You’ve tried it. You ask Claude or ChatGPT to make a PowerPoint. It comes back with something. The layout is technically fine. The content is organized. But something is off, and everyone in the room can feel it.
That feeling has a name: AI default mode. Most people never get past it. Here’s why it happens, and the exact framework I use to get slides that actually look like a consultant made them.
The Real Problem With AI-Generated Slides
AI tools don’t have taste. They have patterns.
When you ask an AI to build a deck without giving it a design system, it pulls from the average of everything it’s seen: generic corporate fonts, rainbow color palettes, bullet point soup, and clip-art-level icons. The result isn’t bad exactly. It’s just instantly recognizable as machine-made.
The fix isn’t a better AI. It’s a better brief.
The McKinsey Design System in Plain English
McKinsey slides look the way they do because of a set of constraints, not freedoms. The discipline is what creates the premium feel. Here’s the full system:
One color only. Navy (#1A2E4A). That’s it. No teal accents, no orange highlights, no gradient fills. When you limit yourself to one color, every use of it becomes intentional. Emphasis comes from weight, fill, and spacing instead of hue.
White background, always. No dark mode, no textured backgrounds, no full-bleed images behind text. White forces clarity. If your content can’t stand on a white background, the content isn’t ready.
One font family, two weights maximum. Mixing fonts is the fastest way to signal that a slide was assembled rather than designed. Pick one and stay there.
Navy header bar, full bleed. Every slide gets a navy bar at the top with the slide title in white. This creates visual consistency across the entire deck without any effort. The eye always knows where to look first.
Three box types only. Navy fill for the most important callout. White background with a thin border for standard content blocks. Light gray for supporting information. You never need a fourth category. If you think you do, you have a structure problem, not a design problem.
Left-side vertical accent line for insights. When you want to highlight a key finding or recommendation, a thin navy vertical line on the left edge of a text block does the job without a colored background that screams “AI made this.”
Generous whitespace. The thing that separates McKinsey slides from corporate slides is not the navy. It’s the breathing room. Resist the instinct to fill every pixel. Empty space is a design decision, not a missed opportunity.
How to Prompt Claude to Follow This System
The actual workflow is straightforward. You create a Claude Project and paste the design rules above as the system instructions. Then every time you ask for a slide, Claude applies the constraints automatically.
Your prompt doesn’t need to be complicated. Something like this works:
“Build a two-slide executive summary on Q2 sales performance. Follow the McKinsey-Navy design system. Use the three-box structure. Lead with the key insight using the vertical accent line treatment. Keep bullet points to three maximum per box.”
That’s it. What comes back looks nothing like the default AI output because you’ve given Claude a design system to work within, not just a content request.
The Difference in Practice
Default AI prompt: “Make me a PowerPoint about our Q2 sales results.”
Result: Seven slides with eight bullet points each, three different font sizes, a teal accent color, and a SmartArt diagram that nobody asked for.
McKinsey-Navy prompt: Same content request, plus the design system.
Result: Three slides. Clean header bars. One callout box per slide with the key number. Supporting detail in bordered white boxes. The insight highlighted with a left-side line. No decoration that doesn’t carry meaning.
The content is identical. The slides are not.
Why This Matters More Than You Think
Executives read slides in seconds. They’re not reading. They’re scanning. A deck built on the McKinsey-Navy system communicates structure and credibility before a single word lands.
When slides look like AI made them, the subtext is that the presenter didn’t take the material seriously enough to control the output. When slides look like a consultant built them, the subtext is the opposite.
The AI did the work. The system gets the credit.