Mark ở vũ trụ khácVIMark học AI
Mark ở vũ trụ khác · Another Universe Common myths · Communication · Tracing the source

Common Myths #4: Words are only 7%

7·38·55

"Communication is only 7% words, 38% tone of voice and 55% body language."

The viral claimwords carry only 7% of the message
Origintwo experiments by Albert Mehrabian, UCLA, 1967
What it really measuredliking, when words contradict voice and face
Mark's verdictthe numbers are real, the reading is wrong
One-minute recap

Essence

Single wordsListeners heard one word such as "dear" or "maybe". No sentences, no conversation.
Feelings onlyThe question was "does this person like me", not "what does the message mean".
Deliberate mismatchA warm word in a cold voice. When words and tone clash, listeners trust the tone.
"55%" is a photo of a faceA still photograph, no posture, no gestures. "Body language" was added later in the retelling.
The author objectsMehrabian wrote that unless you are talking about feelings or attitudes, the equations "are not applicable".

Sources: Mehrabian and Wiener (1967), JPSP 6(1); Mehrabian and Ferris (1967), Journal of Consulting Psychology 31(3); Mehrabian, Silent Messages (1971) and his website (kaaj.com).

Part 1

The question

If words were only 7%, why can't we follow most of a foreign film without subtitles? Because 7-38-55 answers a very narrow question: when words contradict voice and face, which one do listeners believe about your feelings? It never measured how much of a message words carry.

1967Two small experimentsA single word, with voice and face deliberately at odds with it. Measured liking.
→
1971A formula with limits7% words, 38% voice, 55% face, with a warning: feelings and attitudes only.
→
TodayAn unconditional "law""93% of communication is nonverbal", on training slides everywhere.

Sources: Mehrabian and Wiener (1967); Mehrabian and Ferris (1967); Mehrabian, Silent Messages (1971); spread: Lapakko (1997, 2007).

Part 2

Where the myth came from

The claim got trimmed with each retelling. Every step lost one condition.

Four steps of distortion. Step 1, the original result: when talking about feelings, and when words contradict voice and face, perceived liking comes 7% from words, 38% from voice, 55% from the face. Step 2: the mismatch condition is dropped. Step 3: the feelings condition is dropped, giving "words carry only 7% of the message". Step 4: the face becomes body language and the sum becomes "93% is nonverbal". 1 · Original result about feelings + words contradict voice, face: liking = 7% words + 38% voice + 55% face 2 · "Mismatch" condition dropped about feelings: liking = 7% words + 38% voice + 55% face 3 · "Feelings" condition dropped "words carry only 7% of the message" 4 · Face becomes "body language" "93% of communication is nonverbal, words don't matter"
The numbers stayed, the conditions fell away. Step 1 is what Mehrabian wrote. Step 4 is what many training slides say today.Sources: formula and limits in Mehrabian, Silent Messages (1971), quoted on his website (kaaj.com); the distortion as traced by Lapakko (1997, 2007). The four steps are Mark's summary.
1/1967 · The nine-word experimentMehrabian and Wiener played nine single words, read in a liking, neutral or disliking tone. Tone overrode the meaning of the word.
6/1967 · "Maybe" and a photo of a faceMehrabian and Ferris paired a voice saying one neutral word with photos of faces. The face weighed about 3/2 as much as the voice.
1970 · BirdwhistellAnthropologist Ray Birdwhistell suggested words carry "no more than 30 to 35 percent" of social meaning. Later described as a guess.
1971 · The book Silent MessagesMehrabian combined the two results into the 7-38-55 formula for liking, with a note on its scope.
1997 · "Three Cheers for Language"David Lapakko showed the formula in textbooks on public speaking, small groups, organisational and intercultural communication.
2007 · 75 out of 79 websitesLapakko looked again: almost every site gave the formula with no qualification at all.
2025 · A humanlike robotAn experiment with the android Nikola found the same order, face then voice then words, when viewers judged emotion. The true kernel is still there.

Sources: Mehrabian and Wiener, JPSP 6(1), 1967; Mehrabian and Ferris, J. Consulting Psychology 31(3), 1967 (abstract, PubMed 6046577); Birdwhistell, Kinesics and Context, 1970, and Philpott's (1983) comment, quoted on the ROLSI blog 2024 (secondary source); Mehrabian, Silent Messages, 1971; Lapakko, Communication Education 46(1), 1997; Lapakko, CTAMJ 34, 2007; Sato, Shimokawa and Minato, Scientific Reports 15, 2025.

Part 1

Where the myth came from

7-38-55UCLA 19674 steps of distortionquiz
Part 2

What the evidence says

9 single words"maybe"Mehrabian's own wordsKrauss 1981
Part 3

Work lesson

content firstwords, voice, face alignedfeedback
Part 3

What the evidence says

Experiment 1: nine single words, three tones

Mehrabian & Wiener 1967JPSP 6(1), 109–114
A 3 by 3 grid. Rows are word meaning: liking words such as honey, dear, thanks; neutral words such as maybe, really, oh; disliking words such as don't, brute, terrible. Columns are tone: liking, neutral, disliking. Diagonal cells are matched. Off-diagonal cells are mismatched, for example "dear" in a disliking tone. In mismatched cells listeners judged attitude by tone. liking toneneutral tonedisliking tone honey · dear · thanks maybe · really · oh don't · brute · terrible matchmatchmatch "dear", harsh tone→ heard as dislike "terrible", warm tone→ heard as liking
Listeners did not hear a sentence or an idea. They heard one word, read by a stranger on tape, and only guessed that person's attitude. In mismatched cells, tone won.Source: Mehrabian and Wiener, "Decoding of inconsistent communications", JPSP 6(1), 1967, pp. 109–114; the nine words as described by Lapakko (2007) and secondary summaries.

Experiment 2: one "maybe" and one photograph

Mehrabian & Ferris 1967J. Consulting Psychology 31(3), 248–252
Left: photos of a face with liking, neutral and disliking expressions. Middle: a recording of the word maybe in three tones. Listeners were 37 female psychology students. Right: the result, the face weighed about 1.5 times the voice, that is 3 to 2. 3 still face photos + "maybe" × 3 tones 1 neutral word face: 3 parts voice: 2 parts face ≈ 1.5 × voice listeners: 37 female psychology students · no real person present · no gestures
The 55 to 38 ratio comes from here (55/38 ≈ 1.45, close to 3/2). The "words" in this experiment were a single word chosen because it means neither liking nor disliking.Source: abstract of Mehrabian and Ferris (1967), PubMed 6046577 ("the facial component receiving approximately 3/2 the weight received by the vocal component"); sample of 37 per Lapakko (2007), a secondary source who read the original. The 55/38 division is Mark's.

Two experiments glued into one formula

Silent Messages 1971Mehrabian 1971, 1981
Experiment 1 shows tone overrides word meaning. Experiment 2 shows the face weighs about 3/2 the voice. Combined into the formula: total liking equals 7% words plus 38% voice plus 55% face. No experiment measured all three channels at once. Exp 1: voice > words9 words, no photos Exp 2: face ≈ 3/2 voice1 word, with photos liking = 7% words + 38% voice + 55% face the three channels were never measured together "cannot thoughtlessly be combined this way" · Lapakko 1997
The formula extrapolates from two very small samples. Mehrabian called its values approximate and stated that they concern liking, that is, "like versus dislike".Sources: Mehrabian, Silent Messages, Wadsworth, 1971 (2nd ed. 1981), formula quoted on kaaj.com; Lapakko, Communication Education 46(1), 1997, p. 64.

The real scope: one cell out of four

scope of the 7-38-55 ruleMehrabian's own caveat
A 2 by 2 table. Across: words match voice and face, or contradict them. Down: talking about feelings and attitudes, or about information. Only the cell "about feelings and contradicting" is where the formula has a basis. In the other three, such as presenting figures, giving instructions, or telling a story with matching tone, it does not apply. words match voice, facewords contradict voice, face about feelings,attitudes about facts,tasks "thank you", warm toneevery channel agrees 7-38-55 lives here"I'm fine", with a frown quarterly figures reportwords carry nearly all "deadline Friday", jokinglisteners will ask again
"Unless a communicator is talking about their feelings or attitudes, these equations are not applicable." Those are Mehrabian's own words, not his critics'.Source: Albert Mehrabian's website, Silent Messages page (kaaj.com/psych/smorder.html), quoted verbatim by many secondary sites because the original page could not be opened at the time of writing. The examples in the cells are Mark's.

Measured with real speech: words lead

Krauss et al. 1981JPSP 40(2), 312–320
Krauss and colleagues, 1981, two studies with 68 undergraduates. Speakers described pictures in natural speech. Viewers rated emotion from the full recording, or only from a transcript, only from content-filtered voice, or only from silent video. The best predictor of good-bad judgments of the full recording was verbal content, in both studies. full videoimage + sound + words transcript only → best predictor voice only (words filtered out) silent video only 2 studies · 68 students speakers described scenery,buildings, machines, people conclusion: weightsdepend on setting
When words and image are not staged to clash, verbal content is the clearest predictor of how viewers read the speaker. The weights of the channels are not fixed.Source: Krauss, Apple, Morency, Wenzel and Winton, "Verbal, vocal, and visible factors in judgments of another's affect", JPSP 40(2), 1981, pp. 312–320 (original abstract).

The true kernel: feelings travel by face and voice

Sato, Shimokawa & Minato 2025Scientific Reports 15, 25986
A 2025 experiment with the android Nikola, face to face, expressing negative, neutral or positive emotion through facial expression, voice and words. Order of influence on viewers' perceived emotion: face first, then voice, finally words. The study is only about emotion. 1 · face 2 · voice 3 · words when judgingemotion android Nikola, face to face · order only, the paper gives no 7-38-55 split supports the narrow part: emotion. Says nothing about conveying information.
Don't fix one myth with another. "Body language doesn't matter" is wrong too. When the topic is feelings, face and voice really do carry weight.Sources: Sato, Shimokawa and Minato, "Exploration of Mehrabian's communication model with an android", Scientific Reports 15, 25986 (2025), abstract; RIKEN news release, 18/7/2025.
Part 4

Fact check: Solid, Careful, Wrong

11 common claims about 7-38-55, checked against the original research. Tap each line to see why.

SOLIDMehrabian ran the experiments and did publish the 7-38-55 formula

Two 1967 papers (JPSP 6(1); J. Consulting Psychology 31(3)) and the formula in Silent Messages (1971). The numbers are not invented.

SOLIDWhen words contradict voice and face about feelings, listeners tend to trust voice and face

That is exactly what the two experiments measured. The 2025 android experiment (Sato et al.) found the same order, face, voice, words, when viewers judged emotion.

SOLIDWords and facial expression that don't match lower how people rate the speaker

Newcombe and Ashkanasy (2002): 537 people watched videos of 4 leaders giving feedback. Leaders were rated lowest when positive feedback came with a negative face, and highest when their affect was positive and matched the words.

CAREFUL"Body language is very important"

True for feelings and attitudes. But Krauss et al. (1981) found that when people described things naturally, verbal content was the best predictor. Importance depends on the situation; there is no fixed ratio.

CAREFUL"Tone of voice is 38%"

The number only means something inside the experiment: one neutral word, read by a stranger, recorded, paired with a photo. Nobody has measured 38% in a 20-minute presentation.

CAREFUL"About two thirds of meaning is nonverbal" (Birdwhistell)

Birdwhistell (1970) wrote that words carry "no more than 30 to 35 percent" of social meaning. Philpott (1983) noted it was a guess, not a measurement. (Secondary source: quoted on the ROLSI blog, 2024.)

CAREFUL"The 1967 experiments are outdated and worthless"

Small samples, only female students, artificial stimuli: genuinely weak. But the narrow conclusion (feelings, under mismatch) is still supported by later work. The error lies mostly in the broad reading, not in the experiments.

WRONG"Words are only 7% of the message in all communication"

The experiments measured liking, not information. Mehrabian is explicit: unless you are talking about feelings or attitudes, the formula does not apply.

WRONG"93% of communication is nonverbal"

It is just 38 + 55, carrying the same limits. Lapakko (2007) calls it an "urban legend" and found 75 of 79 websites stating it without any qualification.

WRONG"55% is body language: posture, gestures, eye contact"

Experiment 2 used photographs of faces. No posture, no hand movement, no live person. "Body language" was added in the retelling.

WRONG"According to Mehrabian, great presentations are about posture and voice, not content"

The research had nothing to do with presentations. According to Lapakko (1997, quoting a 1995 letter), Mehrabian himself said his findings are "often misquoted". (Quote via a secondary source.)

Sources: original papers in Part 8; Newcombe and Ashkanasy, The Leadership Quarterly 13(5), 2002 (abstract); Lapakko (1997, 2007).

Part 5

Rating the myth

Scores are Mark's assessment based on the sources in Part 8, not figures from any study. Higher is worse.

How wrong
4/5
How persistent
5/5
How harmful
3/5
True kernel
  • Feelings travel strongly through face and voice
  • When words and face clash, listeners trust the face
  • Saying one thing while your face says another costs trust
What's wrong
  • Turns liking into "the message"
  • Turns a face photo into "body language"
  • Makes people practise posture more than ideas

A narrow, real result retold as a broad, false law. How wrong: 4/5 rather than 5/5, because its kernel is still true.

More: why this myth lives so long

It has three things that help a claim spread: a number precise to the percent, the name of a university professor, and a surprising conclusion. It is also useful for people selling courses: if 93% is posture and voice, everyone needs more posture and voice training.

And it matches real experience. Everyone has heard an "it's fine" that clearly wasn't. That true part makes people believe the false part too.

Part 6

Work lesson: content first, then make words, voice and face say the same thing

For giving feedback, presenting, online meetings and messaging colleagues.

Four steps before a feedback session or presentation: write the main point in one sentence, check your real feeling, make words, voice and face say the same thing, and when someone else's words and face disagree, ask. 1 · one-line pointcontent is the corewrite it before rehearsing 2 · real feelingworried? pleased?your face will show it 3 · channels alignedname the feelingdon't smile through criticism 4 · mismatch? ask"fine, but worried?"ask, don't guess Voice and face cannot replace content. They can only make it lose credibility.
Why step 3 matters: in Newcombe and Ashkanasy's experiment, leaders who gave praise with a negative face were rated lowest; leaders whose positive affect matched their words were rated highest.Source: Newcombe and Ashkanasy, The Leadership Quarterly 13(5), 2002, pp. 601–614 (abstract; 537 participants, 4 leaders).
1-on-1 feedbackLead with the main point. If you are worried, say "I'm a bit worried about the deadline" instead of smiling it away.
PresentingSpend most of your time on the message and examples. Rehearse your voice so it doesn't contradict what you say.
Online meetings, chatLess voice and face means words must carry the feeling too. Write "I'm happy with this" or "I'm not convinced yet".
Spot a "7-38-55" slideAsk: measured what, on whom, in what situation? Share Mehrabian's own caveat.
Unless a communicator is talking about their feelings or attitudes, these equations are not applicable.Albert Mehrabian, on the 7-38-55 formula in Silent Messages (his website, kaaj.com).
Part 7

Further reading

Albert Mehrabian, Silent Messages (1971, 2nd ed. 1981)Go to the original book to see the conditions the formula comes with.
David Lapakko, "Three Cheers for Language" (Communication Education, 1997)A short, five-page paper on why the formula cannot be used for communication in general.
Krauss et al., "Verbal, vocal, and visible factors in judgments of another's affect" (1981)A different, more lifelike way of measuring, with a different answer.
Common myths #05: "It takes 21 days to form a habit"Where the 21 came from, and what number a real tracking study found.
Part 8

Sources

Original papers first. At the time of writing most original pages could not be opened from the work machine; details come from the papers' abstracts and verbatim quotes, and anything resting only on a secondary source is labelled as such.

  1. A. Mehrabian and M. Wiener, "Decoding of inconsistent communications", Journal of Personality and Social Psychology 6(1), 109–114, 1967. doi:10.1037/h0024532
  2. A. Mehrabian and S. R. Ferris, "Inference of attitudes from nonverbal communication in two channels", Journal of Consulting Psychology 31(3), 248–252, 1967. PubMed 6046577
  3. A. Mehrabian, Silent Messages, Wadsworth, 1971 (2nd ed., 1981); formula and caveat on the author's website. kaaj.com
  4. D. Lapakko, "Three Cheers for Language: A Closer Examination of a Widely Cited Study of Nonverbal Communication", Communication Education 46(1), 63–67, 1997.
  5. D. Lapakko, "Communication is 93% Nonverbal: An Urban Legend Proliferates", CTAMJ 34, 7–19, 2007. cornerstone.lib.mnsu.edu
  6. R. M. Krauss, W. Apple, N. Morency, C. Wenzel and W. Winton, "Verbal, vocal, and visible factors in judgments of another's affect", JPSP 40(2), 312–320, 1981. doi:10.1037/0022-3514.40.2.312
  7. M. J. Newcombe and N. M. Ashkanasy, "The role of affect and affective congruence in perceptions of leaders: an experimental study", The Leadership Quarterly 13(5), 601–614, 2002. UQ eSpace
  8. Sato, Shimokawa and Minato, "Exploration of Mehrabian's communication model with an android", Scientific Reports 15, 25986, 2025. nature.com
  9. R. Birdwhistell, Kinesics and Context, 1970; Philpott's (1983) comment quoted on the ROLSI blog, 24/5/2024 (secondary source). rolsi.net