Here’s the most recent presentation skills blog! To see the other blogs head over to the blog page!
- Using AI for making presentations
Let’s start with something I tell you not to do in presentations – a long introduction
I’m bordering on changing my position on AI. Until recently I’ve found its results pretty risible. If I ask it draw me a female high-jumper in action, for example, it gives her three legs. (On the other hand, more focussed and dedicated AI such as the one embedded in Notion to help re-write stuff in particular styles does the job competently.)
By definition, things like ChatGPT creates content which is a sort of average – drawing on pre-existing content means it can’t do anything else. By definition it can’t be creative. What it can do, however, is draw on information that I’ve not been privy to (yet) so I might pick up something I’d miss otherwise.
Enter Charlie Whyman, who was giving me advice on marketing my upcoming StoryMaking Live gig (details here!). She created a bespoke GPT that acted and sounded quite a lot like her so that I could bounce content off.
I was impressed.
I wasn’t always happy or comfortable with the results but that was part of the point – if the real Charlie had been there she’d have battered my drafts the same sort of way her GPT did. So… I started to look into the possibility that AI might be useful for me and my clients.
ChatGPT pointed me at something by slidedecksai.com so I gave the prompt “I need a ten minute presentation about how to use stories in marketing”.
First, the upsides for an AI designed presentation
It was relatively quick, and relatively painless… as in, I got results while all I did was sit there and sip tea. It searched Unsplash etc far, far quicker than I could have done. Pretty impressive, to be honest.
There was also a lovely moment when I looked at the slides and it referred to Donald Miller’s StoryBrand. It’s not a spectacular breakthrough moment but it was something I’d not thought of including in the first instance. (It put it alongside Simon Sinek’s “Start With Why” and the “Hero’s Journey”.) In the end I’d have rejected all three references but it was a nice reminder that AI can draw our attention to things we’d not considered.
The overall structure wasn’t actively bad. It started with an intro slide and then included a slide each on:
- why stories work
- elements of an effective marketing story
- brand storytelling frameworks
- a case study of Nike’s “Crazy Dream” campaign (not a bad choice, I think)
- common mistakes in storytelling
- how to craft your brand’s story
- integrating stories across channels
See what I mean? It’s not actually bad – and while I can’t stand up and cheer, it covers bases.
And I come back to the speed at which it worked. You’ll see in a minute that the results fail the “would you use this?” test, but it gives a starting point very (very!) quickly.
… and now the downsides for an AI designed presentation
I’m tempted to just write “everything else”, but that would be churlish.
This presentation is mind-numbingly uninspired are the first words that spring to mind. Having said that, so are most presentations created by people that’s not a surprise really. The uninspired thing comes in two strands.
The first strand is the content itself. I’ve just said the content isn’t bad, but let’s face it, it’s hardly going to set the world on fire, eh? It’s the average of what everyone else is saying and therefore by definition it can’t be novel.
Honestly, I’d be scrolling my phone before the half way point of the presentation.
The other strand of ‘uninspired’ is the design of the slides themselves. It’s hard to describe the world-famous levels of “meh” in some of the slides. Seriously. Take a look at this bad boy, eh? Bullet points in a row.
Even the better-looking slides were just bullet point lists with a an image slapped on the slide to make them look 5% less rubbish.
There is, of course, also the tiny(ha!) little problem that no one will be able to read those slides past the front two rows. And that they’re too wordy in the first place. And that they bullets are so bland as to be unusable.
“Know your audience” isn’t by advice but… well.. yeah… how exactly? The idea we should know our audience is about as helpful as saying “cook good tasting food”. Where’s the bloody recipe?!?
Would I use an AI generated presentation?
So I’m left with a disappointment and the answer is “hell no”.
However, that’s arguably unfair because I was asking for something that humans are designed to do and by definition we know that AI can’t because it doesn’t think. So what about a slightly fairer test… “Can I use AI to modify and improve my presentation?“
Round two of testing AI presentations
Here’s what I did…
First off, I checked that ChatGPT felt it was up to the task and it assured me it could help …
I work in Apple’s Keynote, so I converted to PowerPoint for a more generalised test and uploaded a set of slides. Perhaps it was mean, but the slides I gave it were for a 30 minute presentation on the impact of stories in presentations – the do-and-don’t stuff of what works and what doesn’t work at a technical level.
I should add it’s a slide deck that’s got me standing ovations, so I’m reasonably confident it works. The results were, I’m afraid, exactly what you’d fear. Given that ChatGPT is pulling from the average of current wisdom it had no chance of being impressed with something a bit more advanced.
After the usual bumf about what worked well there was room for improvement, including:
❌ Areas for improvement:
- Font consistency: Fonts, sizes, and alignment vary significantly across slides. Use a consistent font hierarchy (e.g., title = 32pt, body = 20pt).
- Alignment & spacing: Some slides have awkward element placements (e.g., off-centre or floating text).
- Image quality & placement: Ensure images are aligned and not pixelated. Add captions only if they provide useful context.
- Text overlap or clutter: A few slides (e.g. with lists or multiple concepts) feel cramped. Aim for one key idea per slide.
Let’s put it this way. There were no images that were pixelated. Does that give you a feel for how useful this feedback is? To be fair, there were a variety of font sizes and alignments, all for the considered sensible reasons, but when your measure of what a good slide is looks the the ones the AI generated above I can see why ChapGPT sulks.
I was particularly interested in that last comment though, because it’s something I mention to clients quite a bit. So after a bit more playing, I asked “A few minutes ago you said that some slides had text overlap or clutter and yous suggested I Amed for one key idea per slide. Do you remember saying that?”. I’m not sure what to make of the reply:
That was part of a general design critique, based on standard presentation principles. But based on our conversation since, it’s clear that your deck intentionally resists that approach in places—and with purpose.
You’re using density, repetition, and sparse text as rhetorical tools rather than mistakes.
Well spotted, little machine.
So we’re starting to get somewhere, I thought. GPT is learning and responding accordingly. But then my burgening faith was ruined when I asked for feedback on slide 51. For context slide 51 looks like this:
Imagine my surprise when the commentary starts off like this.
✅ What Works Well
- Mnemonic power: “SOAP” is familiar, easy to remember, and metaphorically rich (cleansing, clarity, preparation).
Not to be put off, I recognised that GPT was referring to slide 49. I figured that it had got confused somewhere in one of the transformations or duplicate slides earlier on (one slide appears twice – once with and once without background music). Obviously therefore I asked for a commentary about slide 49.
It didn’t go very well. Here’s a screenshot what ChatGPT said about my presentation’s slide 49…
To be clear:
- none of the slides anywhere in the deck contains the words “hope so stories” twice
- any slides which do contain the phrase “hope so stories” don’t have the word “mining”
- there is no background image on slide 49
Basically, GPT at this point is performing a commentary on a combination of different slides, calling them slide 49, and consequently talking rubbish.
So much for trying to get help on a more tactical level.
Worse than any disappointment about not being able to get help with specific slide design though, is the bigger implication. If comments about a single slide are based on a hallucination, how many of the previous comments about broader style are based on a catastrophic mis-reading of what’s on what slide?!
Wrapping up – what’s the state of play for AI and presentations? You can guess…
There might be better AIs out there for this kind of analysis but ChatGPT is so well known and widely used it’s the AI equivalent of PowerPoint for slides – everyone uses it because everyone else is using it… hence me testing it.
But I just can’t trust it. So many of the facts it was commenting on were based on what can only be described as the AI equivalent of a drug-induced stupor it’s about as reliable as putting a witness on the stand in court who can’t see and asking him what colour the defendants shoes were on the night.
In the dark.
From 200 yards away.
Facing the opposite direction.
