Don't ask AI to build a deck. Do this instead.

Author: Dan Galletta · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Don’t ask AI to build a deck. Do this instead. · Channel: Dan Galletta · 09:05 · Watch on YouTube

Topics: AI evaluation and benchmarks, Data, databases and embeddings, AI content generation, Markets and economy

Summary

The video argues that asking AI to build a complete presentation often produces polished decks that lack the context, reasoning, and audience-specific judgment needed to be reliable. Instead, the presenter recommends using the usual presentation-building process and applying AI selectively across five steps: analyze data, craft the storyline, build the skeleton, build the slides, and review. AI can support analysis, structure and stress-test a storyline, suggest titles and visualizations, and check slides and numbers, but people must verify the work and make the final decisions. For slide creation, the presenter favors reusing previously reviewed, on-brand slides, and promotes SlideBrain as a tool for finding, polishing, and reviewing them.

Key points

A professionally formatted AI-generated deck can still contain unsupported claims or errors; polish is not evidence of sound analysis.
AI lacks the project history, conversations, decisions, and stakeholder nuances required to build a complete and tailored deck from a prompt.
The recommended five-step process is to analyze the data, craft the storyline, build the skeleton, build the slides, and review.
Use AI to summarize clean datasets and identify trends, but verify its analysis, especially when sources are messy or incomplete.
The presenter recommends deciding what the data means yourself, then using AI to structure and stress-test the storyline.
AI can help write action titles, identify supporting evidence, and suggest chart types; reading the titles in sequence tests whether the deck’s skeleton tells a coherent story.
Rather than generating slides with AI, reuse slides that are already on-brand and have been reviewed.
AI can review a deck’s storyline, assess whether slide content supports its action title, and audit numbers and inconsistencies across slides.

Chapters

Start	Chapter	Summary
00:00	Why polished AI decks can be misleading	The presenter uses a flawed report as an example and explains why AI-generated decks can look credible while lacking context, sound reasoning, or audience-specific judgment.
03:07	Steps 1–2: Analyze data and craft the storyline	AI can assist with data synthesis and structuring a story, while people must validate the analysis, interpret its meaning, and account for stakeholders.
05:10	Steps 3–4: Build the skeleton and slides	Use AI for action titles, evidence suggestions, and visualization ideas, then reuse reviewed slides rather than asking AI to build the deck.
08:00	Step 5: Review the presentation	AI can check the deck’s storyline, individual slides, and numbers, but its review tools need careful testing.

Quotes

the polish gives us a false sense of quality — 02:05

AI does the parts it’s good at and you do the parts it isn’t. — 03:07

you decide what the data means, and AI helps you turn it into a story. — 04:08

If they tell the complete story without the slide bodies, your skeleton works. — 05:42

wherever you can reuse slides you’ve already built. — 06:13

Concepts and entities

AI, Deloitte, McKinsey, BCG, SlideBrain, action title, situation-complication-resolution storyline

Full transcript (original language)

00:00 In this video, I’m going to walk you through my exact process for using AI to build corporate slide decks but first let me tell you a quick story In 2025, Deloitte handed the Australian Government a 237-page report and looked exactly like you’d expect a consulting report to look It was clean, structured and properly referenced But as it turns out the report had references to research papers that didn’t exist and a quote from a federal court judgement that was never said and as you can imagine it blew up in the media Now Deloitte says that the report’s recommendations wouldn’t change at all

00:32 even if those issues were found earlier. But that’s not really the point. The point is that none of those errors looked like errors. They were formatted perfectly, they were attributed to real people and they were sitting inside a very professional looking document, and that’s exactly why no one caught them. But I think the same thing happens every day in offices everywhere when people ask AI to build presentations for them. The deck looks finished, but it’s not. So in this video, I’m going to explain why using AI to build a deck for you just doesn’t work and then I’ll walk you through my five step process

01:04 that I use instead where AI does the parts it’s good at and you do the parts it isn’t. Let’s start with what a lot of people do You need to put together a slide deck for a project so you open your AI tool of choice, you paste in all your data and then you ask it to draft a deck You wait a couple of minutes and you get something that often looks pretty impressive Has a decent layout, got some charts, action titles, and it looks kinda like something a consulting firm will produce. But when you really look closely, it’s not quite there. Firstly, AI never has enough information to build a full deck.

01:35 A real project deck draws on months of context and data, conversations, previous decisions, what’s already been tried and so on, and you just can’t paste all that stuff into a prompt. So the AI fills the gaps with things that sound right, but are not right. Secondly, these tools are optimised for looking good, but not for thinking well. So they’ve gotten pretty good at producing something that looks like a McKinsey deck, but the value of a McKinsey deck was never the formatting. It’s the thinking underneath, it’s the analysis, it’s the logic, and it’s the judgement about what actually matters, and that’s the part that’s missing

02:05 from AI Thirdly, it can’t tailor the deck to your audience Now you can say my audience is the board, but that’s not what matters What matters is your CFO hates surprises or a board member blocked a similar proposal last year or your executive stakeholder needs a solution that delivers benefits in the next six months And in my opinion, one of the biggest issues is that reviewing an AI generated deck for hidden errors is actually harder than just building it the right way yourself in the first place Because the polish gives us a false sense of quality For decades, a well formatted

02:36 deck was a reasonable sign that someone had done the work The formatting took effort and the people who put effort into formatting usually put effort into the content as well but AI has kind of broken that link So now you get the formatting, but no thinking behind it, and our brains still read polish as a sign of quality These tools are getting better, I’ll grant you that, but even the perfect AI doesn’t know what it wasn’t told So what do you do instead? Well, my view is that you should use the same process that you would use to build a presentation without AI when you use AI.

03:07 So what changes is that you use AI deliberately and narrowly in the places where it’s good. So my process has five steps. Analyze the data, craft the storyline, build the skeleton, build the slides and then review. So let’s look at how AI fits into each step. Step one is analyzing the data, and I think this is where AI is actually quite strong. It can summarize the data, spot trends, and can draft a first view of what the data means. And this is also where the big firms apparently are seeing real gains. McKinsey says that AI is saving consultants around 30% of the time they

03:37 would spend on research and synthesis. But one caveat though, AI is great when you have one clean, well labeled dataset and it gets much less reliable when you’re trying to connect messy incomplete sources together So that’s where you need to take over And as always, you need to check AI’s work Because even the strongest models still hallucinate So it’s your responsibility to make sure that the analysis is correct Because everything here, all the subsequent steps are reliant on having accurate analysis Once you’ve done that, and you understand what

04:08 the data is saying, you can start crafting the storyline Step two is crafting the storyline The principle here is simple you decide what the data means, and AI helps you turn it into a story. So you should take a first pass at synthesising all the analysis you’ve run to figure out what it actually means for your project. And once you’ve made that call, AI is really good at turning it into a structured storyline. So give it your interpretation and say, AI, give me a situation complication resolution storyline. That actually saves a lot of time.

04:38 But as I talked about earlier, AI is not gonna know the nuances of your client, your project, or your stakeholders, so you need to layer in those things. But AI can help with this too. So for example, you could say, how do I make this storyline more optimistic? It’ll give you a new pass. Or you could say, what in this storyline would a skeptical CFO push back on? And you can make a change. So it’s really good at stress testing your storyline, but you make the final call. Now step three is building the skeleton, and this is where you turn each point in your storyline into action title and then decide

05:10 how you visualize the data that supports it. AI is pretty good at this too, in three different ways. So firstly, it’s good at writing clear action titles. So pull a point from the storyline and you can get a really good action title from it by asking AI. Secondly, you can use AI to suggest what data you need to support each action title. So you can ask it to review your analysis and suggest which data points you should put in the slide. Now this speeds up the slide building process, but it also is really good for spotting gaps in your evidence before you’ve built anything. And thirdly, it can suggest which chart type of visualization would best

05:42 communicate your insight. Now it usually gives a good starting point even if you end up tweaking it. Then once you’ve put together a full skeleton for the entire deck, do a quick test. Read just the action titles one after another. If they tell the complete story without the slide bodies, your skeleton works. If they don’t, fix it now because it’s much easier to fix right now than after you’ve built the slides. Now step four is building the slides and right now I just don’t recommend using AI tools to build slides. Now my principle here is this, wherever you can reuse slides you’ve already

06:13 built. One, they’re done. Two, they’re on brand. And three, they’ve already been through review at least once, so you know that they work. Now the big firms work this way too. Apparently BCG’s deck tool works using hundreds of internal templates in past presentations. The hard part is finding the right slide. Which project was it? Which deliverable was it? Which version was it? And you have to find it and dig it out and put it into your presentation. But that’s one of the things I’ve solved with my own PowerPoint add-in called SlideBrain. You can save your best slides, then search for them in natural language so they’re found instantly.

06:45 Plus it has a whole bunch of AI skills to review slides, interrogate your storyline, review the accuracy of your data, and restyle everything to match your brand. And once you have the slide drafted with appropriate data and some rough insights, you can also use SlideBrain to polish it. It’ll allow you to restyle elements to be on brand with one click, and you can rewrite text for executive audiences without changing the meaning.

08:00 So if you’re interested, can check that out with the link below. Now the final step, step five is review. Now if you think back to the Deloitte report, this is the step that would have caught it, and it’s also where AI really earns its keep. So BCG, apparently the most popular feature of their deck tool is one called review this, which tells you something. And There are a few different ways that you can use AI to review your presentation Firstly, you can ask AI to review the storyline to check that the slide titles read as a story, it builds towards the recommendation and there is a logical flow between different parts of the story Secondly, you can

08:33 ask AI to review the current slide to check that the message is clear, the action title is sharp and the information in the body supports the action title Thirdly, you can ask AI to audit the numbers across the whole deck to identify anything that disagrees across slides, slide text that contradict the charts and math that doesn’t add up SlideBrain has all of this built in and I spent a lot of time testing it against real consulting decks so it’s worth taking a look But of course, you can write your own skills and tools to do it yourself, but my one recommendation is you do a lot of testing because it’s quite hard to get right.

09:05 If you wanna try SlideBrain, you can head to slidebrain.co. And if you want to go deeper on using AI in your presentations, I have a couple of dedicated videos that should appear on the screen now.