Evidence against and open controversies

This file gathers the evidence against the presentation rules: null results, disputed studies, and advice that a study contradicts, by theme.

Against assertion-evidence

One experiment carries most of the weight: Garner 2013 (n = 110, one topic, one 8-minute talk, engineering students).
The coauthor Alley is the main designer and promoter of the approach.
The owner's site says it makes talks more understandable and cites no study on its front page (Leonhard Center 2026, 9 of 20).
Other tests are not clean: Alley 2007 bundles the design with active learning and shares an author with Garner 2013; Root Kustritz 2014 let students choose their session.
The arms differ in words per slide (21.2 against 41.5) and in graphics; the design is a package, so the cause is unknown.
AE viewers rated their slides as having a less appropriate amount of text than the other group did (Garner 2013).
Immediate fact recall did not differ; the gain was in explanation and delayed items.
Later studies measured other things: slide makers' workload (Alzayed 2023), grammar scores (Sripradith 2023), and pilots (Wolfe 2023).

Against the principles as rules for talks

Most principle studies are short narrated lessons on a computer; a live talk adds a person, a room and questions.
Redundancy effects depend on prior knowledge and picture content (Adesope 2012); "never put text on the slide" overreaches.
The modality effect disappears under self-pacing (d = -0.14, 95% CI -0.32 to 0.04, 7 effects; full text); a talk is system-paced, so that result fits it better than it fits a page.
Segmenting helped more for high prior knowledge in one account (secondary).
Seductive details hurt (Sundararajan 2020), yet Wolfe 2023 reports that organiser graphics such as SmartArt can backfire, which supports testing images with an audience.

Against the critics of slides

Tufte 2003 argues the tool is the fault; his case is an essay with an accident report as its example, not a trial.
CAIB 2003 is an investigation of an organisation; slides were one of several factors in what it found. The page numbers 191 and 201 came from a secondary page.
Baker 2018 shows no average harm either, so "slides are bad" is not supported.
Kosslyn 2012 shows violations are common, not that fixing them raises learning.

Against the attention rule

The belief that attention drops after 10 to 15 minutes has little primary support (Wilson 2007, Bradbury 2016).
Short fixed formats (Ignite, five minutes) cannot cite it as evidence.

Against practice formats

Liao 2020: the differences between PechaKucha and a traditional talk mostly reflected the selection of less material to fit the time, not the format.
Bakcek 2020 (randomised, n = 134): equal test scores; only satisfaction was higher.
Warmuth 2021: instructors rated PechaKucha higher, but delayed retention did not differ for presenters; audience retention was self-reported.

Against accessibility advice and the redundancy result

Guides ask speakers to say every slide word aloud (Rector 2019; ACM 2026), which is the redundancy that Adesope 2012 and Yue 2013 found unhelpful for sighted learners.
Rector 2019 itself allows more text on slides when the speaker talks fast. No study of blind or low-vision listeners on this trade-off was found.

Against the speaker-effect claims

A visible face has no consistent effect over 12 studies (Henderson 2021).
Enthusiasm raised engagement but not quiz scores (Marty-Dugas 2024).
Agent studies use animated agents, not live speakers (Castro-Alonso 2021).

Against motion on slides and in talks

The build studies are small and recorded: n = 99 and n = 40, one lesson each (Liu 2022, Ito 2026); one course found exams unaffected by animated slides (Miller 2011).
Learner control beat system pacing on transfer (Tabbers 2009), but cost time; a speaker-paced build is the system-paced arm.
A whiteboard animation raised satisfaction and not knowledge, and its authors suspect seductive details (Meij 2021); engaged viewers comprehended slightly less (Li 2023, secondary).
A slides-only video matched an animation, at lower cost (Pichler 2023).
Watching time is not learning (Guo 2014); its 6-minute rule comes from engagement logs in four edX courses (computer science, statistics, AI, chemistry).
Live coding did not beat static code on learning in one controlled study (Shah 2023). Active live coding did not beat traditional live coding (Shah 2025). The engagement gain is not a learning gain.
Accessibility guides ask for no animation (Rector 2019); the WCAG rule is only to allow pausing (W3C 2023d).

Against colour claims

Emotional design: the colour-only subgroup is 5 of 33 samples (Brom 2018, Table 3); studies disagree on learning and agree on motivation (Mutlu-Bayraktar 2024); designs are too varied to compare (Desiron 2026). Wong 2020's updated values are a little smaller than Brom 2018's.
Colour as retrieval cue did not help (Meusel 2024); hue-signalled sections helped one test and not another (Clariana 2009).
Rainbow maps: a model found they can help some inference tasks (Reda 2021, secondary), so the ban is a default, not a law.
The rule of dark ground for projection has no support in the one experiment found (Greco 2008, secondary).
Harmony: contrasting hue pairs, the "complementary" ones of the colour wheel, were rated less harmonious (Schloss 2011, secondary); preference is not comprehension.
Practice cites weakly sourced figures: a 90% claim for colour in quick judgments, citing a study by name without a working link (Ethos3 2026, 2.5 of 20).
Old colour-versus-monochrome studies were excluded by Brom 2018 because the colours were arbitrary.

Against, for business presentations (added 2026-10-29)

Section to append to against.md, in the same form.

Against slides in organisations

Kernbach 2015: a systematic review finds 18 constraining qualities of PowerPoint in business and education, grouped as cognitive, emotional and social; the number of papers was not read.
Kosslyn 2012 (held): violations of perception and memory principles occur in every field, business included; observers often fail to spot them or to say what is wrong, so critique by eye is unreliable.
Tufte 2003 (held) is an essay; Tufte 2022 reports a company's move away from slides on his own site, with no data (5 of 20).
Baker 2018 (held) shows no average harm from PowerPoint in teaching, so "slides are bad" is not supported even here.
Segel 2010 notes the slide show as one of seven genres of narrative visualisation, which does not condemn it.

Against the memo

Bezos 2018 and Tufte 2022 are claims by the practitioners about their own practice.
No experiment compares memo and deck. F8 returned opinion pieces that agree with one another and cite no measure.
The Amazon rule bundles format, silent reading, writing time and expected discussion; a benefit cannot be put on the format alone.
Bezos 2018 says memo quality varies widely, which is a report that the format alone does not make the thinking.

Against answer first and the pyramid

The evidence for claim-first order is on two-clause texts (Britt 2003, full text), not on decks or talks; the speed gain held only when the claim was clearly marked.
Recall has a serial-position component, so putting a point last is not weak by default; no source for this was kept.
The BLUF standard's origin could not be read; what is said of it rests on secondary pages.
Minto 2026 and Millerd 2025 cite no study; both sell training or a book.

Against passion and for it

Chen 2009: preparedness, not passion, went with funding. Practice advice to "show passion" is not supported by that paper.
Yang 2021b: enthusiasm and preparedness both raised willingness to invest (79 pairs, Chinese VC setting). Cardon 2017 found enthusiasm, preparedness and commitment behave as distinct cues, which sits against treating them as one passion construct (inference; the abstract does not mention Chen 2009).
Pollack 2012 uses funded pitches only, so the sample is cut on the outcome.

Against pitching on merit

Brooks 2014: identical content persuaded more from a male presenter; attractiveness helped men only. A rule "the content carries the pitch" is not supported.
Kanze 2018: the questions themselves differed by presenter gender, and prevention-focused questions went with less funding.
Clark 2008: angels were unaware of the influence of delivery on their ratings.
These results describe bias. A presentation standard can prepare for them; it cannot remove them.

Against reading the practice counts

DocSend 2026 is the owner's data on decks shared through its product (the page says so); the 2022 article does not describe how its sample was drawn.
The two owner figures (just under three minutes in 2022, 2 minutes 42 seconds on an undated page) differ in sample and date, so they are not a trend.
The 3 minutes 44 seconds figure of 2015 is repeated widely and was only seen as a search summary (not used).
Sequoia 2019 offers an anecdote of one success as its evidence.
Huthwaite 2026 is an interview with the originator of SPIN on the training company's own blog.

Where practice contradicts held evidence (Q1 to Q8)

Practice	Held evidence	Verdict
Dense text on investor decks, read alone	Redundancy effects (Adesope 2012, Yue 2013) hold for system-paced material	not a contradiction: a deck read alone is self-paced, and Ginns 2005 finds the modality effect gone under self-pacing (it measures modality, not redundancy; full text)
Memos replace slides because bullets hide weak reasoning	Baker 2018: no average difference; Garner 2013: assertion-evidence slides beat bullets	partial: the held evidence says the slide design matters, not the medium
"Show passion"	Chen 2009: preparedness	practice runs ahead of, and partly against, the evidence
"The content speaks for itself"	Clark 2008; Brooks 2014	evidence against the practice
Consulting action titles	Garner 2013	agree
10 slides, 20 minutes, 30 point (Kawasaki 2005, held, 6.5)	no business study of slide count	convention, untested
Team slide first or always	Bernstein 2017; DocSend 2022	Team content: Bernstein agrees; DocSend counts presence, not position

Against, for casual talks (added 2026-10-29)

Each section: what practice says, what the evidence says, and what is left open.

Against "use humour"

Practice: PyCon says to use humour if you are funny (PyCon 2026); meetups suit entertaining talks (Hilton 2022).
Evidence: related humour is tied to perceived learning, attention and rapport, while gains in learning are mixed and mostly self-reported (Zhou 2025). Concept-related humour helped retention at six weeks, not immediate comprehension (Kaplan 1977, secondary).
Evidence against: interesting but irrelevant material lowers learning (Sundararajan 2020, held; Cheng 2026b, g = -0.16 over 50 studies). Harsh satire, not gentle, lowered perceived message credibility (Freiling 2024).
Frank 2025 measured likability and perceived legitimacy, not trust or learning; its authors warn of backfire effects from satire and did not test harsh satire.
Open: whether a joke in a live talk helps a lay listener remember the point.

Against "tell a story"

Practice: stories are the default advice for lay audiences and for meetups.
Evidence for: narrative offers increased comprehension, interest and engagement (Dahlstrom 2014, abstract). Not verified: recall and the earlier text studies, which sit in the body.
Evidence against: the same review finds stories intrinsically persuasive, with ethical considerations (abstract). Not verified: a health-story meta-analysis, the difficulty of correcting stories with facts, and the human-scale gap, which sit in the body. In health campaigns, statistics moved belief and narrative moved intention (Zebregs 2015, held).
Open: whether a story helps a lay listener understand a mechanism, as against feel interested.

Against "avoid live demos" and against "do live demos"

Practice splits. PyCon says never do a live demo; LornaJane is wary and keeps demos for long slots; Nederbragt teaches by live coding in workshops (PyCon 2026, LornaJane 2017, Nederbragt 2016, held).
Evidence: live coding taught no more than static code, and active live coding taught no more than the traditional kind (Shah 2023, Shah 2025, held).
Both camps rest on experience. Neither is contradicted by a controlled study.
Open: whether a demo helps a lay audience, who were not tested in these studies (students were).

Against "TED views mean it works"

Practice: TED-style talks are the model for popular talks, and views stand for success.
Evidence: clarity scored by a language model predicted views and likes beyond topic and duration, but the model explained 22.5% of the variance in views, and the data are observational (Segal 2026, a preprint). No significant rise in citations was found after a TED talk (uncontrolled; Sugimoto 2013).
Views are not comprehension. The gesture counts quoted for popular talks come from a popular source with no method (seen, not used).
Open: what a viewer remembers a week later.

Against "attention fails at ten or thirty minutes"

Practice: caps at 30 minutes because audiences tire (Hilton 2022).
Evidence: the ten-minute figure has little primary support (Bradbury 2016, Wilson 2007, held).
The 30-minute cap is a courtesy to an evening audience. It is a convention, untested, and not wrong for being untested.

Against "short talks teach more"

Practice: lightning talks, 3-minute theses and PechaKucha are praised as disciplines for the speaker and for the listener.
Evidence: in class trials, PechaKucha gave equal test scores to a normal talk (Bakcek 2020, held); no trial compares a short talk with a long one on a lay audience (G43).
Open: whether the cut helps the listener or only the speaker.

Against "the general public"

Practice: popular talks are written for "the general public".
Evidence: 48% of science-communication articles use a general, non-specified or "public" framing, which the authors call simplistic, and only 4% define their audience (McCarthy 2025).
Attendees of science cafes came from learning and social motives (Childers 2021). The authors say their respondents self-selected, and cite earlier work for cafes reaching mostly people already interested in science; that a general audience is rarely the audience present follows, and remains an inference.

Where practice contradicts held evidence (Q1 to Q9)

Practice	Held evidence	Status
"No more than 60 seconds on any slide" for lightning talks (PyCon 2026; the blog tips were not opened)	numeric slide rules are conventions (position.md)	agree: both untested
Humour as a rule	coherence principle (Sundararajan 2020)	partly contradicts: unrelated humour is detail
Story first	Q9: numbers for belief, story for intent (Zebregs 2015)	partly contradicts for a talk meant to inform
Drop the outline slide	Peyton Jones 1993 says the same (read in the paper), and puts a recap slide about a third of the way through instead; experience, untested	untested
Ask the audience during the talk	Q5: Q&A research is descriptive	no contradiction; classes only (Hunsu 2016: the questions, not the device, carry the gain)

Notes

Seven of the 17 new formal sources are secondary. A claim resting on a secondary number is graded no higher than supported with caveat unless a held full-text review carries it.
Casual-talk practice has no source at 10 of 20. The practice side of this strand is examples only.

Replications and notes

Laptop against longhand notes: the quiz difference did not replicate in a direct replication (Urry 2019).
Handout studies are not randomised (Worthington 2015, full text).
Practice pages report their own success: Designer, Ignite, PechaKucha, the memo, the assertion-evidence site. Of these, all but the memo (Bezos 2018, 10.5) score under 10 of 20.