We try to be the Strategist (crafting a logical argument) and the Designer (working on aesthetics) simultaneously.
The result? We do neither well. We end up with "Frankendecks"—slides cluttered with walls of text, confused logic, and inconsistent fonts.
Here is how I engineered a system that forces me to think like the world's best presentation experts, and why I stopped using Google Slides entirely.
The Core Philosophy: Separation of Concerns
In software engineering, Separation of Concerns is a non-negotiable principle. You don't put your database logic in your user interface code. Yet, when we use standard AI tools (like ChatGPT) for presentations, we ask them to "Create a 10-slide pitch deck."
We are asking the AI to be the architect and the bricklayer at the same time. It inevitably fails.
To fix this, I split the system into two distinct Prompts that work in tandem:
- Maestro (The Writer): Creates the narrative strategy.
- Virtuoso (The Art Director): Creates a beautiful, responsive single-page HTML slide deck.
The Maestro prompt generates a structured JSON output, which is then processed by the Virtuoso prompt. This ensures the "thinking" is finalized before the "drawing" begins.
Let's break it down.
Phase 1: Maestro (The Narrative Creator)
I hypothesized that when you ask an AI to "create a presentation," it defaults to the average way of doing things. To get exceptional results, you must identify the Timeless Wisdom of that domain and codify it into the prompt.
I set out to identify the state-of-the-art methods for structuring narratives. I found that the goal of the presentation dictates the framework:
- To Sell or Persuade: The system selects Duarte's Sparkline or the SCQA Framework (Situation, Complication, Question, Answer). It creates tension between "what is" and "what could be."
- To Educate or Inform: The system selects the CRI Framework (Context, Relevance, Implication) or the Pyramid Principle. It focuses on clarity and logical grouping.
The "Stop and Think" Protocol
Instead of generating slides immediately, Maestro acts as a consultant. It forces me to define the Intent first. It guides me through a dynamic questionnaire that I codified to ensure I never forget a piece of context:
- "Is the goal to Inform/Educate or to Persuade/Drive a Decision?"
- "Who exactly is the audience? What do they believe right now?"
- "What is their likely disposition? (Skeptical, Supportive, Hostile?)"
- "If they were to walk away with just one idea or make one decision, what would it be?"
This last question forces me to define the "Governed Thought." If I can't articulate that one idea, the system refuses to build the deck.
Advanced Workflow: The "Synthesis Engine"
I don't just use this for my own ideas. I often feed Maestro complex research papers, technical articles, or disorganized notes. Because it is grounded in these logical frameworks, it acts as a synthesis engine. It strips away the academic fluff and extracts the core narrative arc, helping me understand (and explain) dense topics faster.
Pro Tip: When the stakes are high, I don't just say "The audience is engineers." I research the specific participants—their backgrounds, their recent tweets, their published work—and feed that context into Maestro. The resulting narrative is hyper-tailored to their specific mental models, not a generic persona.
Phase 2: Virtuoso (The Art Director)
The Problem: Even with a great story, AI struggles with visual layout. When I asked it to "make it look good," it would guess at the design, often resulting in broken layouts or overflowing text boxes.
The Solution: I stopped asking the AI to be artistic. Instead, I codified the principles of Cognitive Load Management directly into the code.
I implemented a strict set of CSS rules—an "Architectural Library"—that the AI must follow.
- Cognitive Load Calculation: The system analyzes the text density of every slide. If a slide is too complex, it automatically triggers a "High Load" CSS class. This increases white space, removes decorative elements, and simplifies typography to protect the audience from exhaustion.
- The Rhythm: The system ensures the deck follows a high-low-high rhythm (alternating between dense data and punchy visuals), so the audience never feels left behind.
Why HTML & CSS?
You might wonder: Why go through the trouble of coding a presentation instead of just using PowerPoint?
The answer is control. Standard presentation tools are static; the web is alive.
- Icons as Visual Anchors: I can instantly tap into massive icon libraries (like Google Material Symbols). I don't have to drag-and-drop SVGs; I just tell the prompt "Use a 'Rocket' icon," and it renders perfectly aligned, scalable vector graphics.
- Animations that Inform: CSS allows for "Progressive Disclosure." I don't just fade things in; I can animate a timeline to draw itself as I speak, or have a chart grow to emphasize a metric.
- Responsiveness: My "deck" is actually a website. It renders perfectly on a projector, a laptop, or even a phone, without me ever having to pinch-to-zoom.
By treating presentations as a code problem rather than a design problem, I didn't just automate the work—I elevated the output. The system prevents me from making the lazy choices that lead to bad slides.
Ready to give it a spin?
Try the Maestro Prompt →