Unlock Your Charisma
A practicum in organising thought while the conversation is still happening: recording what you actually do under pressure, structuring it in real time, and holding the room without performing.
- Levels
- 6
- Lessons
- 30
- Knowledge checks
- 60
- Gates
- 62
- Working tools
- 7
- Simulations
- 2
- Worked exemplars
- 5
Most communication training works on the wrong layer. It works on delivery, because delivery is visible, and it leaves the thing underneath untouched. So people come out of it standing better, pausing more and saying the same unstructured thing.
This course starts from a different premise, and the premise is measurable. The single largest driver of how you sound is not confidence, or voice, or posture. It is how much you are having to organise while you are speaking.
In a corpus of ninety-six speakers producing about 192,000 words, disfluency rates split cleanly by planning load. The speaker who had to generate and structure the content produced 7.00 disfluencies per hundred words. The speaker responding to them produced 4.93. Same people, same conversation, roughly 42 per cent more hesitation on the side that was doing the thinking (Bortfeld, Leon, Bloom, Schober & Brennan, 2001).
That is what "thinking on your feet" costs, and it is why the first capability in this course is structural rather than vocal.
What this course means by charisma
Not charm, not extroversion, not performance. The working definition comes from the only research programme that has trained it and measured the result: values-based, symbolic and emotion-laden leader signalling (Antonakis, Bastardoz, Jacquart & Shamir, 2016), operationalised as twelve nameable behaviours that a trained observer can mark present or absent.
Those behaviours are trainable. Across two studies, a field experiment with 34 middle managers and a within-subjects study with 41 MBA participants, coded use of the tactics roughly doubled and the sample-weighted effect was D = .62 (Antonakis, Fenley & Liechti, 2011). In a randomised field experiment where 106 temporary workers stuffed envelopes for a children's hospital, a charismatic speech raised output 17.4 per cent against a fixed-wage baseline, statistically indistinguishable from paying a piece rate (Antonakis, d'Adda, Weber & Zehnder, 2022).
And the finding that should set your expectations
In a preregistered prospective meta-analysis across five countries, the same tactics produced d = 0.52 in person and d = 0.01 across six virtual samples (Ernst et al., 2022).
Zero. On screens, in that study, the tactics did nothing at all.
The behaviours are real, trainable and context-bound. Anybody selling you a universal method is selling you something the evidence does not support.
Nobody knows which of the twelve matter, or in what combination. The authors say so themselves: "It is not clear whether we have identified the best markers of charisma."
The doses that produced measurable change were 16 hours plus twelve weeks of practice, and 80 to 90 hours. No study tests a one-day workshop against a control with an objective outcome. This course runs fourteen weeks because that is what the evidence base describes.
Three capabilities, and the reason they are in this order
Clarity comes first because it is upstream of everything. If you have not decided what you are saying, no amount of delivery work reaches it.
Judgement comes second and is the part most courses omit: knowing when to answer, when to qualify, when to give a range instead of a number, and when to say you do not know. That last one is better evidenced than almost anything in delivery, and it points the opposite way to what people expect.
Connection comes last because it is the hardest and because it depends on the first two. Listening properly is not a technique you add. It is what becomes possible once you are no longer using the other person's speaking time to compose your answer.
What this course spends most of its time removing
More than half the standard advice in this subject is either unsourced or contradicted by the best available test. Rather than quietly omitting it, this course names each item and shows the study.
The headline case is power posing. The original finding, on 42 participants, reported raised testosterone, lowered cortisol and increased risk-taking. A replication with 200 participants found the hormonal and behavioural effects null. A p-curve analysis estimated the underlying literature's statistical power at 5 per cent, the signature of a true effect of zero. And the first author published a statement beginning: "I do not believe that 'power pose' effects are real."
What survives is smaller and more useful: across 73 studies, the measurable effect is contractive posture against neutral, g = 0.45, while expansive against neutral is g = 0.06 (Elkjær, Mikkelsen, Michalak, Mennin & O'Toole, 2022). The instruction is not to stand like a superhero. It is not to hunch.
By the end of Level 2 you will have stopped counting your filler words, stopped pausing before you answer, and stopped trying to calm down. Each of those is contradicted by a study larger than the one that produced the advice.
The hours you free up go into structure, evidence and listening, which are the parts that hold up.
Cadence
Fourteen weeks. Plan three to four hours a week: about an hour on lessons and checks, two hours on recorded practice against real conversations, and thirty minutes on the record.
Levels 0 to 4 take two weeks each. Level 5 takes four, because it contains conversations with real people that you will want to postpone, and because the Live Room takes a week to arrange.
The source programme runs eight facilitated sessions. This runs fourteen because a session can be attended and a recording of yourself cannot.
Where the evidence comes from, and where it does not
Almost every psychological finding here was established on a Western sample. In an audit of the field, 96 per cent of research subjects came from Western industrialised countries holding 12 per cent of the world's population, and 68 per cent came from the United States alone (Henrich, Heine & Norenzayan, 2010).
For this subject there is a specific and consequential version of that gap. Accent bias in hiring is meta-analytically established at d = 0.47 across 139 effect sizes and 4,576 participants, it is stronger for foreign than for regional accents, stronger in high-communication jobs, and stronger for women. Critically, candidate comprehensibility was not a significant moderator, which removes the usual defence that the penalty is practical rather than prejudicial (Spence, Hornsey, Stephenson & Imuta, 2024).
The penalty is real and it is not about being understood. That is worth knowing plainly rather than discovering.
Two findings point somewhere useful. When 61 lawyers and graduate recruiters assessed interview answers that varied in quality, they tracked answer quality and not accent on competence judgements, while accent still moved likeability. And in an intervention trial with 480 participants, simply raising awareness of accent bias outperformed fairness appeals, accountability measures and diversity appeals (Accent Bias in Britain, 2020).
This course therefore works on the evidence and the structure rather than on the accent, and it says openly that fixing the evaluation is better evidenced than fixing the speaker.
What this course does not claim
It does not claim that better speaking fixes a bad argument. In two preregistered studies with 668 participants, charismatic delivery did not change how carefully people read a message, and the single significant effect ran the wrong way: contra-environmental messages were rated more convincing after charismatic delivery (Engelbert, van Elk, Theeuwes & van Vugt, 2024). Delivery can make a weak case land, which is a reason to be careful rather than a reason to celebrate.
It does not claim more is better. Across two samples of 306 and 287 leaders, the relationship between charismatic personality and rated effectiveness was an inverted U: low-charisma leaders were rated less effective because they lacked strategic behaviour, and high-charisma leaders because they lacked operational behaviour (Vergauwe, Wille, Hofmans, Kaiser & De Fruyt, 2018).
And it does not claim that everyone should end up sounding the same. Accent, personality, introversion and cultural style are not what this course assesses, and the gates are written so that a quiet, slow, heavily accented speaker can pass every one of them.
A level is passed when you have done something in a real conversation and have a recording or a contemporaneous record of it.
Nothing on this page has changed how you will sound in your next meeting, and it is not supposed to.
The levels
- 0Measure what you actually doReplace your impression of how you speak under pressure with a recording, a transcript and three counts.Free
- 1Organise before you speakBuild a structure in the seconds you have, and say the point out loud instead of leaving it to be inferred.Locked
- 2Delivery, and what to stop working onRetire the delivery habits you have been told to fix, and put the hours where the effect sizes actually are.Locked
- 3Answer what you did not prepare forHandle the question you did not see coming, using material you loaded before the room started rather than reasoning you produce inside it.Locked
- 4Make them see itExplain the thing you know best to somebody who does not know it, and find out from them rather than from yourself whether it worked.Locked
Your speaking ground
One recurring professional conversation, one recording setup, one person who will tell you the truth, and one thing you find hardest to explain.
Before you finish Level 0 you fix four things. They stay fixed for fourteen weeks.
One: the recurring conversation
A real professional conversation that happens at least fortnightly, where your contribution matters and where you are sometimes asked something you did not prepare for. A team meeting you attend. A client call. A supervision session. A project review.
It has to satisfy three tests. It recurs, so you can measure change rather than remember it. You speak in it, rather than attending it. And something in it is genuinely unpredictable, because a meeting where you always know the questions trains nothing this course teaches.
If nothing you attend qualifies, the honest answer is to create one: a fortnightly fifteen-minute standing conversation with a colleague about live work, with an agenda neither of you fully controls.
Two: the recording
You cannot do this course from memory. Self-assessment of speaking is among the weakest instruments in this field, and there is a specific demonstration of it in Level 5: in a study of 238 government executives in live disagreements, participants' self-rated receptiveness did not predict their partner's rating at all (β = 0.05, p = .54), while an algorithm reading their actual words did (β = 0.29, p < .001).
So: an audio recorder you will actually use. A phone is sufficient. You need consent from anybody recorded, which in most organisations means recording only your own contributions or getting explicit agreement, and where recording is not permitted you keep a contemporaneous written record instead, made within ten minutes.
Set it up in week one. The Level 0 exercise depends on it and cannot be reconstructed later.
Three: the reader
One person who will listen to a recording and tell you what was unclear, rather than telling you that you did well.
They do not need to work in your field. Two of the exercises specifically require somebody who does not, because a listener inside your profession reconstructs your meaning from their own knowledge and cannot tell you what was missing.
Name them before Level 0 closes, in writing, with the date you asked them.
Four: the hard explanation
One thing you understand well and consistently struggle to explain to people outside your area. A technical process, a regulatory constraint, a methodology, a financial mechanism.
You will carry it through Level 4 and explain it three times to three different audiences. Choose it now, before you know what the level asks.
"The Thursday programme review, fortnightly, eight people, I present the data section and get questioned on it. Recording on my phone from my own seat, consent agreed with the chair on 6 March. My reader is Chidi, who runs a restaurant and knows nothing about M&E. The hard explanation is why a confidence interval is not the same as a margin of error, which I have failed to explain to this team three times."
"I want to become more confident and articulate in meetings and presentations generally, and be better at thinking on my feet." No named conversation, no recording, no person, nothing that could be measured twice.
If two come to mind and one of them contains somebody who challenges you, that is the one with the information in it.
The other one is a meeting where you already perform well, and it will show you nothing for fourteen weeks.
Rules of practice
Seven rules. Four are supported by evidence, three are design choices, and the difference is marked.
1. Everything is recorded or written down within ten minutes
Design choice, with support by analogy. The support comes from goal monitoring rather than from speaking: prompting people to monitor progress raised attainment by d = 0.40 across 138 trials and 19,951 people, and physically recording progress outperformed tracking it without a record, 0.43 against 0.29 (Harkin et al., 2016). What transfers is narrow and sufficient. Your memory of how a conversation went is produced by the person with the strongest interest in the answer, and it is worse than usual here because anxiety distorts it in a known direction.
2. You measure before you change anything
Evidence: strong, by demonstration. People are poor judges of their own communication. Government executives' self-rated receptiveness predicted their partner's assessment at β = 0.05, non-significant, while a language algorithm predicted it at β = 0.29 (Yeomans, Minson, Collins, Chen & Gino, 2020). Writers systematically overestimated their own receptiveness and completely failed to anticipate that receptive language would make them more persuasive. Level 0 is two weeks of measurement for that reason.
3. Structure is upstream of delivery
Evidence: strong for the mechanism. Speech production runs conceptualisation, then formulation, then articulation (Levelt, 1989). Load at the first stage shows up as disfluency at the third: the speaker carrying the planning burden produced 7.00 disfluencies per hundred words against 4.93 for the responder (Bortfeld et al., 2001). Working on articulation while the conceptual load is unmanaged is working on the wrong stage.
4. Uncertainty is stated precisely, not hedged verbally
Evidence: strong. Across five experiments with 5,780 participants, including a preregistered replication and a field experiment on the BBC News website, communicating uncertainty as a numerical range produced only a small decrease in trust in the number and no significant change in trust in the source, while verbal hedging produced a larger decrease in both (van der Bles, van der Linden, Freeman & Spiegelhalter, 2020). A range is nearly free. Waffle is not.
5. You say you do not know, and you say why
Evidence: moderate, and it reverses the expectation. Across three experiments and three surveys totalling about 3,100 participants, "I don't know" responses increased perceived trustworthiness relative to directive advice, and an explained "I don't know" beat an unexplained one on competence. Participants in the surveys predicted the opposite (Mushkat & Mayo, 2026, preprint). The competence cost sits in the unexplained version, not in the admission.
6. Nothing here requires anyone else to change first
Design choice. Every gate can be claimed inside a team that stays exactly as it is, with a chair who does not improve and colleagues who keep interrupting. That is a constraint on the course, not a claim about your organisation.
7. You claim your own gates
Design choice. Nobody marks these for you, and marking one early costs only you.
One lesson block read, at least one knowledge check answered wrong, one recording made and listened to, three dated lines in the record, and one moment named where your structure disappeared and why.
All lessons read, all checks correct first time, no recording made because the meeting was not a good example, the record blank, and a private sense that you already do most of this.
Assessment
What counts, what does not, and what you should hold at the end.
What is assessed
| Component | Weight | The evidence |
|---|---|---|
| The baseline and its re-measurement | 15% | Two transcribed recordings, twelve weeks apart, with the counts done the same way |
| Structure under time pressure | 20% | Recorded responses to unseen questions, with the structure identified afterwards |
| Judgement under questioning | 15% | A question log with your response type against each, including the ones you got wrong |
| The three explanations | 15% | One thing explained to a specialist, a senior generalist and a non-specialist, recorded |
| Listening and receptiveness | 15% | A summarise-before-responding record, and one real disagreement written to the recipe |
| The Live Room | 20% | A multi-stage unscripted session, with the after action review across eight areas |
What is not assessed
Accent is not assessed. Volume, pitch, extroversion and speaking speed are not assessed, except where your own recording shows a change you set out to make. Whether people found you impressive is not assessed. Whether you sounded confident is not assessed, because the evidence that confidence signals competence is weaker than the evidence that it does not.
A measurement warning that applies to everything you are about to do
Your assessment of how a conversation went is a poor instrument, and there are three independent demonstrations of it in this course.
Anxious speakers believe they performed far worse than the assessor thought: interview anxiety correlates between −.15 and −.49 with self-rated performance and between −.07 and −.28 with observer-rated performance (McCarthy & Goffin, 2004), and in a field study of 8,343 real candidates the relationship with observed performance was γ = −.03, which is nothing.
Speakers systematically overrate their own receptiveness, and their self-rating carries no information about how their partner experienced them (Yeomans et al., 2020).
And people wrongly predict that admitting ignorance costs credibility, when it raises it (Mushkat & Mayo, 2026).
That is why the assessment above is weighted toward recordings and contemporaneous records, and why the scorecard below is explicitly not evidence.
The check
Ten statements, each scored one to five, run before Level 0 and again after Level 5. Run it now and record the date.
It is a reflection instrument with no validation behind it, and it is the same weak instrument this page has just spent three paragraphs warning you about. Its only legitimate use is beside the recordings. If line 5 rises and your recorded speech rate is unchanged, the score moved and you did not.
What you should hold at the end
Two transcribed recordings twelve weeks apart with disfluency and structure counted the same way. At least twelve recorded responses to unseen questions. A question log with the response type you chose against each. One thing explained three times to three audiences, recorded, with what changed and what did not. A summarise-first record from at least six conversations. One real disagreement written to the receptiveness recipe, with what happened. And a Live Room recording with eight-area feedback and your own assessment written before you read theirs.
Sources
Every figure in this course, with the study behind it and what it can and cannot support.
Full citations with links are in SOURCES.md in the course directory. This page is the map: what each level rests on, and how strong it is.
How to read the grades
Strong means large samples, randomised or preregistered designs, and either replication or a meta-analytic estimate. Moderate means well conducted but single-source or single-lab. Weak means small samples, one laboratory, or a claim that circulates without a controlled test. Debunked means it failed a well-powered replication or traces to no primary source.
Where a figure is weak, the lesson says so rather than this page.
A structural warning about this subject
Communication training has the worst evidence-to-confidence ratio of any professional subject this course has examined. Its two most repeated statistics come from a 1926 dissertation on 21 farmers and housewives observed for one day, and a 1957 magazine article that reports no sample size, no instrument and no statistics. Its most famous demonstration was disavowed by its own first author. Its most cited number about nonverbal communication comes from an experiment on single words and still photographs whose author has stated it does not apply.
This course names each one, gives the provenance, and replaces it where a replacement exists.
Level 0, on measurement
| Finding | Figure | Grade |
|---|---|---|
| Baseline disfluency in ordinary conversation | 5.97 per 100 words overall; 96 speakers, ~192,000 words (Bortfeld, Leon, Bloom, Schober & Brennan, 2001) | Strong |
| Planning load, not personality, drives hesitation | Director 7.00 per 100 words against matcher 4.93, a within-corpus manipulation of who generates content | Strong |
| Speech production runs in stages | Conceptualisation, formulation, articulation; 60 to 90 per cent of speech errors involve segments (Levelt, 1989) | Strong for the architecture, descriptive rather than predictive |
| Self-rating of your own communication carries no information about how you were received | Self-rated receptiveness to partner rating β = 0.05, p = .54; algorithm β = 0.29, p < .001; N = 238 government executives (Yeomans et al., 2020) | Strong |
| Anxious speakers underrate their own performance | Anxiety to self-rated −.15 to −.49 against observer-rated −.07 to −.28 (McCarthy & Goffin, 2004); γ = −.03 in 8,343 real candidates (McCarthy et al., 2021) | Strong |
| "Listening is 45 per cent of communication time" | Rankin, 1926: 21 adults, one day, 15-minute intervals, 42 per cent. Across nine studies the range is 15 to 55 per cent; the largest modern study, N = 680, found 24 per cent | Weak, bordering debunked |
| "We remember only 25 per cent of what we hear" | Nichols & Stevens, 1957, Harvard Business Review: no sample size, no instrument, no statistics, not peer reviewed. The 25 per cent is a two-month retention figure; their immediate figure is about 50 per cent | Debunked |
| "93 per cent of communication is nonverbal" | Two 1967 experiments on single spoken words and still photographs, 62 female participants, judging attitude only; the author states the equations do not apply otherwise | Debunked |
Level 1, on structure
| Finding | Figure | Grade |
|---|---|---|
| Charisma tactics are trainable and coded behaviourally | Coded use .24 to .48; sample-weighted D = .62 across two studies; 34 managers and 41 MBAs; coder agreement 85.03 per cent, κ = .67 (Antonakis, Fenley & Liechti, 2011) | Moderate |
| The tactics move objective behaviour | Output +17.4 per cent against fixed wage, p = .017, statistically indistinguishable from a piece rate, p = .671; N = 106 field experiment (Antonakis, d'Adda, Weber & Zehnder, 2022) | Strong |
| The tactics do not transfer to virtual delivery | In person d = 0.52 (k = 4); virtual d = 0.01 (k = 6); preregistered prospective meta-analysis across five countries (Ernst et al., 2022) | Strong |
| Nobody knows which tactics matter | The authors state: "It is not clear whether we have identified the best markers of charisma" | Evidence gap |
| Put the conclusion first with an engaged audience | High elaboration produces primacy effects, low elaboration produces recency (Haugtvedt & Wegener, 1994) | Moderate to strong, about competing messages |
| BLUF | US Army Regulation 25-50, paragraph 1-10. A promulgated writing convention with no experimental validation in the doctrinal chain | Doctrine, not evidence |
| The Minto pyramid principle | Provenance fully documented: developed in McKinsey's London office 1966 to 1973. No study tests pyramid-structured against unstructured communication on any outcome | No empirical validation |
| The inverted pyramid | One unpublished dissertation, N = 58, reading rather than listening: no recall difference between structures | Weak |
| "What? So what? Now what?" | Not Matt Abrahams's. Rolfe, Freshwater & Jasper (2001), a reflective-practice model from nursing education | Attribution correction |
Level 2, on delivery and what to stop doing
| Finding | Figure | Grade |
|---|---|---|
| Power posing raises testosterone, lowers cortisol, increases risk-taking | Original N = 42. Replication N = 200: all null, point estimates nominally reversed. P-curve estimated literature power at 5 per cent. First author: "I do not believe that 'power pose' effects are real" | Debunked |
| What survives is felt power | Replicated, d = 0.344; Bayesian meta-analysis pooled N = 1,071 | Moderate, self-report only |
| The postural effect is avoiding contraction, not adopting expansion | Contractive against neutral g = 0.45; expansive against neutral g = 0.06; 73 studies (Elkjær et al., 2022) | Moderate |
| Speech rate and processing are curvilinear | Linear effect p = .203, not significant; curvilinear B = −2.56, p = .031; mediated by perceived ability to process; N = 3,958 across six studies (Guyer et al., 2024) | Strong |
| Perceived confidence is also curvilinear in speed | B = −7.28, p < .001. Both "slow down to sound confident" and "speed up to sound confident" are wrong as general rules | Strong |
| Speed helps a hostile audience and hurts a friendly one | Rapid speech suppressed counterarguing of a counterattitudinal message and inhibited favourable elaboration of a proattitudinal one (Smith & Shaffer, 1991) | Moderate, single lab |
| Fast speech reduces the audience's ability to tell strong arguments from weak | Confirmed at moderate and high personal relevance (Smith & Shaffer, 1995) | Moderate |
| Fillers do not impair comprehension | "Uh" made listeners recognise the following word 47 ms faster, p = .001, N = 34, replicated in Dutch at 28 ms; "um" was neutral (Fox Tree, 2001) | Moderate |
| Fillers do not cost eloquence against the silence that replaces them | Filled and silent pauses did not differ on eloquence, F = 0.26; speakers using filled pauses were rated significantly more relaxed, F = 8.28, p < .005; N = 1,067 (Christenfeld, 1995) | Moderate |
| Counting and eliminating filler words improves outcomes | No study located. No threshold established | No evidence |
| Pausing before you answer signals thoughtfulness | No study located showing a competence gain | No evidence |
| Pausing before you answer costs perceived sincerity | 2 s delay d = 0.32; 3 s d = 0.38; 5 s d = 0.63, rising to d = 1.08 on video; 14 experiments, N = 7,565 (Ziano & Wang, 2021) | Strong |
| Accent bias in hiring | d = 0.47 across 139 effect sizes, N = 4,576; stronger for foreign accents, high-communication jobs and women; comprehensibility was not a significant moderator (Spence et al., 2024) | Strong |
| Professionals attending to answer quality show no accent bias on competence | 61 lawyers and graduate recruiters rated by answer quality regardless of accent; bias remained on likeability (Accent Bias in Britain, 2020) | Moderate, N = 61 |
| Raising awareness beats other de-biasing strategies | Six conditions, N = 480; awareness-raising outperformed fairness appeals, accountability and diversity appeals | Moderate |
| "Executive presence" | No peer-reviewed definition, factor structure, reliability or predictive validity located. Primary source is a trade book on proprietary survey data | Not a research construct |
| Reappraising anxiety as excitement | Speech study N = 140: longer speeches, rated more persuasive, competent and relaxed (Brooks, 2014). A direct replication failed on all observer ratings while the self-report effect held | Moderate for the felt effect, failed replication for the observed effect |
Level 3, on questions you did not prepare for
| Finding | Figure | Grade |
|---|---|---|
| Saying "I don't know" raises trustworthiness | Three experiments and three surveys, about 3,100 participants; explained beats unexplained on competence; participants predicted the opposite (Mushkat & Mayo, 2026) | Moderate, preprint |
| Numerical uncertainty is nearly free, verbal hedging is not | Five experiments, N = 5,780, including a preregistered replication and a BBC News field experiment; ranges produced a small decrease in trust in the number and no significant change in source trust; verbal hedging cost more on both (van der Bles et al., 2020) | Strong |
| Experts gain from expressing minor doubt, non-experts from certainty | Three experiments; works only with strong arguments, and can backfire with weak ones (Karmarkar & Tormala, 2010) | Moderate, consumer-review stimuli |
| Evasion is a set of nameable, gradable moves | Taxonomy from 100+ broadcast interviews and press conferences over two decades, with verbatim transcripts (Clayman, 2001) | Strong as a descriptive taxonomy |
| Question hostility can be coded reliably | Four dimensions across 742 questioning turns from 30 press conferences; intercoder κ ≥ .80 on 7 of 10 indicators (Clayman & Heritage, 2002) | Strong |
| Delay before answering costs sincerity | See Level 2. This is the finding that governs how long you may take to think | Strong |
Level 4, on explaining
| Finding | Figure | Grade |
|---|---|---|
| The curse of knowledge | Camerer, Loewenstein & Weber (1989), the founding experimental demonstration in market settings | Strong |
| The tapping study | Newton (1990), unpublished dissertation. Tappers predicted about 50 per cent and listeners identified about 2.5 per cent. The widely quoted "predicted 80 per cent" is wrong | Weak as evidence, excellent as a live demonstration |
| Curse of knowledge in adults | Birch & Bloom's effect did not survive: a replication with N = 3,074 found d = 0.20 to 0.24 rather than .47 to .65 | Use the replication figure |
| Jargon damages comprehension and confidence | F(1,636) = 76.03, η² = .11, N = 650 (Bullock, Colón Amill, Shulman & Dixon, 2019) | Strong |
| Defining your jargon does not repair the damage | Definitions η² = .0005, p = .543. Removing the term works; glossing it does not | Strong |
| Comparing two analogous cases roughly triples strategy transfer | Working professionals; the intervention is comparing two cases and articulating the commonality, not deploying one decorative analogy (Thompson, Gentner & Loewenstein, 2000) | Moderate |
| Self-explanation improves understanding | g = .55, 69 effect sizes from 64 reports (Bisra et al., 2018) | Strong |
| Preparing to teach helps only if the expectancy is set in advance | g = 0.48 with expectancy against g = −0.02 without; 39 studies (Kobayashi, 2024) | Strong |
| The Feynman technique | Not Feynman's. Traceable to Scott H. Young, around 2011. The underlying mechanisms are evidenced; the name is not | Attribution correction |
| Narrative persuasion | The often-quoted r = .44 is the transportation-to-attitude correlation, not narrative's causal effect. The causal estimate is r ≈ .17 to .23 (Braddock & Dillard) | Correction |
| Plain-language rewriting | Comprehension +19.8 percentage points, 95% CI 14.7 to 24.9, p < 0.001, N = 488 randomised trial, with no accuracy loss; effects varied sharply by document | Strong |
Level 5, on the room
| Finding | Figure | Grade |
|---|---|---|
| Being listened to reduces speaker anxiety and raises self-awareness and attitude clarity | Clarity d = 0.46, 95% CI 0.31 to 0.62, pooled across five studies, N = 717, one preregistered; anxiety d up to −0.86; self-awareness d up to 1.19 (Itzchakov, DeMarree, Kluger & Turjeman-Levi, 2018) | Strong |
| It does not make speakers think they are more right | Attitude correctness η²p ≈ .00 in all five studies; no increase in persuasion intentions | Strong, and it is the finding people misreport |
| Listening training rarely changes what speakers notice | The field's own review states existing evaluations "did not show that speakers notice any change", and that making trainees knowledgeable about the benefits "is unlikely to change behavior by itself" (Kluger & Itzchakov, 2022) | Strong caution |
| The receptiveness recipe works | Four components; human-rated receptiveness β = 0.57, willingness to collaborate β = 0.28, rated persuasiveness β = 0.24, all p < .001; N = 771 writers, 1,548 raters (Yeomans et al., 2020) | Strong |
| People do not anticipate the persuasiveness gain, and the recipe feels harder | Writers' own prediction of the persuasiveness effect β = 0.05, p = .506 against an actual 0.24; the recipe felt harder and they were less inclined to reuse it, while taking no extra time | Strong, and it is the main obstacle to teaching it |
| Low receptiveness predicts being attacked better than it predicts attacking | 585 matched Wikipedia thread pairs: attackers 53.6 per cent, p = .082; victims 60.3 per cent, p < .001 | Strong |
| Men interrupt women far more | Overstated. Meta-analytic d = .15 overall, described as negligible; d = .33 for intrusive interruptions specifically | Correction |
| Equality of conversational turn-taking predicts group performance | Correlation with variance in speaking turns r = −0.41, p = .01; 40 groups (Woolley et al., 2010). Contested: Bates & Gupta (2017) argue member IQ explains the collective intelligence factor | Moderate, and contested |
| Structured disagreement beats consensus-seeking | Achievement d = 0.70, k = 12; perspective taking d = 0.97, k = 8 (Johnson & Johnson, 2009) | Moderate |
| What actually predicts meeting effectiveness | Goal clarity r = .51 to .54; agendas r = .14 to .18; meeting duration and total time show no significant relationship with effectiveness | Moderate |
| Information-rich feedback beats evaluative feedback by about four to one | d = 0.99 for task, process and self-regulation feedback against d = 0.24 for reinforcement and punishment; 435 studies, k = 994, N > 61,000 (Wisniewski, Zierer & Hattie, 2020) | Strong, educational settings |
| Closed-loop communication | The check-back is a three-turn structure published as an operational protocol: sender states, receiver states back, sender confirms. The loop closes on the third turn (AHRQ TeamSTEPPS) | Documented protocol |
Claims removed from this course, and why
"Power posing changes your hormones." Failed a 200-participant replication on every physiological and behavioural measure. The first author states she does not believe the effects are real.
"93 per cent of communication is nonverbal." Two 1967 experiments on single words and photographs of faces, 62 female participants, judging liking. In the thin-slice meta-analysis, transcripts alone reached r = .29 against tone of voice at .26.
"Listening is 45 per cent of what we do." Twenty-one adults, one day, 1926. The range across nine studies is 15 to 55 per cent and the largest modern study found 24.
"We remember 25 per cent of what we hear." A 1957 magazine article with no method, and the figure is a two-month retention number routinely misquoted as immediate recall.
"Eliminate your filler words." Against the silence that would replace them, filled pauses cost nothing on eloquence and are rated more relaxed. No study shows that reducing them improves any outcome.
"Pause before answering; it signals thoughtfulness." No study shows a competence gain. Fourteen experiments with 7,565 participants show a sincerity cost rising to d = 1.08.
"Slow down to sound confident." Perceived confidence is curvilinear in speed. Slow reads as less confident, not more.
"Eleven million meetings a day" and "$37 billion wasted." A 1976 trade book with no source and a 1989 magazine line with no methodology. The widely quoted modern figures are a 2015 blog model whose own ranges are 36 to 56 million meetings and $70 to $283 billion.
"Google proved psychological safety is the number one factor." One company, 180 teams, no published effect sizes, not peer reviewed, and omitted from the 2024 systematic review of the field.
"Agendas make meetings effective." Correlation r = .14 to .18, near the bottom of the ranking. Goal clarity is about three times stronger.
"The Feynman technique." Traceable to a blog post around 2011. Feynman never described it. The underlying mechanisms are real and are taught here under their own names.
"Storytelling is enormously persuasive, r = .44." That figure is the correlation between transportation and attitude, not the causal effect of narrative. The causal estimate is about .17 to .23.
"Men interrupt women far more." The meta-analytic effect overall is d = .15, described as negligible. For intrusive interruptions specifically it is d = .33, which is worth teaching, and it is not the claim usually made.
Worked exemplars
Five published artefacts. Every one is a real document used to assess or produce speech under pressure.
These were chosen because somebody published the instrument rather than describing it.
Two are band scales from assessments that gate careers: the IELTS fluency descriptors, which are unusually precise about why a speaker hesitated, and the World Universities Debating Championship speaker scale, whose band boundaries are entirely about whether a reason survives contact with a counter-argument.
One is a taxonomy of evasion built from verbatim transcripts of broadcast interviews and presidential press conferences, which lets you name what a speaker is doing rather than sensing it.
One is an operational protocol from healthcare and aviation, where the cost of a misheard instruction is measured in incidents rather than embarrassment.
And one is the coding scheme from the only training programme that has raised charisma in a controlled trial, which is a checklist a peer can apply after twenty minutes of training.
Read them for what they attend to, which is almost never what a communication course attends to.
What is open, and what is not
The free levels of this programme are readable with no account at all — the real levels, not samples. An account carries your progress, your gate claims and your saved work. The remaining levels, the tools, the gates and this programme’s worked exemplars are opened together when you enrol.
Create an account