How to Run Structured Interviews: Plan, Questions, Scorecard and the Evidence

A structured interview means every candidate for a role answers the same job-related questions, every interviewer rates the answers on the same written scale, and the decision is made against criteria agreed before the first interview. Run it in four steps: fix four to six criteria from the job's real requirements, write one behavioral or hypothetical question per criterion with 1-4 anchors, time-box the conversation so every criterion actually gets asked, and collect each interviewer's scorecard before anyone talks. Below is the 60-minute plan, a filled four-criterion scorecard and the research that justifies the effort.

This replaces the April version of this article, which had the theory and none of the materials.

What is a structured interview, and why is it worth the discipline?

Google's re:Work guide to structured interviewing defines it as "using the uniform methods to assess candidates applying for the same job": the same questions, graded on an identical scale, with decisions resting on predetermined qualifications. The guide recommends two question types, behavioral ("Tell me about a time...") and hypothetical ("Imagine if..."), and recommends "standardized rubrics so that all reviewers have a shared understanding" of what an outstanding, solid, borderline and poor answer looks like.

The research case is stronger than most vendor glossaries make it. Sackett, Zhang, Berry and Lievens (2022), in the Journal of Applied Psychology, re-analysed the classic meta-analytic estimates after showing that earlier range-restriction corrections had systematically inflated them. Their Table 3 puts the operational validity of structured interviews at .42, the highest of every selection procedure they examined, against .19 for unstructured interviews. Job knowledge tests came in at .40, empirically keyed biodata at .38, work samples at .33 and cognitive ability tests at .31. The older figures from Schmidt and Hunter (1998), .51 for structured and .38 for unstructured interviews, are reproduced in the same table as the estimates being corrected; treat them as the historical numbers, not the current ones.

Two things in that table matter for practice. First, the gap between structured and unstructured is not a rounding difference: the same hour with the same candidate predicts roughly twice as well when the questions and the scale are fixed in advance. Second, structure now sits above the work sample and the ability test, which were the methods people used to recommend over interviews. The interview is not the weak link in selection. The unstructured interview is.

Google's own internal results, as the guide reports them: structured interviews saved interviewers about 40 minutes per interview on average, and rejected candidates who had a structured interview were 35% happier with the process than those who did not. Those are one company's numbers and should be read that way.

How many teams actually run structured interviews?

Fewer than say they do. Survale's 2025 CandE Benchmark Research (vendor data; more than 66,000 candidate surveys and 110 registered organisations) found that 65% of North American employers in the research used a structured interview process, and that among the employers who won a CandE award the figure was 71%. The winners also conducted structured interviews 9% more often than all participating employers combined. The same research names time disrespected during interviews and appointments as the top reason candidates withdrew in 2025, cited by 32% of those who did, which is the candidate-side cost of an interview loop that nobody planned.

The load is not small. Ashby's EMEA Recruiter Productivity report (vendor data; over 180,000 EMEA jobs from January 2021 to June 2026) has EMEA candidates going through 3.9 interview events on average before a hire as of Q1 2026, up from 3.6 in 2021, and spending between 2.4 and 2.9 hours interviewing before receiving an offer. Only 6.7% of applications reach an interview at all, and offer conversion from interview sat in the 6-8% range as of early 2025. Every one of those hours is either producing comparable evidence or producing an impression. The plan below is how to make it the former.

How do you choose the criteria and write the questions?

Start from the job, not from a question bank. Take the job brief and extract the four to six things that separate someone who succeeds in this role from someone who merely qualifies for it. Skills that can be verified from a CV or a test (a certification, a language level) do not need interview time. Behaviors do.

For each criterion, write one primary question and two follow-ups. Behavioral questions ask for a real past situation: "Tell me about the last time a release you owned went wrong in production." Hypothetical questions put the candidate in a realistic scenario: "Imagine a client moves the deadline forward by two weeks on Monday. Walk me through your first day." The follow-ups are where the evidence is: "What did you personally do?", "What would you do differently?", "How did you know it had worked?"

Then write the anchors. An anchor is a one-line description of what a 1, 2, 3 and 4 answer looks like for that specific criterion. Without anchors, a 3 from one interviewer and a 3 from another describe different candidates. We use a four-point scale on purpose: there is no middle to hide in, so an interviewer has to decide whether the answer met the bar or not.

If the interviewers have not seen the criteria, the questions and the anchors before the interview starts, none of this works, which is the subject of the interview prep problem every recruiter ignores.

What does a 60-minute structured interview plan look like?

This is the plan we use for a single-interviewer 60-minute round with four criteria. The minutes are a template, not a measurement; the point is that every criterion has a protected slot, so the fourth one does not get three minutes because the first ran long.

Minute	Block	What happens	What you write down
0-3	Opening	Names, format, that you take notes, that the last ten minutes are theirs	Nothing yet
3-8	Role context	Two or three sentences on the role and the team. No selling, no "tell me about yourself"	Nothing yet
8-20	Criterion 1: technical depth	Primary behavioral question, two follow-ups	Evidence line for criterion 1
20-32	Criterion 2: problem solving under constraints	Primary hypothetical question, one follow-up	Evidence line for criterion 2
32-42	Criterion 3: working with others	Primary behavioral question, one follow-up	Evidence line for criterion 3
42-50	Criterion 4: ownership	Primary behavioral question, one follow-up	Evidence line for criterion 4
50-57	Candidate's questions	They ask, you answer honestly	Note the questions; they are evidence too
57-60	Close	Next steps and the date they will hear from you	The date you just promised
After	Scorecard	10-15 minutes blocked in the interviewer's calendar, same day	Scores, evidence, recommendation

Three rules make the plan hold. The interviewer asks the primary question in the same words for every candidate; the follow-ups may vary with the answer. A criterion that runs out of time gets a 2 with the note "not enough evidence", not a guessed 3. And the scorecard slot after the interview is a calendar block, not an intention; the interview is not over until it is written.

For a loop with several interviewers, split the criteria across them rather than having everyone cover everything. Two interviewers asking the same criterion produce one duplicated impression; two interviewers asking different criteria produce twice the evidence. The scheduling side of multi-interviewer loops, including remote panels, is in how to run structured interviews across time zones.

What does a filled four-criterion scorecard look like?

First the anchors, written for a backend engineer role. Your criteria will differ; the shape should not.

Criterion	1: Below the bar	2: Partial	3: Meets the bar	4: Clearly above
Technical depth	Describes tools, not decisions; cannot explain a trade-off they made	Explains one decision but not its alternatives or cost	Explains a real decision, the alternative considered and why they chose it	Does that, and names what they would change now and why
Problem solving under constraints	Jumps to a solution with no questions; ignores the constraint	Asks a question or two, then follows a template	Clarifies the goal, names the constraint, proposes a first step and a fallback	Does that, and states what they would measure to know the first step worked
Working with others	Conflict is always someone else's fault; no example of changing their mind	Example exists but is vague on their own role	Specific situation, their own action, the outcome, what the other person needed	Does that, and shows they changed their approach because of it
Ownership	Waited for instruction; escalated without a proposal	Took the task but not the result; stopped at the handoff	Owned the result past the handoff, followed up, closed the loop	Does that, and fixed the process so it does not recur

Now the same card, filled in for an illustrative candidate (not a real person) after a 60-minute round. Score, evidence, reasoning: all three, every row.

Criterion	Score	Evidence from the interview	Must-pass?
Technical depth	3	Asked about the last production incident they owned: described a queue backlog after a schema migration, explained why they chose to replay from the dead-letter queue rather than roll back (rollback would have lost 40 minutes of writes), named the alternative. Did not volunteer what they would change now; when asked, said "add a canary", without specifics.	Yes, passed
Problem solving under constraints	4	Hypothetical: deadline moved forward two weeks. Asked which feature the client would drop before proposing anything, then proposed shipping the import path first with the export path stubbed, and said they would check daily active imports after three days to confirm the cut was right. Named the fallback (manual export by support for two weeks).	Yes, passed
Working with others	2	Asked for a disagreement with a colleague about a technical decision. Example was a code-review argument; described the other person's position fairly but could not say what they themselves did differently afterwards. Second follow-up got "we just moved on". Enough to show no red flag, not enough to show the behavior.	No
Ownership	3	Described taking over a flaky integration test suite nobody owned, fixing the three worst tests and writing the runbook. Followed up two weeks later to check the suite was still green. Did not change the process that let the suite rot, which is the difference to a 4.	No

Overall: 12 of 16. Recommendation: advance to the final round. Concern to probe next: working with others; the next interviewer should take a second, different example of a disagreement before this goes to offer.

Three things to notice. The 2 is not a rejection; it is a flag with an instruction for the next interviewer. The evidence fields quote what the candidate said, not what the interviewer felt. And the must-pass column means a 1 on either of the first two criteria ends the process regardless of the total, which is written down before the interview, not negotiated after it. The hiring manager's side of this card, including what to do when the criteria do not match what was actually asked, is in the hiring manager's guide to interview scorecards.

How do you score and debrief without anchoring on the loudest voice?

Collect first, discuss second. In Noise: A Flaw in Human Judgment, Kahneman, Sibony and Sunstein argue that group discussion increases noise and recommend collecting individual opinions before any meeting, and sequencing information so that each judge sees only what they need when they need it. Applied to interviews: every interviewer submits their scorecard before seeing anyone else's, and the debrief opens with the disagreements, not with the most senior person's summary. If the technical interviewer gave a 2 where the hiring manager gave a 4, that gap is the most useful ten minutes of the meeting.

The submission step is where the practice breaks. Ashby's Recruiting Coordination report (vendor data; October 2024, over 2.8 million candidates and 5.1 million interview events) puts average scorecard submission rates between 84% and 89%, with most feedback submitted within a median of about two hours, and both figures flat at a global level from Q1 2021 to Q3 2024. The flat line is the finding: reminders and policy have had years to move it and have not. If one scorecard in eight never arrives, the debrief is running on seven interviewers' evidence and one interviewer's memory, and that interviewer is usually the one whose opinion carries the most weight.

Two mechanical fixes work better than reminders. Scorecards are required before a candidate can be moved to the next stage, so the gap becomes visible to the recruiter the same day. And interviewers cannot read anyone else's card until they have submitted their own, which removes the incentive to wait for the senior view.

What breaks structured interviews in practice?

From our work at ScalingPPL, the failures are rarely in the design. They are a few weeks in.

• Drift. The primary questions get paraphrased, then replaced by the interviewer's favourites. Fix: the question set is in the interview invite, and the recruiter spot-checks transcripts or notes against it.
• Anchors nobody reads. Interviewers rate on gut and write a 3. Fix: the anchor text sits next to the score field, not in a separate document, and a calibration session with two sample answers per criterion runs before the first real interview, which is also what the re:Work guide lists as "interviewer training and calibration".
• The criteria change between candidates. Someone adds a "culture" criterion after candidate two. Fix: criteria are frozen when the role opens; a new criterion means re-interviewing everyone on it or dropping it.
• No evidence, only scores. "Technical depth: 3" with nothing under it is not a scorecard, it is a vote. Fix: an empty evidence field blocks submission.
• The interview that nobody prepared for. The interviewer reads the CV in the lobby. Fix is upstream, in the preparation step, and it is the recruiter's job to make it cheap.

None of these need software. All of them need someone who owns the process and checks it weekly.

Where does software help, and where does it not?

Software does not make an interview structured; a hiring team does. What it can do is remove the effort that makes teams quietly stop. In Pickr, the question set is drafted from the job's requirements and the candidate's background before the interview, and the interviewer edits, reorders or drops questions, so the set is a starting point rather than a script. The conversation is transcribed live, so nobody is taking notes instead of listening, and when two interviewers remember an answer differently the transcript settles it. After the interview, every criterion on the scorecard already carries the relevant passages from the transcript; the interviewer confirms, corrects or disagrees, which takes minutes rather than an evening. The human still evaluates and decides. When a scorecard is outstanding, Pickr nudges the interviewer rather than the recruiter chasing by email. Names and email addresses are redacted from AI prompts by default, and candidate data runs on EU hosting in Frankfurt.

What it will not do is choose your criteria, write your anchors or make a hiring manager accept that an unstructured conversation is the .19 in that table, however experienced the person running it. Those are the parts of this article you have to do yourself, and they are the parts that move the number from .19 to .42.

Sources

• Google re:Work, A guide to structured interviewing for better hiring practices: https://rework.withgoogle.com/intl/en/guides/a-guide-to-structured-interviewing-for-better-hiring-practices (accessed 2026-10-14)
• Sackett, P. R., Zhang, C., Berry, C. M. & Lievens, F. (2022), Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range, Journal of Applied Psychology 107(11), 2040-2068, Table 3: https://doi.org/10.1037/apl0000994 (accessed 2026-10-14)
• Schmidt, F. L. & Hunter, J. E. (1998), The validity and utility of selection methods in personnel psychology, Psychological Bulletin 124(2), 262-274, figures as reproduced in Sackett et al. (2022), Table 3: https://doi.org/10.1037/0033-2909.124.2.262 (accessed 2026-10-14)
• Survale, 2025 Candidate Experience (CandE) Benchmark Research, key statistics: https://survale.com/2025-candidate-experience-cande-benchmark-research-key-statistics/ (accessed 2026-10-14)
• Ashby Talent Trends, EMEA Recruiter Productivity (9 September 2026): https://www.ashbyhq.com/talent-trends-report/reports/emea-recruiter-productivity (accessed 2026-10-14)
• Ashby Talent Trends, Recruiting Coordination (22 October 2024): https://www.ashbyhq.com/talent-trends-report/reports/recruiting-coordination (accessed 2026-10-14)
• Kahneman, D., Sibony, O. & Sunstein, C. R. (2021), Noise: A Flaw in Human Judgment, Little, Brown Spark; Publishers Weekly review: https://www.publishersweekly.com/9780316451406 (accessed 2026-10-14)
• Pickr, Interview Intelligence feature page: https://pickr.dev/features/interviews (accessed 2026-10-14)