Body language in video calls: what you can and can't read on Zoom, Teams and Meet
← All articles · By the RedFlag team · September 26, 2026
Most of the conversations that matter now happen in a grid of webcam tiles: sales calls, investor pitches, remote interviews, the weekly one-to-one. So people ask, reasonably, what a video call can tell you about the person on the other side. The honest answer is: less than in-person advice suggests, and different things. A webcam gives you a large, well-lit face and a voice close to a microphone. It also crops away the body, compresses the audio, adds lag, and puts everyone's eyes on the screen instead of the lens. This page covers what survives that trip, what doesn't, and where analysis should not be used at all.
RedFlag is not a lie detector. Nothing on this page can tell you whether someone on a call is telling the truth. RedFlag measures change in a speaker's own voice and face over the course of a recording. Whether a change means anything is your judgment. On a video call, the most common cause of a change is the call itself.
What a webcam removes
- The body. Posture, hands, feet, distance: most of what classic body-language books talk about is outside a head-and-shoulders frame. Anything you think you read from a shoulder shrug on a 300-pixel tile is guesswork.
- Eye contact. To look at your face, the other person has to look away from their camera. On a call, "not making eye contact" is simply what attention looks like. Glances to the side are notes, a second monitor, or a chat window.
- Timing. A few hundred milliseconds of latency turns natural turn-taking into overlaps and awkward pauses. A pause before an answer on a call is very often just the network.
What a webcam keeps, and even improves
- A large face. A typical webcam framing fills a big share of the frame, well above the size where eyelid movement can be measured. RedFlag records blink data only when the face is at least 22% of the frame height; a normal webcam shot clears that easily, which a wide TV shot often doesn't.
- A close microphone. Pitch and pacing come through a laptop or headset mic well enough to measure, because they depend on the fundamental frequency of the voice, not on audio fidelity.
- A stable setting. One person, one chair, one light, for the length of the call. That's what a person-relative baseline needs.
Call artifacts and what they do to each signal
Every video-calling app changes the signal before you see it. These are the effects worth knowing before you read anything into a moment:
| Call artifact | What it does to the measurement | What to do |
|---|---|---|
| Gallery view | Several similar-sized faces on screen at once. RedFlag leaves such stretches unscored (a second face ≥ 55% of the subject's size and ≥ 13% of frame height gates the shot as multi-person) | Use speaker view, or a recording of a single participant |
| Screen sharing | The speaker shrinks to a thumbnail. Below 10% of frame height nothing is scored | Analyze the stretches where the face is on screen |
| Noise suppression | Clips quiet speech and word onsets, which can change how continuously voiced a stretch looks | Compare the person with themselves on the same call, never across calls |
| Frozen or dropped video | A frozen frame shows no blinks, so a bad-connection patch reads as unusually still | Skip segments with visible freezes |
| Packet loss / robotic audio | Garbles the pitch contour for a few seconds | Rewatch any flag that lands on a glitch; it is almost certainly the network |
| Muted stretches | No voiced speech, so there is nothing to measure. RedFlag gates windows with no voiced speech out | Nothing; these are correctly left blank |
The full gate and threshold list is on the methodology page. The practical rule is the one that applies to every format: compare a person with themselves, on this call, in this setup. A change of headset halfway through is a new baseline.
Where it's genuinely useful: your own calls
The best use of signal analysis on video calls is on the person you have the most right to analyze, and the most to gain from: yourself.
- Record a call or a rehearsal of your pitch, demo or presentation, with everyone's consent. Recording laws differ by country and state, and many require every participant to agree.
- Upload the recording to YouTube as unlisted and paste the link into the analyzer. The analysis runs in your browser.
- Look for where your delivery flattened. Pitch monotony, a voice that goes flat outside the normal 2.3–4.5 semitone range, is the signal RedFlag weights most. In a pitch it usually marks the part you've said a hundred times, or the question you were dreading.
- Look at the long unbroken stretches. A flag on continuous talking often lands exactly where you ran through an objection without pausing to let the other person in.
- Rewatch each flag with the transcript. What was asked, and did anything technical change? What's left is useful coaching information about you, not a verdict about anybody.
Public recorded footage is fair material too: remote TV interviews, recorded webinars and panels, conference talks given over video. The same method applies. Pick a single-speaker stretch, baseline against the easy part, and read flags as "their delivery changed here." For a full walkthrough of single-camera footage, see how to analyze an interview.
Where it should not be used
Don't use behavioral analysis to judge job candidates, employees, students, or anyone on a call who hasn't agreed to it. The science doesn't support it. DePaulo et al.'s 2003 review of over a hundred cue studies found no behavior that reliably marks deception, and Bond & DePaulo's 2006 meta-analysis put human accuracy at judging truth at about 54%. You can check your own number with the 54% test. A nervous candidate and a dishonest one look the same on a webcam, and a well-rehearsed one looks calmest of all. In the EU, the law says the same: Article 5(1)(f) of the AI Act prohibits AI systems that infer the emotions of people in the workplace and in education. RedFlag's privacy policy makes you responsible for having the right to analyze what you submit.
What behavior does track, per Vrij, Fisher & Blank (2017), is cognitive load: effortful recall and careful wording, which show up whether a statement is true or not. For the signals individually, see voice pitch under stress, blink rate and cognitive load in speech.
Review your own recorded call. Paste a YouTube URL (unlisted works) into the RedFlag analyzer. It's free with no signup (1 video a day, up to 10 minutes; 2 a day and 30 minutes once you sign in), and the analysis runs in your browser. RedFlag is not a lie detector: it shows where a speaker's signals diverged from their own baseline, and leaves the judgment to you.
Open the analyzerHow the scoring works · How to analyze a podcast guest · Test your own judgment
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.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 5(1)(f).
- RedFlag scoring method, face-size and multi-person gates: How RedFlag scores a video.