Quick takeaway: a rehearsed answer collapses when you ask for specificity. Instead of accepting the fluent story, ask for the concrete example, the “how exactly” (numbers, names, decisions), and then change the scenario to something that cannot be rehearsed. Whoever lived the situation deepens without effort; whoever memorized it or copied it from an AI stalls, generalizes, or repeats the script.
Everyone who has interviewed people has walked away from a conversation thinking “that person was impressive” and hired a problem. The candidate spoke with confidence, used the right words, had a ready answer for every question. Three months later, none of it showed up in the work. The problem is not that the person lied outright. The problem is that the interview rewarded performance, not capability.
And the problem has gotten worse. Today it is trivially easy to memorize the 20 most common interview questions, rehearse answers in the right format, and increasingly, receive live answers from an AI during the video call itself. Tools like Cluely, which overlay AI-generated responses directly on the candidate’s screen during a live interview, have made the post-Cluely era a real operational concern for any hiring manager running video interviews. Your read of “this person seems sharp” is worth less than ever. The good news: the technique for breaking a rehearsed answer is simple and does not depend on intuition.
Why so many answers sound good
Rehearsed answers sound good because they were optimized to sound good, not to be true. They have clean structure, textbook vocabulary, and zero friction. That very polish should trigger the alert. Real experience is messy: it has specific detail, a difficult decision, a number the person remembers, a mistake that taught something. A prepared script is smooth and generic. The DePaulo et al. (2003) meta-analysis on cues to deception, covering 158 signals across 120 studies, identifies lack of specific detail, lower plausibility of the account, and absence of ordinary imperfections as among the most consistent indicators in accounts that were not actually lived 1.
Decades of selection research show that structured interviews predict on-the-job performance substantially better than unstructured ones: Schmidt and Hunter’s (1998) 85-year synthesis of selection methods reports predictive validity of .51 for structured interviews versus .38 for unstructured ones 2. The reason is simple: structure forces comparable evidence across candidates instead of rewarding whoever converses best. Recrutador applies the semi-structured interview model, which combines the standardized starting point of structured interviews with adaptive depth per answer — delivering comparability across candidates without losing the ability to follow what each person actually brings. If you are hiring without a dedicated HR department, that discipline is even more critical, because there is no institutional safety net to correct a decision made by impression.
The principle that never fails: concrete evidence beats fluent delivery
The only question that matters for each answer is: did the person deliver concrete, specific, verifiable evidence, or did they perform a generic claim? Speed, confidence, and charisma do not count. A short and specific answer is worth more than a fluent and empty one.
Whoever lived the situation can descend into detail indefinitely: exactly what they did, why they decided that way, what the number was, who else was involved, what went wrong. Whoever memorized the answer has one layer and is done. Whoever is being fed by AI has an excellent response and no ability to defend it once you step off script.
Signs that an answer was rehearsed, generic, or copied
- Speaks in “we” and “the team” constantly, never in “I did.”
- Perfectly structured answer, as if reading from a list.
- Corporate vocabulary (synergy, proactive, results-driven) with no example behind it.
- Specificity disappears when you ask for it: “in general we hit our targets.”
- Long, unnatural pauses before perfect answers (classic screen-reading signal).
- Tone and vocabulary in the answer do not match the rest of the conversation.
- Cannot go deeper: the second and third question about the same case add nothing.
No single signal is proof. What confirms it is what happens when you probe.
The questions that break rehearsed answers
Do not try to “catch” the candidate. Just ask for evidence and go deep. Memon, Meissner and Fraser’s (2010) meta-analysis on the Cognitive Interview (a method originally developed by Fisher and Geiselman for forensic testimony) shows that controlled probing techniques significantly increase the volume of correct details retrieved from a real episodic memory, without a proportional increase in incorrect details 3. Whoever lived the case responds more and better when you probe; whoever memorized it or is being fed live by an AI has nothing to retrieve. Use three layers.
Layer 1: Ask for the example
“Tell me about a specific, recent situation where that happened.”
- Strong answer: brings a concrete case, with context, timeframe, and the candidate’s role clearly defined.
- Red flag: responds in the abstract (“I always try to…,” “we usually…”). Whoever lived it tells a case. Whoever memorized it describes a principle.
“What was your role in that, specifically?”
- Strong answer: separates clearly what they did from what the team did.
- Red flag: dilutes everything into “we.” Cannot isolate their own contribution.
Layer 2: Ask how exactly
“How exactly did you do that? Walk me through it step by step.”
- Strong answer: details the process, the tools, the people, the sequence.
- Red flag: jumps to the result and cannot reconstruct the path.
“What numbers do you remember from that? Before and after.”
- Strong answer: gives a concrete range and knows where it came from.
- Red flag: round and magical numbers with no basis, or “I don’t remember” for something they would have lived through.
“Who else was involved, and how did you decide?”
- Strong answer: names, roles, the moment of disagreement.
- Red flag: nobody has a name, no tension, everything went smoothly. Real life has friction.
Layer 3: Change the scenario
“What if the constraint were the opposite? What would you have done?”
- Strong answer: reasons in real time, shows they understand the why behind their own decisions.
- Red flag: stalls, because the variation was not in the rehearsed script or the AI’s response.
“What went wrong in that case, and what would you do differently?”
- Strong answer: owns a real mistake and draws a specific lesson.
- Red flag: “I wouldn’t change anything,” or a rehearsed false flaw (“I’m too much of a perfectionist”).
The practical rule: always ask at least two probing questions about the same answer. Anyone can prepare the first layer. Only whoever lived it can sustain the third.
How to record and score (do not trust your memory)
When it is time to decide, you will be working with impressions and fragments of what you remember. That is why a good interview captures evidence, not sensation. Define the role’s criteria upfront, note the actual evidence per criterion during the conversation, and classify: was it concrete evidence, specific, or just a generic claim? Compare candidates against the same criteria, not by how much each one sparkled.
For the full step-by-step on building this from scratch, see the structured interview guide.
Where AI-assisted answers break down
An AI writes an excellent response to an isolated question. What it does not do is sustain that response under live probing, with real human interaction timing, and follow-up questions that depend on what the person just said. When you ask “how exactly” for the third time, shift the scenario, and cross-reference something said five minutes earlier, the prepared answer cannot keep up. You do not need to flag the AI. You only need to probe faster than it can follow, and demand evidence that exists only in the memory of someone who actually lived it.
In the post-Cluely environment, this probing discipline is not just good interview practice; it is the baseline competency for running a credible evaluation. Any structured interview method that relies on depth and scenario variation is inherently resistant to real-time AI assistance, because AI cannot fabricate episodic memory on command.
Frequently asked questions
How do you tell if a candidate is lying in an interview?
Do not try to detect lying through body language. Bond and DePaulo’s (2006) meta-analysis, covering more than 24,000 judgments, shows that humans classify statements as true or false correctly around 54% of the time, barely above chance 4. Ask for concrete evidence and probe deeper on the same case: whoever lived the situation descends into detail; whoever invented it stalls or generalizes.
What questions can a candidate not memorize?
The questions that depend on the conversation itself: scenario variations (“what if the constraint were reversed”), the same “how exactly” repeated in layers, and cross-references to something said earlier. There is no ready-made script for that, and AI-generated answers in real time also break down when the question demands specific episodic memory.
Is a generic answer always a red flag?
Not always, but it is an invitation to go deeper. Ask for a specific example. If the specificity appears, great. If it disappears, that is the warning.
How do you probe fairly without turning the interview into an interrogation?
Use the same criteria and the same type of probing with every candidate. Going deeper is not hostile; it gives everyone the same chance to show real evidence.
Run this live, without the overhead
Maintaining criteria, probing at the right moment, and capturing evidence while conducting the conversation is a lot to do alone in the middle of an interview. Recrutador is a Hiring Intelligence Platform with five phases: the Strategist (chat-first consultant) defines the role’s evaluation criteria; the system generates a job description from those criteria; triages resumes with per-criterion coverage; the live HUD runs a semi-structured interview (every candidate starts from the same probe library, depth adapts per answer); and generates the Hiring Memo with cited evidence per criterion at the end.
During the live interview, the HUD suggests the next probing question based on what the candidate actually said, signals when an answer was concrete or vague, and produces the Hiring Memo with the captured evidence at the end. You focus on listening; it handles the rest.
Recrutador applies exactly this logic at the pattern-detection level. The Integrity Signals feature surfaces patterns during the live interview that warrant additional human verification — for example, abrupt vocabulary shifts or unusually long pauses before perfectly composed answers. The system never claims cheating occurred, never labels the candidate, never renders a verdict. The interviewer reads the signal and decides.
Talk to the team and we run your first interview with you.
References
Footnotes
-
DePaulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., & Cooper, H. (2003). Cues to deception. Psychological Bulletin, 129(1), 74-118. Reference meta-analysis covering 158 behavioral and verbal signals associated with deception across 120 studies. DOI ↩
-
Schmidt, F. L., & Hunter, J. E. (1998). The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings. Psychological Bulletin, 124(2), 262-274. DOI ↩
-
Memon, A., Meissner, C. A., & Fraser, J. (2010). The Cognitive Interview: A meta-analytic review and study space analysis of the past 25 years. Psychology, Public Policy, and Law, 16(4), 340-372. A synthesis of 25 years of research on probing techniques that increase retrieval of correct details from a genuinely lived memory. DOI ↩
-
Bond, C. F., & DePaulo, B. M. (2006). Accuracy of deception judgments. Personality and Social Psychology Review, 10(3), 214-234. Meta-analysis synthesizing more than 24,000 judgments, reporting mean human accuracy of 54% in distinguishing truth from deception. DOI ↩