Skip to content

How AI decks fail

AI can build you a deck in forty seconds. Sometimes that’s all you need, and honestly, sometimes it’s fine. But there’s a reason briefs now land in my inbox with a note that says “AI did this slide, fix it.” The tools produce something that looks finished and reads as junior. Formatted, not designed. Fine for the Tuesday catch-up. Not fine for the pitch, the board, or the client who’s seen a hundred decks and can tell.

I’m Dan Plumb, a presentation designer with sixteen years of agency-grade work behind me, most of it on decks where the stakes were real. This page is about what AI presentation tools actually get wrong, and what I do about it.

Having fixed a fair few of them now, the failures are consistent enough to name.

They format at best, they don't design

The tools place content competently and evenly. Every slide gets the same weight, the same rhythm, the same stock structure. Design is the opposite of that. It's deciding what matters on this slide, for this audience, and making the important thing unmissable.

They don't fit real templates

Generate into a genuine corporate master and things break quietly. Colours get hardcoded instead of inheriting the theme. Fonts drift. Layouts get invented that your brand team has never seen. The deck looks roughly right until it's next to one that actually follows the system.

Things go missing

Force AI output into a template and content gets dropped, cropped or orphaned. Footnotes vanish. Charts lose their sources. Nobody notices until someone senior does.

Everything looks the same

AI decks have a look now, and audiences are learning to spot it. When your proposal resembles the other side's proposal because you both pressed the same button, distinctiveness has left the room.

The judgement layer is empty

Which numbers deserve a full slide. What gets cut. Where the argument needs to slow down. Tools don't know what the meeting is for. That's the part that was never about software.

Fix-ups

You’ve got an AI-generated deck and a meeting it isn’t good enough for. I take it from formatted to designed. That means rebuilding the hierarchy so each slide has a point rather than a layout, restoring brand fidelity where the tool improvised, finding what got cropped, dropped or orphaned on the way in, and applying the judgement the software doesn’t have: what deserves a full slide, what gets cut, where the argument needs room to breathe. The content you generated stays useful. The credibility gets rebuilt around it. Same-week turnarounds are normal, because these decks usually come with a date attached.

Who this is for

Teams whose decks now start life in Gamma, Copilot or ChatGPT and need some of them to end life somewhere better. Brand and ops people watching AI quietly erode a visual identity one generated deck at a time. Agencies whose client-facing work can’t afford to look like everyone else’s. And anyone with a deck that’s fine for Tuesday but not for Thursday’s meeting.

AI-ready templates

Most templates were built before anyone generated into them. An AI-ready template is proper template craft tuned for the way decks get made now: true theme colours rather than hardcoded ones, fonts set at theme level, clean semantic layouts the tools respect, and a slide library people can pull from instead of improvising. When your team uses Copilot or ChatGPT inside a template built this way, the output inherits the brand instead of fighting it. I’ve spent years building templates that survive thousands of users. Surviving AI is the same discipline.

Training

Your team is already using these tools, whichever policy says otherwise. The gap isn’t access, it’s judgement: knowing what good looks like, what to keep from a generated draft, what to fix, and when a deck needs a designer rather than another prompt. I run practical sessions built around how your organisation actually presents, using your template, your real decks and the tools your people genuinely use. They leave knowing how to get more out of the software and, just as importantly, when not to trust it. Sixteen years of deck craft, taught in an afternoon’s worth of habits.

Where AI tools are genuinely useful

This isn’t an anti-AI page. I use these tools daily and they’re excellent at first drafts, structure, speed and volume. They’re genuinely good writing partners too: summarising a forty-page report into slide-sized points, drafting speaker notes, tightening clunky copy, restructuring an argument that isn’t landing. Used that way, they take real hours out of building a deck. The honest position is that they’ve moved the starting line, not the finishing line. The gap between a generated deck and a designed one is exactly the gap between a deck that circulates and a deck that wins. Knowing which one your meeting needs is the skill. I wrote more about what changed this month in ChatGPT designs decks now. And if you’re choosing between them, here’s my independent comparison of the main tools.

AI can build you a deck in forty seconds. It can’t design one for the meeting that matters.

Got a deck that needs to be better than generated?

Tell me what it’s for and when it’s needed. Get started

Senior only
Direct contact
Fast turnaround
On-brand always
Editable files
NDA as standard

FAQ

Can you fix a presentation that was made with AI?

Yes, that’s now a regular part of the work. The content usually survives. The hierarchy, brand fidelity and judgement get rebuilt around it.

What is an AI-ready PowerPoint template?

A template built so AI tools generate into it correctly: theme-level colours and fonts, clean named layouts, and a slide library the tools can draw from. The brand holds even when nobody’s watching.

Should we stop using AI for presentations?

No. Use it for drafts, speed and structure. Bring in design where the outcome matters. The expensive mistake is not knowing which deck is which.

Do you use AI in your own work?

Yes, where it helps, never where judgement is the job. Clients get senior design either way.

How do I know if my deck is good enough?

If the meeting matters and you’re asking, it probably isn’t. Send it over and I’ll tell you straight.