How to analyze an interview: body language, voice, and where the signals diverge

← All articles · By the RedFlag team · September 22, 2026

Anyone can watch an interview and get a feeling about the person on screen. The problem is that the feeling is usually wrong. Bond & DePaulo's 2006 meta-analysis of over 200 studies found that humans judging truthfulness average about 54% accuracy — barely above a coin flip, professionals included. This guide is about doing the watching more carefully: not to detect lies, which no method can do, but to notice when a speaker's own behavior changes, and to hold that observation honestly.

Why "does this look off" is the wrong question

DePaulo et al.'s 2003 review of over a hundred cue studies found no single reliable behavioral marker of deception — not gaze aversion, not fidgeting, not pauses. The behavior that actually shifts under pressure isn't a "tell," it's a response to cognitive load: Vrij, Fisher & Blank (2017) describe how effortful recall, careful wording, and self-monitoring all draw on the same mental resources, and that strain can show up as changes in pace, pitch, blinking, or filler words. Crucially, the same strain shows up for someone struggling to recall a true detail under stress. A change is a question. It is never, by itself, an answer.

The method, step by step

1. Pick a single-speaker clip. Choose a public interview with one person answering on camera — a press conference, a sit-down, a testimony. A few minutes of continuous speech gives you enough to work with; a 20-second soundbite doesn't.
2. Establish their baseline first. Before judging anything, watch the opening minute or two and just notice: how fast do they normally talk, how much does their pitch move, how often do they blink, do they say "um" a lot or rarely? This is the single most-skipped step, and skipping it is why casual lie-spotting fails — there is no universal normal, only this person's normal.
3. Watch for changes at specific moments, not absolute levels. A naturally fast talker who speeds up further on one question is doing something; a naturally fast talker who simply talks fast throughout is not. Compare each moment against the baseline you just built, not against how you imagine an honest person "should" sound.
4. Name the innocent explanation before any other one. Hard recall (an exact date, a name), genuine emotion, fatigue, answering in a second language, or a heavily rehearsed talking point can all flatten pitch or slow speech exactly the way discomfort can. If you can name a boring explanation, you don't get to skip past it for a more dramatic one.
5. Let RedFlag automate the baseline-relative part. Building a mental baseline and holding twenty minutes of it in your head while watching for deviations is genuinely hard. RedFlag's free analyzer does that arithmetic for you: paste a public YouTube URL and it tracks 13 signals — blink rate, gaze and head stability, brow furrow, jaw tension, smile authenticity, voice volume, pitch, verbal stability, vocal activity, speaking rate, and filler ratio and count — then computes a person-relative binary read (🟢 Full Harmony / 🔴 Red Flag) from pitch monotony and blink and voiced-continuity elevation, each measured against that speaker's own median, not a universal number. The screen tints red during flagged stretches of at least 3 seconds, so you know exactly where to look again. The full math, thresholds and gates are published at the methodology page — nothing about how a flag fires is hidden.
6. Read the result honestly. A red flag is not a verdict. It means: at this moment, on these signals, this person's behavior diverged from their own baseline by a measured amount. That is a prompt to go rewatch the moment with the question in mind, not a conclusion to repeat. Whatever you decide after rewatching is your judgment, not RedFlag's.

Signal changes: what they might mean, and what they don't

See it worked through on a real clip

The clearest way to understand this method is to watch it applied. In our walkthrough of Lance Armstrong's 1999 doping denial, we ran a famous on-camera denial through RedFlag and got 6 flagged moments, almost all driven by pitch flattening rather than anything facial. What makes that clip useful is that the documented record is now settled — Armstrong confessed in 2013 — so we could ask a narrow, honest question afterward: on a statement now known to be false, what did the measured signals actually do? The answer, laid out in that piece, is more modest than a TV lie-detector narrative and more useful precisely because of that.

What this is for

This method — and RedFlag itself — will not tell you whether someone is lying. Nothing will; that's not a limitation of this particular tool, it's the state of the science (see the 54% test for why even trained judges do this badly). What this gives you is a disciplined way to notice, instead of a vague impression, exactly where a person's own behavior moved — so that when you do form a view, it's about a specific moment you can point to and rewatch, not a general feeling that happened to land on them.

Sources

• Bond, C. F., & DePaulo, B. M. (2006). Accuracy of deception judgments. Personality and Social Psychology Review, 10(3), 214–234.
• DePaulo, B. M., et al. (2003). Cues to deception. Psychological Bulletin, 129(1), 74–118.
• Vrij, A., Fisher, R. P., & Blank, H. (2017). A cognitive approach to lie detection. Legal and Criminological Psychology, 22(1), 1–21.
• RedFlag scoring method and thresholds: How RedFlag scores a video.