The 93% Myth: What Mehrabian Measured
By Amanuel Bekele · The Science of Persuasion
Is the 7-38-55 rule true? Not as it is usually quoted. Mehrabian’s 1967 studies used single words and asked only how the speaker felt, when words, tone and face disagreed. He combined two small results into 7, 38 and 55, and has said the formula does not apply beyond feelings and attitudes. What holds up: when words and delivery conflict, listeners believe the delivery.
If you've sat through communication training, you've heard the number: communication is "7% words, 38% tone of voice, 55% body language." It's on slide decks, in coaching certifications, in articles telling you your words barely matter. The number comes from real research. The claim built on top of it goes well beyond what that research tested.
What Albert Mehrabian studied
In 1967, the psychologist Albert Mehrabian and colleagues published two small studies on a narrow question. When a listener hears a single word spoken with emotional tone, and the word's literal meaning, the tone it's said in and the speaker's face disagree with each other, which one does the listener believe?
The first study (Mehrabian & Wiener) played recordings of single words whose meaning was positive, neutral or negative, spoken in a tone that was positive, neutral or negative.1 Listeners judged how the speaker felt. When meaning and tone conflicted, tone won.
The second study (Mehrabian & Ferris) set a single word, "maybe," said in different tones, against photographs of faces showing different expressions.2 Again listeners judged the speaker's feelings. When the channels conflicted, the face carried more weight than the voice.
Neither study used sentences, arguments or speeches. Neither measured whether anyone was persuaded, informed or convinced of anything. Both asked one question only: how does this person feel about me?
Where 7/38/55 actually comes from
No experiment ever produced the three numbers together. Mehrabian derived them by combining the two results: the first gave a weight for tone against words, the second a weight for face against tone, and putting them on one scale produced 7, 38 and 55.3 It was a rough summary of two narrow lab conditions, offered as such.
What Mehrabian said about it
Mehrabian spent decades pushing back on the popular version. In Silent Messages he set out the formula with its conditions attached: it applies to communication of feelings and attitudes, and only when the channels are inconsistent with each other.3 He has said plainly since that unless a speaker is talking about their feelings or attitudes, the equation does not apply.
A speaker delivering a well-reasoned argument in a steady voice is not in the experimental condition the studies measured at all.
A simple test of the myth
If 93% of meaning travelled through tone and body language, you could follow most of a lecture in a language you don't speak. You can't. You might tell whether the speaker is enthusiastic or bored, which is exactly the narrow thing the studies measured, but you would not know what they argued. That gap is the whole difference between the finding and the myth.
Why the myth survives anyway
It's a satisfying number for people who are anxious about their words and relieved to hear that body language matters more. It's also useful to anyone selling delivery coaching, since "93% of communication is nonverbal" is a stronger pitch than "nonverbal cues dominate when they contradict what you say about how you feel." Both of those explain why the number spreads. Neither is a reason to believe it describes ordinary persuasion.
What persuasion research says instead
When researchers study persuasion directly, the content of an argument matters a great deal, under the right conditions. The elaboration likelihood model found that listeners who care about a topic and have the attention to spare are moved mainly by the quality of the arguments. Cues like a speaker's confidence or attractiveness do most of their work when the audience is distracted or uninvested.4
So for a room that came to hear you, your words are close to the opposite of 7%. For a room that half-listens, delivery carries more of the load. Neither case looks like the slide.
The finding that holds up
The durable part, replicated well beyond Mehrabian, is narrower and less quotable: when verbal and nonverbal signals conflict, listeners tend to trust the nonverbal signal, especially about emotional state.
That's a real and useful principle. It means incongruence is costly. Saying "I'm confident about this" in a voice that wavers reads as not confident, because the audience resolves the conflict in favour of the tone. "We're excited to announce" said flatly sounds like bad news. A smile on a sentence about layoffs reads as callous rather than warm.
It does not mean your words are 7% of your talk. It means your words and your delivery need to agree.
What to do with it
Don't cite 93%. If someone else does, the useful correction is short: it applies to feelings, when the channels disagree.
Check for mismatch, not for "more body language." Record a run-through and listen for the moments where your tone and your words say different things: a key claim delivered with a rising, uncertain ending, or a serious point said with a nervous laugh.
Put the effort where the room is. For an audience that came to hear the argument, build the argument first. Delivery is how it arrives, and it matters, but it carries the content rather than replacing it.
References
- Mehrabian, A., & Wiener, M. (1967). Decoding of inconsistent communications. Journal of Personality and Social Psychology, 6(1), 109–114. https://doi.org/10.1037/h0024532
- Mehrabian, A., & Ferris, S. R. (1967). Inference of attitudes from nonverbal communication in two channels. Journal of Consulting Psychology, 31(3), 248–252. https://doi.org/10.1037/h0024648
- Mehrabian, A. (1971). Silent Messages. Wadsworth. https://openlibrary.org/works/OL15856720W
- Petty, R. E., & Cacioppo, J. T. (1986). The Elaboration Likelihood Model of persuasion. Advances in Experimental Social Psychology, 19, 123–205. https://doi.org/10.1016/S0065-2601(08)60214-2