A good interview question is a short, open, non-leading prompt built around one topic, phrased so a witness can answer from memory before you narrow it. Use open forms (“Tell me how…”) for recall first, then specific closed questions to close gaps. Keep each question under about 20 words, ask one thing at a time, and never embed the answer you expect.

Interview questions fail in two directions. They fail as memory probes, when the wording pushes a witness toward a plausible story instead of a true one. They fail as evidence, when the answer cannot be compared across candidates, sessions, or interviewers. The science below treats the question as an instrument with measurable properties: cue strength, suggestibility load, and scoring reliability.

Most “how to ask better questions” content is a list of question stems. That is thin because it skips the mechanism. A question is a retrieval cue, and the wording of the cue changes what gets retrieved. Psychologists have studied this for decades under the labels of questioning, eyewitness memory, and structured interviewing, and the findings are consistent enough to set practical thresholds.

Retrieval cue mechanics between question and memory

Human memory does not store events like video files. It reconstructs a record at the moment of retrieval, from fragments plus expectations plus the cues in the room. A question is the largest controllable cue in an interview. Its wording activates related concepts, and the activated concepts bias what the witness assembles.

This is why question form matters more than most interviewers assume. Say “Tell me what happened” and you activate a broad autobiographical search. Say “Was the man wearing a hat?” and you have already introduced a man and a hat, whether or not either existed. The witness now has to reject them rather than generate them, and rejection is harder than generation.

The encoding–retrieval match principle sharpens this. Retrieval works best when the cue overlaps with how the information was originally encoded. A witness who saw a chase encodes motion, sound, and spatial layout, not a checklist of facts. Questions built around those encoded dimensions produce richer, more accurate recall than questions built around a form the interviewer wants to fill.

Timing changes the mechanism too. Immediately after an event, memory is detailed but pliable, because the witness has not yet rehearsed a version. Days later, memory is more stable but thinner, because details fade and reconsolidation blends sources. The practical consequence: ask open recall questions early, and treat the first account as the anchor against which later specifics are checked.

Wording also sets the witness’s response threshold. A question that implies the interviewer already knows something raises the pressure to confirm it. A question that signals “I don’t know, you tell me” lowers it. Small phrasing changes shift answers measurably, which is why the same witness can give different accounts to two polite interviewers using different scripts.

Quantified context: how much question wording moves answers

The classic demonstration is the verb manipulation study, in which witnesses to a car crash were asked how fast cars were going when they “contacted,” “hit,” or “smashed” each other. Estimated speeds rose with verb intensity, and a week later more participants asked with “smashed” reported broken glass that was never shown. The wording changed both the number and the memory.

The misinformation effect literature generalizes this. Roughly a third of participants in typical post-event misinformation designs adopt a suggested detail into their later reports, with rates varying widely by delay, source credibility, and whether the suggestion is repeated (Loftus, 2005). Some studies report adoption rates above half under high-pressure conditions. Treat one-third as a planning baseline, not a fixed constant.

Field interviewing research gives parallel numbers. Evaluations of the cognitive interview, a structured recall protocol, report increases in correct detail of roughly 25 to 50 percent over standard police questioning, with modest increases in errors and confabulation (Memon, Meissner & Fraser, 2009). The gain comes mainly from open recall prompts and from giving the witness time, not from clever closed questions.

Question length has a weaker evidence base, so treat thresholds as practitioner heuristics rather than measured constants. In training settings, questions beyond roughly 20 words are harder to answer, harder to score, and more likely to contain two questions. That is an operational rule, not a finding to cite.

For structured hiring, the strongest quantitative anchor is validity research on structured interviews. Structured interviews, with the same scored questions asked in the same order, outperform unstructured “tell me about yourself” conversations for predicting job performance, according to the meta-analytic summary in Schmidt and Hunter’s work on selection methods. The mechanism is scoring comparability, not question cleverness.

Edge cases where standard advice reverses

The open-question rule is right for recall and wrong for verification. Once a witness has given a free account, targeted closed questions are appropriate and often necessary. “What color was the door?” is a legitimate question after the account exists. Applying “always ask open questions” mechanically produces long, unscoreable narratives and missing facts.

Children reverse another rule. Preschool-age children, roughly under 6, are highly susceptible to repeated and suggestive questioning, and open prompts must be kept simple and concrete. Research in developmental forensic interviewing (Lamb et al., 2018) shows that children can give accurate accounts with open prompts, but repeated “are you sure?” pressure degrades accuracy. The practical adjustment is more patience and simpler wording, not more closed questions.

Cross-cultural interviewing reverses the assumption that silence means evasion. In many cultural and institutional settings, a pause is respectful deliberation. Interviewers trained to fill silence with a rephrased question often end up suggesting an answer. Wait roughly 5 to 10 seconds after a witness stops before speaking, especially on emotionally loaded topics.

High-stakes stress reverses the value of elaboration. In safety-critical post-incident interviews, the first goal is a factual timeline, and creative prompts can introduce error. A checklist-style scaffold, asked as open prompts (“Walk me through the sequence from the alarm to your arrival”), serves better than free association.

The counterintuitive professional caveat: more training in questioning does not reliably reduce suggestibility. Interviewers who learn the science sometimes become more confident in their own leading questions, because they believe they can detect influence. Detection of subtle suggestion is poor, which is why protocols and taped review matter more than intuition.

Common misconceptions about question quality

The first misconception is that a “good” question produces a good answer. Question quality and answer quality are separate properties. A well-formed open question can produce a vague answer from a tired witness, and that is information about the witness, not the question. Judge questions by their cue structure, not by the story they elicit.

The second is that leading questions are only a problem in court. Leading questions distort memory during the interview itself, which is why the effect survives into later interviews and even into courtroom testimony. The distortion happens at retrieval, so chronological order of the transcript does not protect you.

The third is that hypothetical questions are harmless. “What would you have done if…?” invites construction, not recall, and witnesses frequently blend constructed answers with genuine memory. If you need a hypothetical, label it explicitly and score it separately.

The fourth is that the best interviewer asks the fewest questions. Few questions reduce suggestion but also reduce coverage. The goal is a balance: broad open recall first, then narrow verification, with each narrow question tied to a specific gap in the account.

The fifth is that question lists must be followed in order. A rigid script protects comparability across candidates but can bury a witness’s natural narrative. Structured protocols handle this by fixing the core questions while allowing planned follow-ups, so scoring stays comparable and recall stays natural.

Format Typical length Best use Main risk Scoring reliability Suggestibility load
Broad open prompt 6–15 words Initial free recall Vague or short answers Moderate Low
Focused open prompt 10–20 words Expanding one episode Drifting off topic Moderate–high Low–moderate
Closed verification 3–10 words Confirming a specific detail Guessing to please High Moderate
Leading question 5–15 words Rarely justified Fabricated memory Appears high, is not High
Forced-choice pair 5–12 words Narrowing options, known choices False forced choice High if options verified Moderate–high
Hypothetical Any Training, policy design Blending with real recall Low High
Multiple questions at once 20+ words Never Confused, partial answers Low Moderate

The question-to-answer timeline

Question design is a sequence, not a single sentence. Each stage has a different job, and skipping stages moves the suggestibility cost onto the witness.

Stage Timing after event Question job Example prompt Failure mode if skipped
Free recall Minutes to hours Capture the account before contamination “Tell me everything you remember, in any order.” Early suggestions anchor the record
Structured recall Same session Reorganize memory by time or sense “Take me through it again, starting from the sound.” Detail stays untapped
Gap-filling open Same session Expand specific episodes “Tell me more about the moment the door opened.” Accounts stay thin
Closed verification Same session, after recall Confirm or reject specific details “Was the door open when you arrived?” Errors go unchecked
Cross-check Later session Compare to prior account “Earlier you said the door was closed. Walk me through that again.” Inconsistencies go undetected
Score and record Within 24 hours Rate against fixed criteria (Use a rubric, not a question) Impressions replace evidence
Review for suggestibility Within a week Audit transcript for leading wording (Audit checklist) Bias survives into decisions

The timeline matters because each suggestion narrows later options. If a gap-filling question introduces a detail before the witness has generated it, every later question inherits the contamination. Front-loading recall protects the record.

Practical actions: writing and testing questions

Start with a topic map, not a question list. Write down the five to seven things you need to learn, in the witness’s likely chronological order, and one open prompt per topic. This keeps coverage without scripting every sentence, and it makes gaps visible before the interview.

Draft each core question, then run four mechanical tests. First, count the question marks; if there are two, split the item. Second, count the words; if over about 20, simplify. Third, check for embedded assumptions (“When you saw the manager, did she…?” assumes the manager was seen). Fourth, check whether the expected answer is visible in the wording.

Convert closed questions to open forms unless verification is the goal. “Did you sign the form?” becomes “Tell me about the signing.” “How angry was he?” becomes “Describe what he did.” The conversion keeps the topic and removes the suggested emotion or event.

Pilot the list on a colleague who does not know the case. Ask them to answer without inventing, and note every question they ask you to repeat or misinterpret. A question that needs repeating in a pilot will need repeating under stress.

Record and score. Rate answers against predefined criteria, not against interviewer impressions. For hiring, structured interviews with anchored rating scales are the standard, and the evidence favors scoring each answer on a fixed scale rather than holistic judgment.

Warning: any question that contains the answer, a suspect, or a numeric estimate is leading. Warning: “Are you sure?” asked repeatedly pressures guessing and degrades accuracy. Warning: two questions in one sentence produce one answer and one lost data point. Warning: rapid-fire closed questions on emotional topics raise confabulation risk. Warning: untrained interviewers who believe they can spot suggestion still miss it.

Reconciling thresholds: the under-20-word rule is a readability heuristic; the 5-to-10-second silence rule is a training convention; the one-third misinformation adoption figure is a vary-by-design statistic. None of these are hard limits, and none should be quoted as measured thresholds. Use them as defaults you can justify and adjust with evidence from your own transcripts.

Limitations and what the evidence does not settle

The science of questioning is strongest on suggestibility mechanisms and on structured interview validity. It is weaker on exact numbers. Misinformation adoption rates range widely across studies, correct-detail gains from cognitive interviewing vary by field setting and training quality, and the effect of question length has not been isolated cleanly. Do not treat any single percentage as a reliable constant.

Culture, age, trauma, and interpreter use change the calibration of every rule above. Trauma-exposed witnesses may give fragmented accounts that look evasive, and children may need simpler prompts and shorter sessions. Interpreters add a layer of question transformation that can introduce or remove suggestion, so brief them on the question plan.

The deepest limit is detection. Interviewers cannot reliably tell when their own questions have contaminated an answer, which is why protocols, recording, and independent review carry more weight than self-assessment. A good question is not one that feels clever in the room; it is one that survives an audit of the transcript.

This article is general educational information about interviewing technique, not legal, forensic, or mental-health advice, and it does not promise any outcome in an investigation, hiring process, or clinical setting. It does not provide diagnostic or treatment guidance; anyone conducting investigative or clinical interviews should follow their jurisdiction’s protocol and consult a qualified professional. If a participant is in immediate danger or at risk of self-harm, contact local emergency or crisis services.