How to Use AI to Build Better Presentations (Without Losing Your Own Point)

The blank slide deck is its own particular kind of stuck. You know what you need to present — a project update, a proposal, a strategy brief — but getting from that knowledge to something on screen that actually communicates it clearly is often harder than the work itself. Most people open PowerPoint or Google Slides, stare at the first slide, type a title, and then spend an hour rearranging things that aren't quite right yet.

AI tools can genuinely help with this. Not by making slides for you automatically — the tools that promise that usually produce something generic enough to be useless — but by helping you think through structure, sharpen your language, and work through the parts that slow you down. After using ChatGPT and Claude on a range of presentation tasks over the past year, a clear pattern has emerged: AI is most useful in the scaffolding phase, before you've touched a single slide, and in the editing phase, after you have a draft. In between, you're mostly on your own — and that's appropriate, because the middle is where your actual thinking lives.

Start With the Argument, Not the Slides

The most common presentation mistake is opening the slide software before you've figured out what you're actually trying to say. AI can help you avoid that by acting as a thinking partner before any slides exist.

Here's a practical approach that works well. Open ChatGPT or Claude and describe your presentation in plain language — not what you want the slides to look like, but what you're trying to communicate and who you're communicating it to. Something like this:

What comes back is usually a proposed narrative arc — not slide titles, but the logical sequence of the argument. Something like: establish the cost of the current problem, show why the status quo is more expensive than it appears, present the migration as the lower-risk option, address the objection directly, close with a specific ask. That structure is often sharper than what most people would produce starting from a blank outline, because you've forced the AI to think about the audience's likely resistance rather than just the information you want to share.

You might not use that structure exactly. But having it on screen gives you something to push against — you can say "no, actually the objection needs to come earlier" or "we don't need a slide on the current system, everyone already knows it's bad" — and that reaction is productive. It moves you forward faster than staring at nothing.

Where AI Is Genuinely Useful Slide by Slide

Once you have a structure and you're building slides, there are specific moments where AI saves real time.

Sharpening headline text. The most effective presentation slides have headlines that make an assertion — "Q1 customer retention dropped 12% despite higher spend" — rather than a topic label — "Q1 Retention Results." Most people default to labels because assertions require committing to a point. If you're not sure how to phrase a slide's key takeaway, describe the slide to ChatGPT and ask it to write five possible assertion-style headlines. You won't always use one directly, but seeing the options usually clarifies what you actually want to say.

Simplifying dense text. If you've pasted too much into a slide — which happens to everyone — you can paste the text into Claude and ask it to reduce it to two or three sentences without losing the essential point. Claude is particularly good at this kind of reduction task. It tends to preserve meaning more carefully than ChatGPT when cutting text down, though both are worth trying for your own content.

Writing speaker notes. Speaker notes are the part of a presentation most people skip or fill with the same text that's already on the slide, which defeats the purpose. If you describe a slide's content and what you want to say out loud when it appears, AI can draft speaker notes that bridge the gap between the slide and the spoken explanation. You'll want to edit them into your own voice — AI speaker notes tend toward the formal — but the draft is usually 80% of the way there.

Anticipating questions. Before you present, paste your slide outline into ChatGPT and ask: "What are the three most likely objections or questions this audience will have?" This is the same technique that works well for meeting prep, and it works here for the same reason: AI can simulate a skeptical reader quickly, and that simulation surfaces gaps in your argument that are easier to address before the room than during it.

The Things AI Cannot Do for You

There's a version of this that goes wrong, and it's worth being direct about it. AI cannot decide what your argument is. It can help you structure and phrase an argument you've already formed, but if you ask it to figure out what you should say, you'll get something plausible and generic that doesn't actually reflect your judgment or your knowledge of the specific situation.

Presentations are persuasion, and persuasion requires a point of view. "Here is information about the project" is not a presentation — it's a document. A presentation takes a position: we should do this, not that; this is the problem worth solving; this is why the numbers tell a different story than they appear to. That position has to come from you, because it requires knowing things the AI doesn't — the history of the project, the particular sensitivities of the audience, the political context of the ask, what failed last time.

When people come away from an AI-assisted presentation feeling like it missed the mark, it's usually because they let the AI set the argument rather than the structure. The slide deck looked coherent but didn't land, because the point it was making wasn't quite the right one for that room. That's not an AI failure, exactly — it's a misunderstanding of what AI is actually good for here.

A Practical Workflow That Actually Holds Up

Here's how I'd put this together as an actual process, rather than a list of tips to remember in the abstract.

Before you open your slide software, spend ten minutes with ChatGPT or Claude working out the structure. Describe the presentation, the audience, the ask, and the main objection you expect. Ask for a narrative arc. Push back on what it gives you until the structure feels right. This is the highest-value use of AI in the whole process.

Then build your slides from that structure — your content, your data, your judgment about what each slide needs to say. Don't ask AI to fill the slides in; that's where your actual thinking needs to go.

Once you have a rough draft of the full deck, go back to AI for the editing pass. Sharpen headlines. Tighten text. Write speaker notes. Paste in your summary slide and ask if the conclusion follows logically from what came before. Ask for the likely objections so you can prepare for them.

The whole process takes roughly the same amount of time as building a presentation the old way, at first. But the output is usually more structured and the argument is cleaner — because you've been forced to articulate it out loud to an AI before you got anywhere near a slide. That articulation is the thing most presentations skip, and it's the thing that makes the difference between a deck that lands and one that leaves people nodding politely and unsure what they were supposed to do with it.

The goal isn't a faster slide deck. It's a presentation that actually makes your point. AI is a reasonable tool for getting there — as long as you're still the one deciding what the point is.