Amy Cuddy · "Your body language may shape who you are"
June 2012, Edinburgh. A Harvard Business School social psychologist walks on stage and does something very few speakers dare: instead of opening with her research, she makes the room audit what their own bodies are doing right now. How many of you are making yourselves smaller — hunching, crossing your legs, wrapping your ankles? Some of you are spreading out. "Sometimes we spread out. (Laughter) I see you." A thousand people simultaneously notice, within two seconds, how they are sitting.
Over the next twenty minutes Amy Cuddy tells a story with an unusually clean structure. Body language shapes how others see you — everybody knows that. What we forget is that there is a second audience being influenced by our nonverbals: ourselves. She and Dana Carney brought people into a lab and had them hold a pose for two minutes — expansive, space-taking "high-power" poses, or folded-up, neck-protecting "low-power" ones. Then they collected saliva.
The numbers on stage were almost too good. Testosterone up about 20 percent in one group, down about 10 percent in the other; cortisol — the stress hormone — down about 25 percent versus up about 15 percent; willingness to gamble, 86 percent versus 60 percent, "a whopping significant difference." She states the causal chain without hedging: two minutes of posture "configure your brain" to be either assertive and comfortable or stress-reactive and shut down.
Then she turns on herself. At 19 a car accident threw her from the vehicle; she woke in a head-injury rehab ward, was withdrawn from college, and learned her IQ had dropped by two standard deviations — and she had been someone who identified with being smart. It took her four years longer than her peers to graduate. The night before her first-year talk at Princeton she called her advisor to quit; the advisor ordered her to fake it, talk after talk after talk. So what the room actually remembered was not the hormones but a rewritten slogan: "Don't fake it till you make it. Fake it till you become it." She closes by giving the method away, because the people who can use it most have "no resources and no technology and no status and no power" — they need only their bodies, some privacy, and two minutes.
The talk became the second most-played talk on TED's own counter: 77,575,739 plays as of September 2026. It also became the most famous exhibit in psychology's replication crisis.
Those three numbers came from a 42-person experiment — split two ways, roughly 21 people per cell. In 2015 a Zurich replication with about 200 participants found this: "feeling more powerful" replicated; the hormones and the behavior did not, at all. In the autumn of 2016 the paper's first author, Dana Carney, posted a statement on her own faculty page itemizing how the study had actually been run: subjects added in batches while the p-value was checked along the way ("something like 25 subjects run, then 10, then 7, then 5"); many different power questions asked on the self-report measure, with the ones that "worked" selected; on the gambling measure, two tests giving p = .052 and p = .05, with the smaller reported. Her conclusion left no room: "I do not believe that 'power pose' effects are real."
Cuddy did not retract. She renamed the phenomenon postural feedback and counter-attacked in 2018 with 55 studies, showing that the evidence for posture affecting emotion and feeling is genuinely solid. She won that round — but in that same rebuttal, the pile of variables she declares to have no evidential value contains, by her own enumeration, "hormones … performance in a job interview, gambling." Which is to say: everything the talk was about.
A 2022 meta-analysis of 128 studies and more than 10,000 participants gives the current consensus: opening up your body does make you feel somewhat better, and that is real; on hormones, "hardly any effect"; on behavior, after correcting for publication bias, "the effect almost disappeared."
The quietest revision belongs to TED itself. The talk used to be called "Your Body Language Shapes Who You Are." Open the old link today and it 301-redirects to a new address, with one extra word in the title: may.
For the first fifty seconds of this talk, nobody on stage discusses any research. She is directing the whole room to do one thing: turn its attention onto itself.
"So I want to start by offering you a free no-tech life hack, and all it requires of you is this: that you change your posture for two minutes." Then, instead of giving the hack, she demands an inspection: "So how many of you are sort of making yourselves smaller? Maybe you're hunching, crossing your legs, maybe wrapping your ankles. Sometimes we hold onto our arms like this. Sometimes we spread out."
Sometimes we spread out. (Laughter) I see you.— Amy Cuddy, TEDGlobal 2012, 00:50
The laugh is real, because something physical has just happened in the hall: a thousand-odd people uncross their legs in the same second. A talk about the body, and the most effective prop in it is not a slide — it is making the audience verify the claim on themselves first. The prop is also free.
She adds: "I'm hoping that if you learn to tweak this a little bit, it could significantly change the way your life unfolds." Note that the fuse for the next decade of argument is already lit inside that sentence, and it is the word significantly. It belongs both to statistics and to ordinary speech. The fate of this talk gets caught precisely between those two meanings.
Amy Cuddy was 39, an associate professor at Harvard Business School, studying stereotyping and prejudice. Posture was not her academic home turf; her work with Susan Fiske on the stereotype content model — which explains how we judge social groups along two dimensions, warmth and competence — is standard textbook material in social psychology. She explains on stage how she got from there to power poses: "I study prejudice, and I teach at a competitive business school, so it was inevitable that I would become interested in power dynamics."
The lead came from her classroom. She noticed that MBA students entered the room in radically different ways. Some were "caricatures of alphas," walking to the middle of the room before class started, sprawling once seated, raising their hands like this (she demonstrates something enormous). Others were "virtually collapsing when they come in." And it correlated with gender, and with how much and how well students participated — which matters, because in an MBA classroom participation counts for half the grade. Business schools had been struggling with exactly that gender grade gap: equally qualified women and men come in, and the grades come out different.
So her question was not mystical but operational: could you get people to fake it, and would it lead them to participate more?
And 2012 was a peculiar position for this idea. The original paper, "Power Posing," appeared in Psychological Science in 2010 and had already had one media cycle. The second study — posing affecting job interview performance — had been posted as a Harvard Business School working paper in September 2012 and had not yet been peer reviewed. So what she carried on stage was: one published small-sample paper, plus one freshly written working paper. In front of her sat a TEDGlobal audience, not reviewers.
More importantly, the earthquake had not happened yet. What would later be called the replication crisis was at the foreshock stage — the methodological papers warning that small samples manufacture false positives had been written (Ioannidis 2005; Simmons, Nelson & Simonsohn 2011), but systematic large-scale replication was still ahead. She walked on stage during that window, and the window closed behind her.
Her sequence runs: other people's bodies → your body → your hormones → your performance → your life. Every step is propped up by something concrete.
Step one establishes that body language predicts real outcomes. Her three examples are all other people's research. Nalini Ambady at Tufts found that when people watch 30-second soundless clips of real physician-patient interactions, their judgment of the physician's niceness predicts whether that physician will be sued — "so it doesn't have to do so much with whether or not that physician was incompetent, but do we like that person and how they interacted?" Alex Todorov at Princeton found that one-second judgments of political candidates' faces predict 70 percent of U.S. Senate and gubernatorial race outcomes. Then she rolls a video: outside Number 10 Downing Street, the narrator says, "This lucky policeman gets to shake hands with the President of the United States. Here comes the Prime Minister -- No." The hall detonates into laughter and applause. An argument about the importance of a handshake, closed by a handshake that didn't happen.
Step two is the pivot, and the hinge of the whole talk. We obsess over how others judge us, she says, and forget the other audience:
So we know that our nonverbals govern how other people think and feel about us. There's a lot of evidence. But our question really was, do our nonverbals govern how we think and feel about ourselves?— Amy Cuddy, TEDGlobal 2012, 06:40
Step three lays an evolutionary floor. Nonverbal power is about expanding — making yourself big, stretching out, taking up space — and it holds across the animal kingdom, not just primates. Her prettiest piece of evidence is the display Jessica Tracy studies, known as pride: arms up in a V, chin slightly lifted. People born blind do it too when they win a physical competition, crossing the finish line having never seen anyone do it. Powerlessness is the exact inverse: we close up, wrap ourselves, make ourselves small. And when high and low power meet, we do not mirror each other, we complement: if someone is being powerful at us, we shrink.
Step four is hormones. "Minds," in her usage, have matter in them: thoughts, feelings, and the physiological things that make up thoughts and feelings — "in my case, that's hormones." High-power primate alphas run high testosterone and low cortisol, and so do effective leaders. She presses the second half of that: power is not only dominance, it is how you react to stress. You don't want a leader who is dominant and high on testosterone but highly stress-reactive; you want the one who is assertive and "laid back." And in primate hierarchies, when an individual suddenly takes over an alpha role, within a few days testosterone rises significantly and cortisol drops significantly. So the body can shape the mind, and role changes can shape the mind — which raises the question of doing it at the most minimal level possible. "For two minutes," you say, "I want you to stand like this, and it's going to make you feel more powerful."
Step five is the experiment. People come in, spit into a vial, hold poses for two minutes (she shows five; subjects did two — one of them "has been dubbed the 'Wonder Woman' by the media"), report how powerful they feel, get a chance to gamble, spit again. "That's it. That's the whole experiment." Then the numbers: gambling 86 percent versus 60 percent; testosterone up about 20 percent versus down about 10; cortisol down about 25 percent versus up about 15.
So two minutes lead to these hormonal changes that configure your brain to basically be either assertive, confident and comfortable, or really stress-reactive, and feeling sort of shut down.— Amy Cuddy, TEDGlobal 2012, 12:00
Step six is where she dismantles half her own case. After publication, the media said: so this is what you do when you go in for the job interview, right? "You know, so we were of course horrified, and said, Oh my God, no, that's not what we meant at all." This is not you talking to other people; it is you talking to yourself. — And then she describes the interview study: subjects pose, then go through a five-minute mock job interview with evaluators trained to give no nonverbal feedback at all, which is "worse than being heckled," what Marianne LaFrance calls "standing in social quicksand." Four coders, blind to hypothesis and condition, watch the tapes and say, "We want to hire these people" — all the high-power posers. And what drove it was not what they said: on the competence variables — how well-structured the speech is, how good it is, what their qualifications are — no effect. The effect was in the presence they brought.
The last step turns the talk back onto herself. At 19, a bad car accident; thrown from the car, rolled several times; woke in a head injury rehab ward, withdrawn from college, IQ down two standard deviations. As she puts it: "having your identity taken from you, your core identity, and for me it was being smart, having that taken from you, there's nothing that leaves you feeling more powerless than that." It took her four years longer than her peers to graduate. She convinced Susan Fiske to take her on, got to Princeton, and the night before her first-year talk phoned to say she was quitting. The answer: you are not quitting, because I took a gamble on you — you are going to fake it, do every talk you are ever asked to do, until you reach the moment where you say "Oh my gosh, I'm doing it. Like, I have become this. I am actually doing this."
Then the student. At the end of her first year at Harvard, a student who had not spoken in class all semester — one she had warned to participate or fail — came into her office, totally defeated, and said, "I'm not supposed to be here." Two things happened at once, Cuddy says: she realized she no longer felt that way herself, and — "she is supposed to be here!" Months later the student came back, and Cuddy realized she had not faked it till she made it; she had faked it till she became it.
And so I want to say to you, don't fake it till you make it. Fake it till you become it.— Amy Cuddy, TEDGlobal 2012, 18:30
The close does something rare on a TED stage: it gives the method away, and specifies who to. She asks the room "to share the science, because this is simple. I don't have ego involved in this. (Laughter) Give it away." The people who can use it most are "the ones with no resources and no technology and no status and no power. Give it to them because they can do it in private. They need their bodies, privacy and two minutes."
Pull up the 2010 paper and the clean story deforms.
Scale. "Power Posing" ran on 42 participants (26 female, 16 male), randomly split into two conditions — roughly 21 people per cell. The "about a 20-percent increase" in testosterone was computed on about 21 people. And the paper's text contains no percentages at all; it reports absolute values. Baseline testosterone across all participants averaged 60.30 pg/ml with a standard deviation of 49.58; at Time 2 it averaged 57.40, SD 43.25. The standard deviation is roughly eighty percent of the mean — and pooled across conditions, mean testosterone actually fell (the reported effect is a between-condition difference controlling for baseline, not an overall rise). The percentages on stage are a reading off a figure, not numbers in the text.
Procedure. That line — "For two minutes," you say, "I want you to stand like this" — is precisely not what the experiment did. Participants were told the study concerned the science of physiological recordings, specifically how placement of electrocardiography electrodes above and below the heart might influence data collection. An experimenter attached ECG leads to the back of each calf and the underbelly of the left arm, then physically configured their bodies by hand, with verbal instruction as needed ("Keep your feet above heart level by putting them on the desk in front of you"), and left the room. Each of two poses was held for one minute, during a filler task: viewing and forming impressions of nine faces. The gamble: you are endowed with $2, keep it, or roll a die at even odds for $4. "Feelings of power" was two items — powerful, in charge — on a 1-to-4 scale: 2.57 in the high condition, 1.83 in the low.
Statistics. All three headline effects sit against the threshold: testosterone F(1, 39) = 4.29, p < .05; cortisol F(1, 38) = 7.45, p < .02; gambling 86.36 percent versus 60 percent, chi-square 3.86, p < .05. "A whopping significant difference," on stage, is that chi-square of 3.86.
The second study has a different shape than the stage version too. The interview experiment was HBS Working Paper No. 13-027 in September 2012; it appeared as a peer-reviewed article only in 2015, in the Journal of Applied Psychology. Read the original and the key divergences show up. Sixty-six students took part; five were excluded for not understanding the instructions or not maintaining the poses, leaving N = 61. The study was presented to them as "Physical Motion and Performance," and paid $15. And the posing was not "hide in a bathroom for two minutes" — participants held a pose for one minute, and then, while maintaining that pose, prepared a five-minute speech, holding the pose throughout the entire preparation period, standing freely only for the delivery. So in the lab the posing lasted about six minutes and was embedded in the preparation work itself, not the two-minute elevator-or-bathroom-stall prescription given on stage. The results themselves are not small: performance F(1, 60) = 8.33, p = .005, d = 0.73; hireability F(1, 60) = 7.22, p = .009, d = 0.68; and mediation ran only through nonverbal presence (β = .772), not through what was said (β = .049, p = .692) — which the talk reports accurately. But the same paper carries a number that never reached the stage: for these participants, the "felt more powerful" check was only marginally significant, F(1, 60) = 3.258, p = .076. The paper's explanation is that the question was asked after the stressful speech, which may have depleted conscious feelings of power. (In 2018 the article also ran a correction: the degrees of freedom on three F-tests should have been 1 and 59 rather than 1 and 60, the mediation regression direction was written backwards in the text, and two figures reported the wrong beta weights.)
The reverse question the stage never asks. Neither design had a neutral control condition — only expansive versus contractive. So even if the effect were real, you cannot tell whether expanding helps or contracting hurts. This is not hindsight: it determines whether "two minutes will make you stronger" is a sentence anyone should say. Later meta-analyses return to it repeatedly and lean towards contraction doing more harm than expansion does good — in which case the practical advice this talk should have given is "stop folding up," not "stand like a superhero."
PARTLY HELD UP · as of 2026-09
"Opening up your body makes you feel more powerful" has been replicated repeatedly; that evidence is solid. But it is the shortest link in the chain. None of the three numbers on stage — testosterone, cortisol, gambling — replicated; and the paper's first author has publicly stated she does not believe the effect exists, itemizing the p-hacking that produced it.
May 2015. Eva Ranehill, Anna Dreber, Magnus Johannesson and colleagues published a replication as a commentary in the same journal, Psychological Science. They took the sample to about 200 participants (the risk-task test in the paper is t(198)), extended each pose to three minutes, and delivered instructions by computer — which meant the experimenter was blind to condition and experimenter effects were cut out. They also used no deception: the consent form stated that the study investigated whether physical position influences hormone levels and behavior. The result divides cleanly:
Their wording is restrained: perhaps subtle differences between the protocols omitted some crucial factor, "it is far from clear from the literature what these conditions are."
Autumn 2016 changed the nature of the thing. Dana Carney — first author of the original paper — posted a document on her Berkeley page titled "My position on 'Power Poses'." It is not a paper; it is an itemized self-account. Under "Here are some facts" she writes: "The data are real." "The sample size is tiny." "The data are flimsy." Subjects were run in chunks with the effect checked along the way ("It was something like 25 subjects run, then 10, then 7, then 5" — back then, she writes, this did not seem like p-hacking, it seemed like saving money). Five participants were excluded on bases such as "didn't follow directions." On the gambling measure, the Pearson chi-square p was .052 and the likelihood ratio was .05, and "the smaller of the two was reported." And: "The self-report DV was p-hacked in that many different power questions were asked and those chosen were the ones that 'worked.'"
She then lists three confounds nobody had seen in 2010. The most damaging is the second: participants were told immediately after the gambling task whether they had won, and winning paid an extra $2 — while winning is known to raise testosterone (Booth et al., 1989). High-power posers were likelier to take the bet, and you can only win if you bet. So that testosterone effect — "if it is even to be believed" — "may merely be a winning effect, not an expansive posture effect." The first confound: the experimenters knew the hypothesis. The third: testosterone was not z-scored within gender.
Her conclusion leaves no gap:
I do not have any faith in the embodied effects of "power poses." I do not think the effect is real. … I do not teach power poses in my classes anymore. … The evidence against the existence of power poses is undeniable.— Dana Carney, "My position on 'Power Poses'," 2016
2017 moved the fight to methodology. Joseph Simmons and Uri Simonsohn published a p-curve analysis in Psychological Science concluding that the distribution of p-values in the published significant results is indistinguishable from what you would expect if the true average effect were zero and selective reporting were doing all the work. The same year, Joseph Cesario and colleagues organized a special issue of Comprehensive Results in Social Psychology containing a batch of new preregistered replications; within it, Gronau and colleagues' Bayesian meta-analysis reported that "our meta-analysis yields very strong evidence for an effect of power posing on felt power" — while on the behavioral outcomes, that set of studies did not replicate the effect.
In February of that year Cuddy gave a precise concession in a Q&A on TED's own blog: "The current body of evidence regarding hormones effects is mixed, so I am currently agnostic about how posture affects hormones." She added that her unintentional oversimplification "may have allowed people to miss the forest for the trees."
2018 was her counter-attack, and she won part of it. Cuddy, with S. Jack Schultz and Nathan Fosse, published a reply in Psychological Science widening the scope to 55 studies found through systematic review — and renaming the phenomenon postural feedback rather than power posing. They split the effects into piles and p-curved each separately. The result looks like this:
That 69 percent bar on the left is a genuine victory: posture affects emotion, mood recovery, memory retrieval, self-evaluation, and the evidence is robust. "Feelings of power" holds up on its own too (estimated power 58 percent, CI 24–83 percent). But the pile on the right? Their own verdict on it is "an absence of evidential value." And what is in that "theoretically heterogeneous" remainder, they list in the text: "number of calories consumed at a meal, pain threshold, vengeful intentions, performance on creativity tasks, hormones, beliefs about religion, performance in a job interview, gambling."
Line that sentence up against the 2012 talk: hormones, job interview performance, gambling — the three things the talk was about, all sitting inside the pile ruled to have no evidential value. That is not a critic's summary. It is the list in her own paper.
2022 is where it currently rests. Robert Körner, Lukas Röseler, Astrid Schütz and Brad Bushman published a meta-analysis in Psychological Bulletin covering 128 studies and more than 10,000 participants from 1982 to 2022. Three sentences: on subjective feeling, people who adopted dominant poses did feel somewhat more confident, powerful and positive than people who sat or stood slumped; on behavior, "we did find an effect of body positions on behavior, but the findings were not very robust," and after accounting for publication bias "the effect almost disappeared"; on physiology — hormones, blood pressure, heart rate — "we found hardly any effect of body positions in this domain." Their summary: a simple intervention worth doing, but one whose effects are restricted to subjective experience. The same authors, along with Emma Elkjær and colleagues, keep returning to the missing-neutral-control problem, leaning towards contraction hurting more than expansion helps.
Two quiet revisions close the record, one belonging to TED and one to the idea itself.
First: TED now prefixes the talk's description with a parenthesis — "NOTE: Some of the findings presented in this talk have been referenced in an ongoing debate among social scientists about robustness and reproducibility." Second, and more interesting: the talk used to be titled "Your Body Language Shapes Who You Are." Visit the old address today and it 301-redirects, with the title now reading "Your Body Language May Shape Who You Are." A word was inserted into a title with 77 million views, and the word is "may."
The human side of this dispute — what it is like to be publicly audited by a field after becoming its public face — was written up at length by Susan Dominus in The New York Times Magazine on 18 October 2017, under the title "When the Revolution Came for Amy Cuddy." This page did not read that article; it reports the bibliographic fact only and does not restate its argument.
Worth noting, finally: the thing the talk is actually remembered for was never in dispute. "Fake it till you become it" is not an experimental finding. It is what an advisor said on the phone to a graduate student who wanted to quit. It has held up, and it has nothing to do with p-values.