Structured interviews and scorecards: a practical guide

A structured interview asks every candidate for a job the same job-related questions, in the same order, and scores each answer against a rating scale agreed before the first interview. The interview scorecard is the form that holds that scale: one row per competency, a description of what each score looks like, and the interviewer's notes on what the candidate said. Structure raises how well an interview predicts job performance: in a 2022 re-analysis of the research, Sackett and colleagues estimated the validity of structured interviews at .42, against .19 for unstructured ones.

Structured, semi-structured and unstructured interviews

The US Office of Personnel Management (OPM) sets out the difference in its 2008 guide, Structured Interviews: A Practical Guide. "Semi-structured" is not one of OPM's categories. The term usually covers the middle of Huffcutt and Arthur's 1994 scale, where the topics or a list of questions are set but the interviewer picks which to ask and how to follow up.

OPM's guide is blunt about the last column: "Research consistently indicates unstructured interviews have little value in predicting job performance." It also notes that the lack of a standard procedure makes unstructured interviews "susceptible to legal challenges".

What the research says

A validity figure is the correlation between interview scores and later job performance, where 0 means the score tells you nothing. The figures below come from meta-analyses, which pool many studies.

Sackett and colleagues re-ran the older estimates in 2022 after finding that earlier meta-analyses had over-corrected for range restriction. Most methods dropped, and "structured interviews emerged as the top-ranked selection procedure." How the interview compares with work samples, knowledge tests and cognitive tests is set out in skills-based hiring.

You do not need every possible element of structure. In Huffcutt and Arthur's 1994 analysis, validity rose as structure was added through most of the range, then flattened. Of the 15 components of structure that Campion, Palmer and Campion listed in 1997, the research interviews that Levashina and colleagues coded in 2014 used about six on average. The ones used most were a job analysis, the same questions for everyone, better types of questions, anchored rating scales, rating each question and interviewer training.

Google reports similar results from its own hiring. Its re:Work guide, updated in March 2026, says that structured interviews were more predictive of job performance than unstructured ones, that ready-made questions and rubrics saved interviewers 40 minutes per interview on average, and that rejected candidates who had a structured interview were 35% happier than those who had not.

How to build a structured interview

OPM's guide sets out eight steps. In short:

1. Analyse the job. List the tasks and the competencies needed to do them, and mark which are needed on day one. Position descriptions, performance standards and interviews with strong performers and their supervisors are the usual inputs.
2. Choose the competencies the interview will measure. OPM says a structured interview typically assesses between four and six, and that some, such as oral communication and interpersonal skills, suit an interview particularly well. Decide which of the rest another method, such as a written test, measures instead.
3. Write the questions. Each question measures at least one competency. OPM suggests a group of about six or seven experienced, high-performing employees or supervisors write them, and that they write more than needed so weak ones can be dropped.
4. Write the rating scale. Use one scale for every competency, at least three levels and ideally five to seven, with at least three of the levels labelled. Then write example answers for each level of each question.
5. Write the probes. Decide in advance whether interviewers may follow up, how often, and with which questions, so every candidate gets the same chance to fill in a thin answer.
6. Pilot the questions on colleagues to check they are clear and draw a range of answers.
7. Write the interviewer's guide, with the competency definitions, questions, scales, example answers and probes, plus the common rating errors.
8. Keep records of how the interview was built: who helped, what they used, and how the questions and scales were tested.

Structured interview questions

Structured interview questions come in two kinds, and one interview can mix them. OPM and Google describe them the same way.

Both sets of figures are as reported in Levashina and colleagues (2014). The same review found that situational questions held up less well for high-complexity jobs in one meta-analysis, while past-behaviour questions did not.

Some rules for writing either kind, from OPM:

• Tie each question to a competency from the job analysis, and keep it realistic for the job.
• Make it open-ended, clear, free of jargon and at a reading level that suits the candidates.
• In behavioural questions, use superlatives ("the most difficult", "the last time") so the candidate picks one specific incident.
• Ask for the situation, the action and the result. Candidates often know this as the STAR method.

Google's version adds planned follow-ups to each question. For a behavioural prompt, "Tell me about a time your behavior had a positive impact on your team", the set follow-ups are "What was your primary goal and why? How did your teammates respond? Moving forward, what's your plan?"

The interview scorecard

A scorecard puts the scale in front of the interviewer while they rate. Google's re:Work guide describes rubrics that set out "what outstanding, solid, borderline, and poor response looks like" for each attribute a question tests. OPM's 2008 guide does the same with a five-level proficiency scale and example behaviours for each level.

A worked scorecard row

This is OPM's own example for the competency Interpersonal Skills and the behavioural question above. The level definitions are OPM's general ones; the examples are a selection from its table.

OPM tells interviewers to treat the examples "as a general guide (not an absolute)", because candidates draw on different experiences. Anchors at every point help: one study reviewed by Levashina and colleagues found that a five-point scale anchored at each point left ratings less open to disability bias than a scale anchored only at the ends, or not at all.

A scorecard template

Copy one row per competency; four to six rows is typical. Attach the rating scale for each competency, with its level definitions and example answers like the worked row above, so every interviewer scores against the same descriptions.

On weighting, OPM recommends equal weights for each competency. Its assessment FAQ calls this "generally the most effective and defensible course of action in the absence of a clear and documented rationale for doing otherwise." If one competency must count for more, write the reason down before the interviews.

Running the interview and using the scores

Scores can be compared only when every candidate gets the same interview. OPM's guide asks for the same questions, the same amount of time, and similar probes for every candidate. It also says interviewers should not look at a candidate's CV or other documents during the interview, because a strong CV can lift the ratings of the answers.

Take notes on what the candidate says as they say it. Rate each answer right after the interview, while the notes are fresh, and quote the answer in the evidence column. OPM wants ratings "tied to specific behavioral examples". When several people interview together, OPM's FAQ says the panel must come to a consensus on its ratings.

OPM's guide asks interviewers to know the common rating errors:

Source: OPM (2008), Appendix F.

When every candidate has been rated, OPM asks the interviewer to review the ratings, check each one is tied to a behavioural example, and sign and date the forms. Keep them with the questions and notes: they are the record of why each candidate was or was not hired.

Hong Kong and Singapore notes

The guidance in both places matches what a structured interview already does: job-related questions, the same criteria for everyone, and written records.

In Hong Kong, the Equal Opportunities Commission's Code of Practice on Employment under the Sex Discrimination Ordinance recommends that employers "only ask questions at job interviews that relate directly to the essential requirements of the job". Where hours or travel matter, it says interviewers should discuss them "without questions about marital status, children and domestic obligations". It advises keeping interview records that show why applicants were or were not appointed for not less than 12 months, and recording each interviewer's assessment immediately after the interview.

In Singapore, TAFEP's guidance on conducting job interviews tells employers to create a list of selection criteria and apply it "consistently for all candidates", to ask questions "directly related to the selection criteria", and to keep a record of the interview, the assessment and any offer for at least one year. It also suggests an interview evaluation form and, where possible, more than one interviewer. The Workplace Fairness Act will prohibit adverse employment decisions based on protected characteristics, and TAFEP's April 2026 article on the Act lists screening applicants and inviting candidates to interview among the decisions it covers. The Ministry of Manpower said on 4 November 2025 that it aims for the Act to take effect at the end of 2027. TAFEP's advice is to support hiring decisions with "documented, job-related reasoning", which is what a completed scorecard holds.

The rules for AI in hiring in both places are covered in what is an AI interview. This is general information, not legal advice; check the official source for your case.

How ROUND.1 fits

ROUND.1 is a hiring marketplace for Hong Kong and Singapore, built by h.work. Each candidate takes one AI first-round video interview in a web browser, with a camera and a microphone. It lasts about 20 minutes by default (roles can be shorter), 30 at most, and each role is interviewed in one language, English or Mandarin Chinese.

The interview is scored competency by competency, and each score is backed by quotes from the transcript, the same evidence a scorecard's notes column asks for. Accent, pronunciation, grammar, fluency, personality traits and protected characteristics (such as age, gender, race, religion, disability or marital status) are not scored. The same interview is reused for the other roles the candidate chooses.

A company first sees the scored evaluation without the candidate's name or contact details, which open when it shortlists the candidate after the interview. A person at the company decides whom to hire. The research figures on this page describe structured interviews in general; none of them measures ROUND.1's interview.

The full list of facts and limits is on ROUND.1, explained. The candidate's side of a recorded interview is in one-way video interviews.