A working method for the talk that lands: narrative-first, figure-driven, and AI-assisted. This module teaches the method and demonstrates what Tōge does — it measures presentation intuition the same way it measures clinical intuition, by comparing your judgment to expert consensus.
A killer talk is not a slide-building exercise — it's a strategic communication event. The audience leaves with one idea they can act on. Everything in this module works backward from that: decide the message, shape the story, then — only then — open PowerPoint. The same skill Tōge measures in surgeons (expert intuition, externalized and compared) is the skill you are building here.
The Three Questions, the narrative arc, and the visual rules are the tacit moves a seasoned speaker makes without naming them. Made explicit, they become a method any speaker can practice.
The 5-Level Review turns "did it work?" into a structured check across narrative, content, design, motion, and assets — the same move IOH makes when it converts opinion into a measured hierarchy.
Every interactive below shows where a natural instinct (more text, more slides, decorative motion) diverges from what experienced presenters actually do. The gap is the lesson.
This is the rule underneath every other rule. The moment a new slide appears, the audience reads it — automatically, before they listen to you. Until they understand what's in front of them, your voice is just noise competing with their own reading. So the slide has to be instantly legible and worth their attention before your narration can land. Design for that, and everything else follows.
The audience arrives motivated mostly by extrinsic factors — they're here, they want it to be worth their time. Before you can invoke the intrinsic reward (the elegant science, the intellectual beauty), you first have to clear the extraneous load and make yourself worth listening to. Earn attention before you spend it. A clean, legible slide is how you earn it.
The single most violated rule in scientific talks: don't put your narration on the slide. When the text and your voice say the same words, the brain wastes effort reconciling the two — and the audience reads instead of listens. If you're going to say it aloud, it does not belong on the slide.
A slide should be understood in about half a second — a single visual beat — so the audience's ears are free for your voice. That is why slides are figure-driven, why titles argue instead of label, and why anything you can say aloud stays off the screen. The next steps build the talk on this foundation.
If you can't answer these three, the talk isn't ready to build — no matter how good the slides look. Fill them in. The build button stays locked until all three are real.
A killer talk follows a story arc, not a manuscript. Click each stage to see what it does — and what it sounds like.
Correct for a paper, deadly for a talk. The audience has to hold everything in memory and assemble meaning at the end.
The audience solves the problem alongside you. By the reveal, the finding feels earned — not dumped.
The five stages are the simple view. Underneath is a three-act structure with two characters: a villain (the problem) the audience can root against, and a hero (your mechanism, made clinically actionable) that defeats it. The villain is revealed progressively — surface problem first, root cause later; the hero is mechanistically linked to it.
Open with what's known, then introduce the problem and the inciting question. Answer "why should I care?" before anything else.
Each experiment builds on the last. Include at least one dramatic reversal where the data contradicts expectation. Color and animation hold the rising complexity together.
The "aha" lands. Mechanism becomes clinical action. End on 2–3 take-home statements (complete thoughts, not bullets) and forward momentum — the next question.
The cardinal rule: if it can be said aloud, it does not go on the slide. The speaker provides the narrative; the slide provides the evidence. Toggle the slide below to see the fix.
Not "Results." Instead: "Plasmin accelerates fracture healing at day 14." The title states the finding; the figure proves it.
Declarative titles, axis/figure labels, key numbers (p, n, effect size), and one take-home sentence. Nothing else earns a place.
Uncluttered slides feel sure of themselves. Cramped slides feel anxious. Leave room for the audience's eyes to rest.
The audience has a fixed processing budget. Every watt spent decoding your slide is a watt not spent on your idea. Three loads to manage:
The effort of processing the presentation itself. Kill it: no text walls, consistent formatting, clean alignment, no decoration.
The material's inherent difficulty. Tame it: simplified schematics over busy clinical photos; break complex processes across sequential slides.
Meaningful processing — the good kind. Feed it: arrows and highlights that guide the eye, morphs that show transformation, links to what they already know.
A simple ratio for where the slide's visual weight goes — and a check against clutter.
60% primary content (the figure or graph) · 30% supporting (labels, axes, legends) · 10% accent (one highlight or arrow). Hold to 3–4 colors per slide, test it in grayscale, and if a slide needs more than ~20 words it needs a figure instead.
The test for any motion: does this help the audience understand something they couldn't understand without it? If not, cut it.
Morph creates visual continuity between slides — but only when the objects share names across consecutive slides. Mismatched names fall back to a generic crossfade, defeating the purpose. The naming convention keeps morph targets sorted at the top of the Selection Pane:
Morph-heavy talks move far faster than the textbook "2 minutes per slide." A calibrated reference point: 374 slides ran in 40 minutes — roughly 6–7 seconds per slide, because each slide is a single visual beat, not a paragraph to read. The next section turns this into a calculator.
Finishing early beats rushing. Set your slide count and your slot; the calculator uses the cinematic calibration (374 slides ≈ 40 min) or the standard 2-min/slide model, and tells you whether you'll make it.
Cinematic: ~6.4s/slide — each slide is one visual beat. Calibrated from 374 slides = 40 min.
Runtime is only half of pacing. The other half: something visual must change every ~10–15 seconds, or attention drifts. A change is a new slide, a build that reveals data, a highlight moving, or a morph — not decoration, but engagement. If you've spent 20 seconds on a static slide, something needed to change at the 10-second mark. Morphs beat hard cuts.
A live demo of the rule: every ten seconds the timer flips and names the next thing to change. Real talks feel like this — never static for long.
Every talk gets reviewed at five levels before it's ready. Always start at Level 1 — a talk can look gorgeous (Level 3) and still tell the wrong story (Level 1). Check what's true; watch your readiness move.
Alongside what you're saying runs an invisible document: what the audience is thinking — the questions forming in their heads. Before each major finding, write the three most likely questions and preempt them in your narration: "You might wonder… here's why that's not quite right." It's a preparation tool, internalized — never written on a slide.
AI is fastest at the parts that are mechanical: drafting a deck skeleton from your narrative, polishing layout, and generating clean figures. It is worst at the part that matters most — deciding the one idea. Use these copy-paste recipes after you've answered the Three Questions, never before. Each prompt has placeholders in [brackets] — fill them in.
Understand first, design second. AI never skips step 1.
A killer talk ends by telling the audience what to DO. So does this module. Here is the whole method as a pre-talk checklist — download it and run it before every talk.
"The slides are the last thing you build. The narrative comes first."