Presenters

The last section lists claims that failed verification. Several are near-universally repeated. Do not cite them.

Contents

1. The methods
2. Measured artifacts
3. What the research actually supports
4. Where the canon agrees
5. Where it genuinely conflicts
6. The numbers
7. Do not cite

The methods

Michael Alley — assertion–evidence

The most rigorously tested slide structure in existence. Penn State’s published checklist prescribes: “Begin each body slide with a sentence-assertion headline that is left justified and no more than two lines”; “Support the assertion headline with visual evidence… avoid bullet lists”; “design your slides so that the audience reads no more than 20 words per minute”; keep lists to two, three, or four items; and “leave at least a half-inch of white space below the headline.” Typography: bold sans, 28 pt headline / 18–24 pt body / 12–14 pt references — all on a 4:3 canvas, so normalize by ×1.3333 for widescreen (see slide-craft.md).

Effect sizes exist but weren’t obtained at source; describe the structure as well-supported without quoting numbers.

Nancy Duarte — the artifact taxonomy

Her most useful contribution isn’t a slide rule, it’s the three-way split: document (prose, read) → slidedoc (prose + visuals, read, one point per page, self-explanatory) → presentation (visuals, spoken, requires a presenter). She classifies “Research Findings,” “Reports,” and “Status Update” as slidedocs — which means most analytics work is not a presentation by her own taxonomy.

The Glance Test: “People should be able to comprehend each one in about three seconds.” (Note: this is verifiable from her HBR writing; the duarte.com blog URLs that circulate for it are dead, and there is no separately named “3-second rule” for annotation — use “Glance Test, 3 seconds.”)

Her Sparkline structures persuasive talks: what is → call to adventure → alternating what-is / what-could-be → call to action → new bliss. Full slidedoc specs are in deck-genres.md.

Garr Reynolds — Presentation Zen

Restraint, simplicity, naturalness. Coined “slideument” as a pejorative for the hybrid artifact. For charts specifically: “write a declarative statement rather than a title.” Prescribes going analog before opening slide software, and a separate handout rather than a dense projected deck.

Cole Nussbaumer Knaflic — Storytelling with Data

The live↔sent continuum, storyboarding with post-its before software, decluttering, preattentive attributes, and greying the context to colour the point.

Attribution correction: the Big Idea’s three components (point of view, what’s at stake, a complete sentence) are Duarte’s, from Resonate — Knaflic credits her explicitly. Our own dreambase-data-stories currently calls it “Knaflic’s test”; it should be attributed to Duarte.

Barbara Minto — the Pyramid Principle

Governing thought at the top, 3–4 grouped supporting arguments, MECE ordering. Two corrections worth carrying: Minto trademarked SCQ (situation, complication, question) — the Answer is the pyramid apex, not a fourth element of the acronym. And she coined the MECE acronym while crediting Aristotle for the principle itself.

Gene Zelazny — Say It With Charts

First published 1985. His chart-selection framework predates Cleveland & McGill, so follow the perceptual literature where they conflict (notably on pie charts) and treat his matrix as heritage. His framework also maps two valid forms to several comparison types — the popular one-message-one-chart-type table that circulates is a simplification. Cite the books; his often-quoted McKinsey job title is unverified.

Consulting deck craft

Action titles (“so what” titles), horizontal logic (the titles read in sequence are the argument), vertical logic within a slide, storyline-first ghost decks, and mandatory source footers. Documented convention caps titles at two lines / ~15 words.

Important sourcing caveat: McKinsey’s own site could not be reached in this research — zero McKinsey exhibits were actually read. Everything above comes from ex-consultant writeups and published books. Describe these as consulting conventions, not as “McKinsey’s rules,” and never reconstruct a firm’s house style from memory.

Tufte and Doumont — the density argument

Tufte’s The Cognitive Style of PowerPoint argues that slide bullet outlines destroy analytical reasoning, and his NASA analysis is genuinely damning — but it is about Columbia, not Challenger. (The Challenger material is the O-ring chart analysis in Visual Explanations. These get conflated constantly.)

Jean-luc Doumont’s rebuttal exists, is fully verified (Technical Communication 52(1), Feb 2005, pp. 64–70, free at MIT), and is the best counterweight in the canon. He shows the “extremely low resolution” claim is technically wrong — a projected slide is roughly 300 dpi equivalent — and argues Tufte’s density prescription misunderstands oral communication. His own guidance: a headline that is a complete sentence with subject and verb, ≤12 words, ≤2 lines, “not a question… but an answer”; three main points, maximum five; and rules-of-thumb like “N bullets × M words” “make little sense” — use visual limits instead.

The resolution: Tufte’s density argument properly applies to the paper, not the projected slide. A skill that applies it to slides has misread him. Both men agree you should also write the document.

Chris Anderson / TED

The throughline — one idea, stated early, that the talk is about. The 18-minute cap. And a correction: TED does not ban bullet points — their own guidance permits progressive reveal. That myth circulates widely.

Measured artifacts

These patterns come from primary files that were downloaded and measured, not from articles about them.

Benedict Evans — the most transferable slide grammar found. Five type roles on every slide: headline (the claim, serif) / subhead (the because) / chart definition with units / source / identity footer. Serif for argument, sans for data. One signal red plus charcoal, extended by tinting the same hue rather than adding colours. Zero annotation on the plot — the benchmark comparison lives in the subhead. 59–106 slides, always 16:9, median ~48 tokens per slide, and only 36–74% of slides are charts.

Mary Meeker / BOND — the title grammar is literally Metric = verdict + number vs. number (“Global Internet User Growth = Solid But Slowing +6% vs. +7% Y/Y”), applied to content slides and section dividers. ~68% of slides carry a source line including the date of data collection. The encoding goes in the axis label (“Internet Users (Global Blue Bar)”) rather than a legend. Deliberately 4:3 — it’s a document that happens to be paginated as slides. Don’t copy the density unless you’re also building a reference document.

Amazon — “We don’t do PowerPoint (or any other slide-oriented) presentations at Amazon. Instead, we write narratively structured six-page memos. We silently read one at the beginning of each meeting in a kind of ‘study hall.’” On why quality varies: great memos “are written and re-written, shared with colleagues who are asked to improve the work, set aside for a couple of days, and then edited again with a fresh mind.” Offer this when the material is genuinely narrative — and be honest that it can’t be done in an afternoon.

Netflix Culture deck — declares the format contract on slide 1: “These slides are meant for reading, rather than presenting.” Visually plain, black text with one green italic emphasis, and a “Seven Aspects” spine slide repeated at every section boundary. Proof that substance beats polish — and that saying which artifact you built is itself a design act.

Hans Rosling / Gapminder — one encoding, taught once, then reused by swapping a single axis. The animated bubble chart is narration, not decoration.

Bret Victor — the standard the medium is capable of, and the sharpest test for interactivity: can the audience rebut you by modifying your model?

Publication conventions worth stealing: Our World in Data’s footer contract (Data source: in bold, sources semicolon-separated with years, then the URL and licence) — which exists precisely because charts get screenshotted and separated from their page. And the UK Analysis Function’s rule, stated outright: headline title = the message, statistical subtitle = metric + geography + period.

What the research actually supports

The empirical record contradicts the folklore in both directions.

• “Death by PowerPoint” is not an empirical finding. PowerPoint vs. no PowerPoint is a clean null (Baker et al. 2018, 48 studies, g = 0.067). The tool isn’t the problem; the defaults and the culture are.
• “One word per slide” is wrong. Yue, Bjork & Bjork (2013) found abridged on-screen text beat a no-text control on transfer (d = 0.60) — and narration over a blank screen was catastrophically worst. Far-paraphrased text was as bad as verbatim. The winning condition is a sentence headline plus minimal keyed labels.
• Never validate a design against audience preference. Three independent studies show audiences prefer the design they learn less from. Test comprehension, not approval.
• Mayer’s effect sizes are inflated in popular retellings. The independent meta-analysis of his own corpus (Cromley & Chen 2025, 591 effects) finds g = 0.37, declining year over year. The principles still point the right way — coherence, signalling, segmenting, redundancy — but don’t quote the headline numbers.
• Animation is not automatically better than static (Tversky et al. 2002), and where it appears to win, the comparison is usually unfair — the animated version carried more information. See interactive.md for the conditions under which it genuinely helps.

Where the canon agrees

Safe to encode as defaults, because every authority arrived independently:

1. Decide the artifact before designing. Reynolds (“slideument”), Duarte (three-way split), Knaflic (live↔sent), Doumont (companion handout), Alley (projected ≠ handout), Tufte (write the report).
2. The title states the conclusion, in a complete sentence, in ≤2 lines. Six traditions. The most over-determined rule in the field.
3. Reading only the titles should tell the whole story.
4. Never start in slide software. Analog first — cards, post-its, a dot-dash outline.
5. One message per surface.
6. Grey the context, colour the point.
7. Never read the slide aloud — and this one has real evidence behind it.
8. Strip logos, footers, and template chrome from body slides. Consulting decks are the deliberate exception, because they’re read documents where the source line and page number are load-bearing.

Where it genuinely conflicts

The numbers

• ≤2 lines — headline cap. Alley, Duarte, Doumont, and consulting convention, independently.
• ≤12 words (Doumont) / ~15 words (consulting action titles).
• 28 pt headline / 18–24 pt body — Alley’s tested typography, on 4:3. Normalize ×1.3333 for widescreen.
• ~21 words per slide — the tested assertion-evidence density (~42 was the “common practice” comparison).
• 3 seconds — Duarte’s Glance Test.
• 100 words/page concise, up to 175, >250 → write a document — Duarte’s slidedoc thresholds.
• 5 slides — “the tolerance level for people in power.”
• 10 pages / 10 minutes read at a meeting; 20 minutes for a pre-read.
• 3 main points, max 5 — Doumont.
• 18 minutes — TED’s cap.
• ~20 pages, ~50 words/slide, 11–23 seconds per slide — DocSend/HBS, for decks read alone.
• g = 0.37 — Mayer’s corpus, independently meta-analysed. g = 0.067 — PowerPoint vs. no PowerPoint.

Do not cite

Widely repeated, and either untraceable to a primary source or contradicted by one.

Fabricated content. At least one well-ranking site hosts ~40 plausible-sounding but AI-generated Zelazny and Minto quotes. If a Minto or Zelazny quote can’t be traced to the actual book, don’t use it.

Misattributions.

• The Big Idea’s three components are Duarte’s, not Knaflic’s.
• SCQ, not SCQA, is Minto’s trademark.
• The consulting “Dear box” and “so what box” do not exist. The element is a takeaway box or kicker box. (“Dear box” is probably a garbling of Zelazny’s Dear Shirley letter.)
• 10/20/30 is Kawasaki’s, not Steve Jobs’ or Apple’s.
• Tufte’s PowerPoint analysis is about Columbia; Challenger is the O-ring material in a different book.

Myths.

• TED does not ban bullet points.
• Steve Jobs slide “rules” — one word per slide, a minimum font size, a maximum bullet count, a three-act keynote structure — are not verifiable from any Apple source; they trace to third-party books.
• “YC recommends 100 pt type.” Kevin Hale’s actual post gives no numeric font size — only “use large type, bold text, a simple font, good contrast.” Never attribute a size to YC.
• “Death by PowerPoint” as a research finding.
• Tufte’s “PowerPoint costs the US economy $X billion” and the “~40 words per slide” statistic.

Numbers that don’t check out.

• DocSend’s “21 seconds per slide” — unverifiable at source. The verified figures are ~19.2 pages, ~50 words/slide, 11–23 seconds. (“19 slides” is secondary rounding of 19.2 pages.)
• BOND’s “339 pages” — a template artifact in their viewer; the 2019 PDF is 334.
• Any generic slide count for “Benedict Evans’ deck” — his decks run 59, 79, 90, and 103 pages depending on the year. Measure the file you’re citing.
• Amazon six-pager mechanics beyond the 2017 letter (fonts, margins, appendix caps, “banned company-wide”) — folklore.

Institutional conventions that were not verifiable. McKinsey/MGI exhibit conventions, IMF chart-pack conventions, a World Bank data-visualization style guide (no such document was found — don’t assert it exists), Pew’s chart palette, and WEF caption formats. If asked to match one of these houses, say what’s actually known and ask for a real example file.

Licensing corrections that matter if you vendor anything. owid-grapher is no longer MIT — it’s source-available and reuse requires written permission. bbc/bbplot has no licence file at all. And #FFF1E5 is the FT’s paper colour, not “FT pink” (#fcd0b1).