behavior toward colleagues similarly well, and that structured scoring helps especially with the latter.

A job interview without structure mainly measures how good someone is at job interviews.

Two views: gut feeling or rating form?

Many owners have reservations about structured interviews, and not all are unfounded. The first view is: anyone who knows their business senses whether someone fits. A rating form takes the warmth out of the conversation, unsettles applicants, and crowds out the question of whether one wants to work with this person every day. In a small team, this question is legitimate.

The second view rests on the findings. Gut feeling is not a neutral sensor but is prone to similarity, first impressions, and the mood of the day. Levashina and colleagues explicitly treat two questions as separate strands of research in their review: how structure can reduce such biases and how applicants react to structured interviews. Whether the objections outweigh the advantage is answered by the validity values above: in my assessment, they do not.

I consider a combination sound. The core of the interview is structured: the same questions, the same rating. Before and after, there is room for an open getting-to-know-you, for the applicant's questions, and for an honest picture of the position. Earnest, Allen, and Landis (2011) showed in a meta-analysis of 52 studies that realistic job previews reduce later turnover mainly because the business is perceived as honest. That belongs in every interview, structured or not.

Building the interview guide in five steps

For a position in a small business, one hour is enough to build a usable guide. The basis is a description of what the position actually requires. What such a description looks like is shown in the article job descriptions that help in daily work.

1. Set three to five requirements

   What must the person be able to do or bring from day one? Examples: reliability in following up, careful work according to specifications, calm handling of angry customers. What is learned only in the business does not belong here.

2. Write two questions per requirement

   One question about past behavior (“Tell me about a situation in which a customer was very angry. What did you do?”) and one situational question (“A customer calls, the appointment is tomorrow, the material is missing. What do you do?”).

3. Set the rating standard

   For each question, note in advance what makes a weak, a medium, and a strong answer, on a scale from 1 to 5. This is the step most people skip, and the most important one.

4. Include a work sample

   A short task from the position's daily routine: answering a real, anonymized customer inquiry, checking a measurement, evaluating a spreadsheet. It is the most direct form of sample.

5. Rate separately, then compare

   If two people conduct the interview, each rates on their own before you exchange views. Otherwise the louder one takes over the quieter one's judgment.

For sales positions, the guide hiring your first salesperson describes which work samples have proven themselves and how not to be dazzled by demeanor.

What the law says

In Germany, job interviews are subject to the General Equal Treatment Act (Allgemeines Gleichbehandlungsgesetz, AGG). It prohibits discrimination on grounds of ethnic origin, sex, religion or belief, disability, age, and sexual identity. Questions aimed at these are generally inadmissible, for example about pregnancy, family planning, religious affiliation, or origin. Questions with a clear relation to the position are permitted.

• Compensation: If an applicant is discriminated against, they can claim compensation. If they would not have been hired even with a non-discriminatory selection, it may not exceed three monthly salaries (Section 15(2) AGG).
• Deadline: The claim must be asserted in writing within two months of receipt of the rejection (Section 15(4) AGG).
• Documentation: A structured interview with recorded questions and ratings helps to justify a selection objectively.
• Rejection: A rejection does not have to be justified. A justification that mentions a protected characteristic, by contrast, creates a risk.

This overview is not legal advice. In case of uncertainty, for example about questions on criminal records or health for certain activities, have the individual case reviewed by an expert.

What you can do before the next interview

• Set three to five requirements the person must meet from day one.
• For each requirement, write one question about past behavior and one situational question.
• For each question, note what makes a weak, medium, and strong answer.
• Prepare a short work sample from the position's real daily routine.
• Ask all applicants the same questions and rate immediately after the interview.
• Describe the position honestly, including what is demanding about it.

The decision does not end with the offer. How the first weeks and the probationary period are used to confirm or correct the choice is described in the guides onboarding new employees and managing the probationary period deliberately.

Conclusion: how does a small business choose better?

The opening question was
a small business conducts a job interview that predicts later performance as well as possible. Research gives a clear answer. Structured interviews with the same questions derived from the position and fixed rating standards are, by current knowledge, the most accurate common method. Unstructured interviews reach less than half their predictive power, and years of experience say almost nothing.

In practice, this does not mean turning the interview into an interrogation. It means structuring the core, gathering samples of the work instead of self-descriptions, and making the judgment only after the rating. Gut feeling then gets its place back, as one voice among several, not as the only one.