Tag: presentation AI workflow

I watched a board ignore 22 perfect AI-written slides — because not one of them asked for a decision.

Quick Answer: AI generates content — clear sentences, reasonable data points, professional formatting. What it can’t generate is AI presentation structure: the decision architecture that determines which slide goes where, what the room needs to decide, and why the evidence is sequenced to lead them there. If you ask AI to “create a board presentation,” you’ll get 15-20 slides of competent content with no argument. The fix: build the structural skeleton first (what decision, what recommendation, what evidence in what order), then use AI to fill each section.

A client — a VP at a technology company — sent me his board presentation and asked for feedback. It was 22 slides. Beautifully written. Consistent formatting. Every slide had clear bullet points and supporting data.

He’d used ChatGPT to build it, and the output was impressive. Clean language. Professional tone. Relevant content.

One problem: nowhere in 22 slides did it say what decision the board needed to make.

There was no recommendation. No “I’m asking for X by Y date.” No comparison of options with trade-offs. No cost of inaction. Just 22 slides of well-written information, sequenced in the order the AI had generated it — which was the order of his prompt, not the order of a decision-first argument.

I asked him: “If the board reads only slide 1, do they know what you’re asking for?” He looked at slide 1. It was a project overview. They wouldn’t know the decision until slide 19.

We restructured in 90 minutes. Same data, same AI-written content — but reorganised around a decision architecture. Recommendation on slide 2, evidence supporting it, options with trade-offs, specific ask with a deadline.

The board approved it in the first 10 minutes.

The Difference Between Content and Structure (And Why AI Only Gives You One)

Content is what your slides say. Structure is the order they say it in and why.

AI is extraordinarily good at content. Ask ChatGPT to “write a slide about Q3 revenue performance” and you’ll get a clear, professional summary with relevant data points. Ask it to “write 15 slides for a board presentation on Project Phoenix” and you’ll get 15 clear, professional slides.

What you won’t get is an argument. Because an argument requires something AI doesn’t have: knowledge of the decision-maker, the political context, the urgency, the alternatives, and the specific outcome you need from the room.

AI presentation structure fails because AI sequences content in the order it was prompted, not in the order that leads a room to a decision. It generates in narrative order (background → context → analysis → findings → recommendation) when executive communication requires decision-first order (recommendation → evidence → options → ask).

This is the fundamental gap. It’s not about better prompts, more specific instructions, or a different AI tool. It’s about the structural logic that determines what goes on slide 1, what goes on slide 5, and what the room is doing on slide 10.

For more on the difference between AI-enhanced and AI-generated presentations, see the full comparison.

Why do AI-generated presentations fail with executives?

Because executives read slides in decision mode — they’re looking for the recommendation, the risk, the cost, and the ask. AI generates slides in information mode — sequenced to inform, not to persuade. When an executive hits slide 5 and still doesn’t know what you’re asking for, they check out. The content might be better than anything you’d write manually. But without decision architecture, it’s like having a perfectly worded email with no subject line.

Why AI Presentations Fail in Executive Settings

After reviewing hundreds of AI-generated executive decks — from clients using ChatGPT, Copilot, Gamma, and others — I see the same three structural failures every time.

Failure 1: The recommendation is buried. AI typically generates in chronological or logical order: background first, analysis second, conclusions third, recommendation last. In a 20-slide deck, the recommendation lands on slide 17-20. By then, three executives have left and two more are on their phones. Executive presentations need the recommendation on slide 1 or 2 — everything after that is evidence supporting the ask.

Failure 2: No options or trade-offs. AI generates a single recommendation because that’s what it was asked for. But decision-makers need options. “I recommend A” gives the room two choices: yes or defer. “Here are three options with costed trade-offs, and I recommend A because…” gives them agency. AI doesn’t create options unless specifically prompted — and even then, it doesn’t quantify the trade-offs the way an executive audience needs.

Failure 3: No cost of inaction. The most powerful slide in any decision deck is the one that shows what happens if the room doesn’t decide. AI never generates this slide because it doesn’t understand that executive meetings exist to make decisions, and that deferral is the default outcome unless you make it expensive. The decision slide structure includes this by default — AI doesn’t.

The Structure-First AI Workflow: Decision → Skeleton → AI

The fix is simple but counterintuitive: you need to build the structural skeleton BEFORE you open AI. Most people do the opposite — they prompt AI first, then try to restructure the output. That’s backwards.

Step 1: Define the decision. Before you write a single prompt, answer: “What specific decision do I need from this room?” Not “inform them about the project.” Not “update them on progress.” A decision: “Approve £400K additional budget by March 7.” If you can’t state the decision in one sentence, you’re not ready to build slides — with or without AI.

Step 2: Build the skeleton. Choose a template that matches your scenario. A board presentation needs a different skeleton than a project status update, which needs a different skeleton than an investment proposal. The skeleton determines what goes on each slide and in what order — recommendation first, evidence second, options third, ask last.

Step 3: Prompt AI to fill each section. Now — and only now — use AI. But not with a single prompt like “create a board presentation.” Instead, prompt section by section: “Write the executive summary for a £400K technology investment. The recommendation is to approve. The key evidence is…” When AI fills a pre-built structure, the output has the decision architecture the room needs.

This is the approach that turned my client’s 22-slide information deck into a 12-slide decision deck — same data, same AI-generated language, fundamentally different outcome.

For a library of proven prompts, see the complete guide to ChatGPT prompts for presentations.

The 3-Prompt System: Draft → Refine → Executive Polish

One prompt doesn’t produce executive-quality output. Three prompts do — if they’re sequenced correctly.

Prompt 1: Draft. Generate the content for a specific slide or section. Be specific about the scenario, the audience, and the data. “Create content for a Q3 business review for the finance committee. Include: revenue vs target, three significant wins with quantified impact, two challenges with root causes, and three priorities for next quarter.”

Prompt 2: Refine. Sharpen the output for the specific audience. “Make this more impactful for a CFO audience. Each win should quantify business impact. Challenges should include what we’re doing about them. Remove metrics that don’t connect to business outcomes.”

Prompt 3: Executive Polish. Stress-test it. “Review this through the eyes of a CEO with five other meetings today. What would they skip? What questions would they ask? Strengthen the ‘so what’ for each point. Ensure the decision is specific and time-bound.”

Each prompt layer adds something the previous one didn’t: the Draft gives you content, the Refine makes it audience-specific, and the Polish makes it decision-ready. Without the structural skeleton underneath, all three layers produce better-written information. With the skeleton, they produce an argument.

What AI IS Good At (Once the Structure Exists)

This isn’t an anti-AI article. AI is transformative for presentations — but only when it fills a structure rather than creating one.

Once you have the decision architecture in place, AI excels at: generating clear, professional language for each section; stress-testing your content from the audience’s perspective; finding gaps in your logic that you’ve become blind to; polishing language to be more concise and direct; and creating supporting data visualisations.

The combination of human structure + AI content is more powerful than either alone. You bring the judgement (what decision, what audience, what politics). AI brings the execution speed (clear language, consistent tone, gap identification). The structural skeleton is the interface between the two.

The professionals who are most effective with AI aren’t the ones writing the best prompts. They’re the ones who know what the room needs BEFORE they open ChatGPT. Structure first. AI second. That’s the workflow that gets decisions.

Frequently Asked Questions

Can’t I just write better prompts instead of using templates?

Better prompts produce better content — but content isn’t the problem. The problem is structural logic: what goes on slide 1, what goes on slide 5, why the evidence is sequenced the way it is. No prompt, however sophisticated, gives AI the knowledge of your decision-maker, the political dynamics in the room, or the specific decision the meeting exists to make. Templates provide the structural skeleton that prompts can’t. Then prompts fill it brilliantly.

Does this work with ChatGPT, Copilot, and other AI tools?

Yes — because the structural problem is universal across all AI tools. ChatGPT, Copilot, Gamma, Claude, and every other AI presentation tool generates content in information mode. None of them generate in decision-first mode unless you provide the structure first. The templates work with any tool. The 51 AI prompts are written for ChatGPT-style interfaces but adapt to any conversational AI.

How long does the structure-first workflow take?

About 30 minutes for a complete executive deck. Five minutes to choose the right template for your scenario (the playbooks tell you which one). Five minutes to define the decision, recommendation, and key evidence points. Twenty minutes to run the three prompts per section and review the output. Compare that to 3-4 hours of prompt-iterate-restructure-prompt cycles when starting with AI alone.

What if my presentation is informational, not decision-based?

Most presentations that claim to be “informational” actually contain an implicit decision. A project status update implicitly asks “should we continue as planned?” A quarterly review implicitly asks “is this team performing?” If you genuinely need to inform without seeking a decision — a training session or a knowledge-share, for example — AI alone works fine. But for any presentation to leadership, there’s almost always a decision embedded. Find it, make it explicit, and build the structure around it.

If your board pack goes out tomorrow morning — or your SteerCo pre-read is due by 5pm — don’t let AI decide the slide order. Build the structural skeleton first. Then let AI fill it. That’s how 22 slides of information become 12 slides that get a decision.