Twenty research-backed rules — pulled from Tufte, Reynolds, Refactoring UI, NN/g — codified once into the Power Design Claude skill. So the design takes care of itself, every slide.
Twenty rules — one viewport. Each links to its full plate below.
If a slide needs a second headline, it's two slides. Pack two ideas onto one canvas and the audience picks neither.
Slides are scanned, not studied. Three seconds is the budget. If you can't get the point in three, you don't have a slide — you have a document.
Miller said 7±2; Cowan revised to 4. Slides are scanned, not studied — use the stricter limit. Group atoms with proximity so the brain perceives chunks, not items.
Whitespace isn't where the design ends — it's where the design lives. Cramped slides feel anxious. Hero slides want ≥60%; standard ≥40%.
Borrowed from broadcast TV. No critical content within 5% of any edge. On a 1920×1080 canvas that's a 96-pixel breathing band. Crops, projector overscan, and audience eye-paths all start there.
Type sizes shouldn't be invented. Pick one ratio — 1.25, 1.333, 1.414, 1.5, or the golden 1.618 — and let every size on the slide derive mathematically.
Display, subhead, body, caption. That's the budget. A fifth size signals you didn't make a decision.
Body text below 24px on screen — or 28pt for projection — is unreadable from row 10. Title floor: 48px. Caption floor: 18px. Anything smaller assumes the audience moves to your laptop.
Line-height inverts with type size. Big display type wants 1.05–1.2 (almost touching). Body wants 1.4–1.6 to breathe.
Slides shouldn't have paragraphs to begin with. If they do, cap line length at 60 characters. Anything wider and the eye loses its place between lines.
WCAG AA — 4.5:1 body, 3:1 large — is the floor for screens. For projection, design at AAA (7:1) because projectors wash 30–50% of contrast in a real room.
Sixty percent dominant (usually background). Thirty percent secondary (surfaces, structure). Ten percent accent — that's where the eye lands. Reverse the proportions and the slide screams.
Two accents are no accent — emphasis cancels emphasis. One color earns the role of "the eye lands here." Everything else is neutral.
8% of men are red-green colorblind. Encode meaning with shape, weight, position, label, or icon — color reinforces, never substitutes.
Every margin, padding, and gap a multiple of 8 (4 allowed for tight icon work). Use the canonical scale: 8, 16, 24, 32, 48, 64, 96, 128. Never 13. Never 27.
12-column grid with 24–32px gutters covers almost every slide. Every element snaps. No optical drift, no eyeballed centers, no nearly-aligned-but-not.
Proximity is the cheapest grouping signal. Items related to each other ≤16px apart. Items unrelated ≥48px apart. The eye groups by gap.
Strip every chart pixel that isn't carrying meaning. No 3D. No gradients. No drop shadows on bars. No redundant legends. No decorative gridlines. The data IS the design.
Eye-tracking is unambiguous: attention enters top-left, sweeps right, drops, sweeps right, drops. The first 200 vertical pixels are the primary attention zone. Headlines and key visuals belong there.
Two valid modes exist. Presenter mode — sparse, image-led, ≤15 words/slide, narrated live. Document mode — denser, hierarchical, sent as PDF, read on its own. Both are correct. Mixing them in the same deck is what makes audiences hate slides.
Get the skill, the brand library, the full HTML explainers, and weekly walk-throughs inside the AI Automations community.
Join the community