AI and PowerPoint: Why Your Next Presentation Might Take Minutes, Not Days
On how to create the (almost) perfect PowerPoint slide deck using AI without using third part tools
If you have a primarily desk-based job, then you likely spend roughly a fifth of your working week on presentations. That’s an entire day gone to slide creation. And nearly half of that time isn’t spent on thinking or analysis, but on design, aligning boxes, fixing fonts, wrestling with layouts. If you’ve ever stayed up until 2 am reformatting a slide deck for a board presentation, conference or lecture delivery, you know exactly what I mean.
Enter Generative AI
I’ve been working directly, getting my hands dirty and teaching AI skills to business people since 2023. I’ve seen how AI handles work tasks, writing and design, well and not so well. A reliable PowerPoint solution has successfully eluded me, perhaps until now. I enjoy designing materials and writing content, so I don’t mind spending time in the process, making sure it’s my voice and that “I know” the content. But sometimes, I just don’t have the time in my schedule to spend 8 or 10 hours on a 50+ slide presentation.
Generative AI for writing code? From my amateur standpoint, it seems remarkably good. Text? Very good, but with limitations you need to be aware of. Spreadsheets? I’m impressed so far, and AI seems to be increasingly competent. PowerPoint, however, has been the stubborn holdout—the task that AI seems to find genuinely difficult (think layout, image alignment, whitespace, text boxes, and animations).
There are third-party solutions like Beautiful AI, Make, and GenSpark AI, all of which are impressive based on my testing. But I don’t want another third-party tool to manage. I don’t want another subscription and the increased security concerns that come with it. I want to run the solution on my own machine, using my personal files and the local Microsoft PowerPoint application.
The caveat here is that to use an AI to produce a complex presentation from a detailed document or collection of documents, you’ve got to already know the content. Otherwise, you’re just chancing your arm. Work away if you must, but there seems too much risk to credibility in that for me.
The Problem with Presentations
Here’s the thing about PowerPoint that makes it genuinely difficult for AI. It’s two jobs pretending to be one. You’re doing analysis, synthesising data, building an argument, and structuring a narrative. And you’re doing design, arranging elements on a canvas, managing visual hierarchy, making sure the thing is readable from the back of a conference room and so on. This is a different kind of thinking that incorporates several elements in parallel.
Large language models are mostly trained on text. They understand narrative, argument, and logical flow (arguably). But design, alignment, alternative storytelling maybe not so much. Where should this chart sit relative to that text block? Is there enough contrast between the background and the font? That’s a different kind of intelligence entirely. They don’t possess the intuition and understanding that you do. They don’t understand at all, in fact. They are statistical predictive machines that don’t reason.
I’ve tested several tools and became frustrated and impatient with most of them. The AI gets the content right, but not all the time. The visuals are usually unimpressive. Text sliding underneath decorative boxes. Charts with black text on navy backgrounds. Executive summaries are hidden in tiny fonts while irrelevant details sprawl across the screen. It’s disappointing because you can see the intelligence at work, but the output can’t be used. To boot, they don’t necessarily remember what you told them last time.
Claude Code running your AI Business OS does, however.
Magic Single-Shot Prompts Won’t Work
If you’re looking for a single prompt that solves this, I haven’t found one, and I doubt you will either. What actually works is a system with specific guidelines and constraints, dos and don’ts that guide the AI and prevent it from making formatting and design errors, like what I have built into my AI Business OS. The more complex the task for humans, the more brainpower is required, and the more exact our guidance of the AI needs to be.
Since using Claude Code in my local developer application (no third parties), I’ve learned, for example, that running /command to execute a specified and predetermined workflow (rather than letting it decide) produces dramatically better results. Specifying in my workflow “no border boxes around text elements” eliminates a whole category of layout disasters. Requiring minimum font sizes and contrast ratios catches accessibility issues before they become embarrassing. Specifying brand colours and providing examples also helps guide the AI.
This isn’t about creating unnecessary work for yourself. It’s about acknowledging that presentation generation requires a different kind of precision than text generation. The AI needs guardrails not because it’s stupid, but because the task itself is genuinely complex. You’re asking a language model to think and plan as you do. So give it rules to follow. Take the time to build it once and take a sigh of relief.
Simple PowerPoint Versus Big Ones
Not every presentation needs the same approach. A weekly status update—six to eight slides summarising progress, risks, and next steps or whatever, can arguably be generated in a single-shot prompt. You provide your data, specify your corporate styling, describe the slide structure, and get something usable in only a few minutes. But a presentation of 50+ slides with multiple data sources and a complex narrative? That’s a different story altogether.
I’ve had success using what you might call a multi-stage approach. One conversation for planning the structure, separate conversations for generating each section, and a final pass to check consistency across the whole presentation. I’m asking myself, do the numbers match? Does the story flow? Are the section layouts consistent? Is the brand and styling maintained throughout?
It sounds like a lot more work, and you’d be right, it is more upfront effort. But compare that to multiple people spending a week preparing a quarterly review, or you alone spending two days building a presentation for a conference and putting every other important thing on hold. Suddenly, working with an AI for two to three hours seems reasonable.
What AI Can’t Do
The time you save on producing the presentation doesn’t necessarily allow you to lie carefree in your hammock-that’s not what we are chasing. We want to spend our time working on things that stir our creativity and genuinely excite us. We reallocate this saved time to the parts of presentation work that AI genuinely can’t handle, or something else like rehearsal.
Interpreting the data, anticipating responses and questions, knowing how this thing we’ve studied and analysed is likely to impact the business, the broader market or society. Deciding what to emphasise, what to leave out, and where to adlib. Navigating the politics of who gets credit and who gets blamed. Working out what your audience actually needs to hear versus what you want to say.
These are human elements, and they require understanding context, reading rooms, and managing relationships. No prompt will automate these. And honestly, I’m not sure we’d want it to.
What AI does is strip away the repetitive mechanical labour—the formatting, the layout, the tedious back-and-forth of “can you make the font bigger” and “can we try a different chart type.” That work added little value. Now it happens in minutes instead of hours.
The Human In The Process
None of this works without clear thinking on your part.
AI can’t generate a good presentation from vague inputs. It needs to know what story you’re telling, what data supports it, and what decisions you’re asking for. The logic that used to live in your head, how you reconcile conflicting figures, which metrics matter most, what trade-offs you’re willing to accept, all of that has to become explicit. You have to articulate it and direct the AI accordingly.
In a sense, AI forces us to be clear and unambiguous. Think Daniel Kahneman’s System 2, that aspect of mind that requires deliberation and planning. The AI machine requires precision direction, and in that, it is not truly intelligent-you are, however. If you haven’t thought it through properly, the AI will show it in its output.
Some people will find this frustrating, though, I accept that. I think it’s an opportunity, however. The people who’ll benefit most won’t be the ones with the best prompts. Instead, they’ll be the ones who actually know what they’re trying to say and take the time upfront to guide their AI in the way that it needs to.
🗓️ Learn AI for PowerPoint Live
If you want to see how this actually works in practice, I’m running a free workshop this Friday, 2nd January, at 11 am (Irish time). We’ll cover the basics of AI for both Excel and PowerPoint, with live demonstrations of the techniques I’ve described here. No fluff, just practical workflows you can use immediately. The recording will be available afterwards.