Skip to content

Structured interview best practices for hiring teams.

Structured interview best practices backed by research: job-related questions, the same for everyone, scoring against anchored criteria, limiting interview bias, and a checklist.

Updated 17 November 2026 · 15 min read

On this page

In short

  • Structure predicts performance. A 2022 re-analysis ranked structured interviews the top selection procedure, at about twice the validity of unstructured ones.
  • Same questions, from the job: base them on a job analysis and ask every candidate the same ones, with planned follow-ups.
  • Set the length from the content: length doesn't predict performance, structure does; a screen takes 20 to 40 minutes, a technical round up to an hour.
  • Score each answer against anchored criteria right after it, with notes on what the candidate said and did.
  • Combine by formula: average several raters' scores before discussing; formulas beat combining in your head.
  • Structure limits bias from first impressions, halo and similarity, but doesn't remove it; keep questions job-related and offer adjustments.

Why structure matters

Structured interview best practices come down to a simple idea: ask every candidate the same job-related questions and score the answers against criteria you wrote in advance. Research shows it predicts job performance better than a free conversation and leaves less room for bias. Here are the evidence, eight practices for any format and a checklist.

What a structured interview is

In a structured interview every candidate for a role gets the same job-related questions, and every answer is scored against the same criteria, written before the first interview. In an unstructured interview the interviewer improvises: different questions for different people, judged on an overall impression.

Structure isn't all or nothing. Campion, Palmer and Campion (1997) described 15 components of structure. As the paper is usually summarized, they split between the content of the interview and how it's evaluated:

ContentEvaluation
Questions based on a job analysisRate each answer, or use several rating scales
The same questions for every candidateAnchored rating scales
Better types of questions (past behavior, situational)Detailed notes
Limited prompting and follow-upsSeveral interviewers
Longer interviews, or more questionsThe same interviewers for every candidate
No extra information (such as the CV) during the interviewNo discussion between interviews
No questions from the candidate until the endInterviewer training
Combine ratings statistically, not by discussion

You don't need all 15. The practices below have the clearest evidence.

Structured vs unstructured interviews: what the evidence says

Validity here means the correlation between interview ratings and later job performance (0 is none, 1 is perfect).

EstimateStructuredUnstructured
Schmidt and Hunter (1998).51.38
Sackett, Zhang, Berry and Lievens (2022).42about .19
  • Structured interviews came out on top. The 2022 re-analysis found older studies overcorrected, so most validities fell. Structured interviews emerged as the top-ranked selection procedure, at .42 (with a wide range, .18 to .66, across settings). Unstructured interviews fell to about .19, roughly half the structured figure.
  • The older numbers were higher, but the ranking held: in Schmidt and Hunter's review, structured interviews matched ability tests at .51.
  • Structure makes interviewers agree. A meta-analysis of 111 reliability studies (Conway, Jako and Goodman, 1995) found that standardized questions, standardized evaluation of answers and combining ratings all raise reliability.

Asking the right questions

1. Start from a job analysis

Before writing a question, write down what the job needs: the main tasks and the skills needed on day one. The US Office of Personnel Management calls job analysis "the foundation for all assessment and selection decisions"; it also documents that your questions relate to the job, as the US Uniform Guidelines (under revision) expect.

A light version works: ask the hiring manager what separates a good hire after six months, and turn it into three to six must-haves and a few nice-to-haves. For a customer support specialist, a must-have might be "collects screenshots and steps to reproduce before escalating a bug". See generate interview questions from a job description and must-haves vs nice-to-haves.

2. Ask every candidate the same questions

Same questions, same order, same time: the only way to compare answers rather than conversations. If you interview in more than one format, keep one format for everyone at a stage: the mode of an interview changes ratings on its own (see AI interview vs one-way video interview).

3. Use past-behavior and situational questions

  • Past behavior: "Tell me about a time you had to explain a technical problem to a customer who wasn't technical. What did you say?" The candidate describes what they actually did.
  • Situational: "A customer says the export button stopped working after yesterday's update, and they need the report within the hour. What do you do?" The candidate describes what they would do.

Both work. With anchored rating scales, Taylor and Small (2002) found mean validities of .63 for past-behavior questions and .47 for situational ones, and Huffcutt and colleagues (2004) found situational questions lose validity for highly complex jobs while past-behavior questions don't. Lean on past behavior for complex roles, situational questions for people with little experience. For skills you can see, add a task: a short coding exercise, a whiteboard design or a role-play interview.

4. Plan the follow-ups

Follow-ups get past a vague first answer ("What exactly did you ask the customer?"), but unlimited probing makes the interview unstructured again. Research on structure treats limited, planned follow-ups as part of it (Levashina and colleagues, 2014), and Google's guide writes follow-ups for each question in advance. Write two or three per question, and note when a candidate needed a hint.

How long should a structured interview be?

As long as its questions need, and no longer. A meta-analysis by Thorsteinson (2018) found that interview length was unrelated to both reliability and validity: a longer interview isn't a better one, a better-structured one is. So set the length from the content: count the questions and tasks you'll score, give each enough time, and add nothing else.

Length also costs candidates less than people fear. In HireVue's analysis of assessments taken by nearly 30 million applicants (2025 white paper), length did not predict completion: most dropout happens in the first 5 to 10 minutes, and cutting a 45-minute assessment to 30 was expected to raise completion by only 1 to 2%. Criteria Corp's data on hundreds of thousands of tests (2022) shows a high completion rate, around 80%, up to 40 minutes early in the process, with more drop-off beyond. What matters more is that each part is clearly relevant to the job and that candidates know the length before they start.

Match the length to the round

RoundTypical lengthWhat it covers
First screen20–40 minutesThree to five must-haves: motivation, core experience, one or two job situations.
Technical or skills round45–60 minutesOne or two tasks, such as a coding exercise, a whiteboard design or a written case, with questions about the candidate's choices.
Role-play or case round30–60 minutesA realistic conversation, such as a sales call or a support escalation, then questions about how it went.
Final human conversation15–30 minutesConfirm what the earlier rounds found, explore open points from the reports, answer the candidate's questions.

We suggest keeping each session to about an hour, so candidates stay fresh and can fit it into their day. If a role needs more, split it into rounds the candidate takes separately, each with its own topics, rather than one long sitting.

Several structured rounds, then a short human interview

A process can run most of its rounds as structured AI interviews and keep the team's time for the end: for example a 30-minute screen, then a 60-minute technical round, then a 20-minute conversation with the hiring manager. The structured rounds give every candidate the same questions and give the team the evidence; the human interview checks it, asks about what the reports flagged, and lets the candidate meet the people they'd work with. The decision stays with the team.

More rounds aren't automatically better either. Google's people analytics team reported that four interviews were enough to predict a hiring decision with about 86% confidence, and each extra interview added little (as described by Laszlo Bock in Work Rules!). Give each round topics the others don't cover, so no round repeats another.

Practical guidance

  • Budget per question: about 4 to 6 minutes for a conversation question with its follow-ups, 15 to 30 minutes for a coding or design task, 10 to 20 for a role-play, and about a minute per quiz question.
  • Cut what you won't score. Every question should map to at least one must-have or nice-to-have; drop the ones that don't, rather than adding time.
  • Tell candidates the length up front, both the total and the time per part. A realistic estimate, slightly generous rather than short, builds trust and lets them plan a quiet slot.
  • Use the same time limits for everyone in a round, and add time only as an accommodation for a candidate who needs it (see fair AI interviews).
  • Check the real durations. If most candidates run out of time on one part, the part is too long or the question too broad; if most finish early, shorten it.

Interview scoring: rating answers, not people

5. Score against anchored criteria

An anchored scale describes what an answer at each level sounds like, so ratings mean the same to every interviewer. The idea goes back to Smith and Kendall (1963); with anchored scales, interviewers in Taylor and Small's review agreed closely (reliability around .77 to .79). Write the anchors before the first interview.

Rating"Gets the facts before escalating a bug"
StrongAsks for a screenshot, the steps and the device or browser, tries to reproduce it, then escalates with a clear write-up.
AcceptableAsks what went wrong and escalates, but leaves out details the engineers will need.
WeakEscalates at once with "export is broken", or tells the customer to clear their cache and closes the case.

6. Rate each answer right after it

Rate each answer as soon as it ends, before talking to anyone. Rating at the end of the day lets a general impression overwrite the details. The OPM has panel members rate individually, then resolve differences.

7. Write down what was said

Notes help memory more than judgment: Middendorf and Macan (2002) found note-taking improved recall but not the accuracy of judgments, while reviewing notes before rating helped. What you write matters: Burnett and colleagues (1998) linked notes on what the candidate said and did to higher validity, and notes on the procedure to lower. Write "Asked for a screenshot of the error and the browser version", not "good communicator".

8. Use several raters, and combine by formula

Conway and colleagues found several ratings helped when combined mechanically, with no evidence of benefit when combined subjectively. Kuncel and colleagues (2013) found the same for combining assessment data in general: a fixed formula predicted job performance at .44 against .28 for experts combining the same data in their heads. Average first, then discuss only large gaps.

Interview bias, and how structure limits it

  • First impressions. In one study of structured interviews, ratings of the rapport-building opening, the small talk before the questions, correlated with the final interview ratings (r = .42) and with job offers (Barrick, Swider and Stewart, 2010). Part of it was early judgments of competence. Start the scored questions quickly and rate each answer on its own.
  • Halo. A general impression spills over into ratings of unrelated traits, and people don't notice it happening (Nisbett and Wilson, 1977, a classic experiment outside hiring). Separate criteria, each rated against its own anchors, give it less room.
  • Similarity. Interviewers may favor people like themselves. In a highly structured interview of about 20,000 applicants for a managerial job, McCarthy, Van Iddekinge and Campion (2010) found no effects of applicant gender or race on ratings, and no effect of applicant–interviewer similarity. That's one organization: structure can hold these effects down, not remove them everywhere.
  • Group differences. Across 31 studies, Huffcutt and Roth (1998) found that high-structure interviews showed smaller rating differences between Black, Hispanic and White applicants than low-structure ones.

Structure alone doesn't make an interview fair. Never ask about health, family, age, religion or origin, and offer adjustments: see interview accommodations.

Candidate experience in a structured interview

  • Tell candidates what you assess. In two studies, telling interviewees the dimensions being assessed improved their performance and measured the intended dimensions more cleanly, without hurting prediction (Klehe and colleagues, 2008). Share the topics, not the expected answers.
  • Explain the format. Say how long it takes, that everyone gets the same questions, and when they'll hear back. Google reports that rejected candidates who had a structured interview were 35% happier (re:Work; Google's internal data).
  • Move fast and close the loop. In the CandE Benchmark 2025 (the program's own summary of more than 66,000 candidate surveys), 56% of candidates got no feedback after being rejected, and resentment rose by 70% when there was no interview preparation or follow-up. A short, honest message beats silence.

For invitation wording and preparation tips, see how to invite candidates to an AI interview.

Structured interview checklist

Before, during and after

  1. Analyze the job. List the main tasks and what a good hire shows after six months.
  2. Write must-haves and nice-to-haves related to the job, three to six must-haves at most.
  3. Write the questions: past-behavior, situational or a task, each tied to one or more criteria, with two or three planned follow-ups.
  4. Write the anchors: strong, acceptable and weak answers for each criterion.
  5. Set the length per round from its content: about 20 to 40 minutes for a screen, up to an hour for a technical round, a time limit per part, the same for everyone.
  6. Tell candidates the format, length and topics; offer adjustments.
  7. Rate each answer right after it, with notes on what the candidate said and did.
  8. Combine the ratings by formula, then discuss only large gaps between raters.
  9. Decide and record why, with the evidence, and tell every candidate the outcome.

How testee applies these practices

PracticeIn testee
Job-related questionsDraft from a job description, a role title, a template or the question library; your team checks it.
Same questions for everyoneThe same blocks, questions and time limits for every candidate.
Set criteriaEach block has must-have and nice-to-have expectations, hidden from candidates.
Limited follow-upsThe interviewer asks follow-ups within the block's time; when it gives a hint, the status says "covered after prompt".
Evidence for each ratingEach expectation is marked covered, covered after a prompt, partial, incorrect or not covered, with the candidate's quote and its moment in the recording.
Combining by formulaThe coverage score is computed in code from the statuses, never by the AI, the same way for everyone. It's for review, not a hiring decision.
Several ratersReviewers comment and can change any status; the team decision needs a note.
FairnessNo emotion, voice, face or accent analysis; text mode; extra time on request.

A person always decides: the AI never makes a hiring decision or rejects anyone on its own. To send every applicant for a job the same interview automatically, see AI interview ATS integration.

Summary

  • Structure predicts. In the 2022 re-analysis structured interviews were the top-ranked selection procedure (.42), about twice the unstructured figure.
  • The questions: from a job analysis, the same for everyone, past-behavior or situational, with planned follow-ups.
  • The scoring: anchored criteria, each answer rated right after it, notes on what was said, several raters combined by formula.
  • The bias: first impressions and halo have less room when each answer is rated against set anchors.
  • The candidate: tell them what you assess and how it works, move fast and tell everyone the outcome.

Questions and answers

How long should a structured interview be?

As long as its questions need: about 20 to 40 minutes for a first screen, 45 to 60 minutes for a technical or skills round, and 15 to 30 minutes for a final conversation with the team. Research found that length doesn't predict how reliable or valid an interview is, and that most candidates who drop out do so in the first minutes, so set the length from what you'll score and tell candidates up front.

What is a structured interview?

An interview where every candidate for a role gets the same job-related questions and every answer is scored against the same criteria, written before the first interview.

What are examples of structured interview questions?

Past behavior: "Tell me about a time you had to explain a technical problem to a customer who wasn't technical. What did you say?" Situational: "A customer says the export button stopped working after yesterday's update, and they need the report within the hour. What do you do?" Each comes with planned follow-ups and an anchored scale.

What are the advantages of structured vs unstructured interviews?

Structured interviews predict job performance better (.42 against about .19 in the 2022 re-analysis by Sackett and colleagues), interviewers agree more, and group differences in ratings are smaller.

What is a competency-based interview?

A structured interview built around competencies, such as customer focus or problem solving, usually with past-behavior questions ("Tell me about a time…") scored against what each competency looks like at each level. The term is common in the UK.

What questions should you ask candidates in a first-round interview?

Two or three questions on the must-haves for the role, each tied to set criteria, plus a short task where the job calls for it. Leave logistics such as salary and start date to the application form or a short call.

Sources

  1. Schmidt and Hunter (1998), The validity and utility of selection methods in personnel psychology
  2. Sackett, Zhang, Berry and Lievens (2022), Revisiting meta-analytic estimates of validity in personnel selection
  3. Sackett, Zhang, Berry and Lievens (2023), Revisiting the design of selection systems in light of new findings regarding the validity of widely used predictors
  4. Campion, Palmer and Campion (1997), A review of structure in the selection interview
  5. Conway, Jako and Goodman (1995), A meta-analysis of interrater and internal consistency reliability of selection interviews
  6. Levashina, Hartwell, Morgeson and Campion (2014), The structured employment interview
  7. Taylor and Small (2002), Asking applicants what they would do versus what they did do
  8. Huffcutt, Conway, Roth and Klehe (2004), The impact of job complexity and study design on situational and behavior description interview validity
  9. Smith and Kendall (1963), Retranslation of expectations
  10. Middendorf and Macan (2002), Note-taking in the employment interview
  11. Burnett, Fan, Motowidlo and DeGroot (1998), Interview notes and validity
  12. Kuncel, Klieger, Connelly and Ones (2013), Mechanical versus clinical data combination in selection and admissions decisions
  13. Barrick, Swider and Stewart (2010), Initial evaluations in the interview
  14. Nisbett and Wilson (1977), The halo effect
  15. McCarthy, Van Iddekinge and Campion (2010), Are highly structured job interviews resistant to demographic similarity effects?
  16. Huffcutt and Roth (1998), Racial group differences in employment interview evaluations
  17. Thorsteinson (2018), A meta-analysis of interview length on reliability and validity
  18. HireVue (2025), Applicant dropout: completion rate drivers and consequences
  19. Criteria Corp (2022), How much testing is OK?
  20. ERE (Dr. John Sullivan), As Google found out, it takes a lot of work to really hire smarter
  21. Klehe, König, Richter, Kleinmann and Melchers (2008), Transparency in structured interviews
  22. US Office of Personnel Management, Structured interviews
  23. US Office of Personnel Management, Job analysis
  24. Uniform Guidelines on Employee Selection Procedures, 29 CFR 1607.14
  25. Google re:Work, A guide to structured interviewing
  26. Survale, CandE Benchmark 2025 key statistics

Try it onyour next role.

Build a structured AI interview from your job description in minutes.14 days free, no card needed.