How to Use AI for Presentations: A 10-Step Workflow

AI has made one part of presentation-making almost ridiculously easy.

Give an AI presentation tool a topic, a document, or a rough prompt, and within minutes you can have a deck with an outline, written content, visuals, layouts, and a coherent visual style.

The technology has moved fast. Current tools can take everything from a simple topic to PDFs, documents, and existing material, then turn that input into a structured presentation. Many now also let you define the audience, presentation type, and tone before generating the deck.

But there is a problem hiding underneath all that convenience.

Making slides faster is not the same as making a better presentation.

For years, one of the biggest frustrations of creating a presentation was production: formatting slides, finding images, adjusting layouts, rewriting bullet points, and making everything look consistent. AI is rapidly reducing that friction.

Once production becomes cheap, something else becomes more valuable.

Judgment.

What belongs on the slide? What should be removed? What does this audience already know? What do they actually need to understand? Which piece of evidence changes the argument? And what should they remember five minutes after the presentation ends?

The bottleneck is moving from making slides to deciding what matters.

That changes how we should use AI. The best workflow isn't Prompt → Presentation → Done. It is Research → Objective → Narrative → Generate → Critique → Verify → Feedback → Iterate → Finalize.

1. The first draft is the easy part

The temptation is obvious. Open an AI presentation tool and type:

Create a 10-slide presentation about AI in marketing.

A few moments later, you have a deck. It may look surprisingly good. There is a title slide. There are sections. There are charts. There are images. Everything is aligned.

And yet the presentation can still be terrible.

Why? Because the AI successfully answered the wrong question. You asked it to make slides before deciding what the presentation was supposed to accomplish.

This is why modern AI presentation workflows increasingly include steps such as defining the audience and tone, and reviewing the generated outline before the final deck is produced. That outline stage is more important than it looks.

The first deck shouldn't be treated as the final answer. It should be treated as a first hypothesis. The AI is effectively saying: based on the information you gave me, this is the story I think you are trying to tell. Your job is to decide whether it understood you correctly.

2. Start with the argument, not the template

Before asking AI to generate a presentation, answer three questions:

• Who is this for?
• What should they understand, believe, or do?
• Why should they care?

These questions sound simple, but they completely change the quality of the output. Compare these two prompts.

Weak prompt

Create a presentation about AI in marketing.

Strong prompt

Create a 10-slide presentation for marketing directors at mid-sized SaaS companies. The goal is to convince them that AI should first be used to accelerate research and content testing rather than eliminate human review. Assume the audience already understands generative AI. End with three practical actions they can implement this quarter.

The second prompt gives the AI something much more valuable than a topic. It gives it a communication objective.

A vague prompt asks what can you tell me about this? A good brief asks what does this particular audience need from me? That is a much better starting point.

3. Give AI better raw material

AI can generate a presentation from almost nothing. That doesn't mean it should.

If you give a model only a topic, you're asking it to fill the gaps with general knowledge. Instead, give it the material that actually matters: research reports, customer interviews, survey results, internal documents, product information, financial data, meeting notes, case studies, previous presentations, relevant articles, your own analysis.

Now AI isn't inventing a presentation around a subject. It is helping you transform your actual knowledge into a communication format.

This is where tools that accept source documents become useful: current presentation systems can turn articles, reports, PDFs, and other text into presentation structures rather than starting from a blank screen.

A useful rule: don't just give AI instructions. Give it evidence.

4. Use slide titles to test the story

Here's one of the simplest tests you can run on any presentation.

Ignore the design. Ignore the images. Ignore the bullet points. Read only the slide titles. Now ask: does the presentation still make sense?

Consider this sequence:

Introduction · Market Overview · AI Trends · Benefits · Challenges · Case Study · Future · Conclusion

Nothing is technically wrong with it. But there is almost no argument.

Now imagine this:

AI Is Changing How Customers Discover Products · Most Teams Are Still Using It for the Wrong Tasks · Research Is Where AI Can Create the Fastest Advantage · Automation Without Review Creates a New Risk · One Team Cut Research Time Without Removing Human Judgment · The Winning Workflow Combines AI Speed With Human Decisions · Here Is How to Implement It This Quarter

You can already feel the difference. The second sequence tells a story before you have seen a single slide.

If your titles cannot communicate the story by themselves, the story may not be clear enough yet. This is exactly why outline review matters: the outline is not just a technical intermediate step. It is an opportunity to fix the narrative before spending effort on the final design.

5. Generate the first deck, then critique it

Once the audience, objective, evidence, and narrative are clear, let AI do what it does exceptionally well: make the first version quickly.

Generate the deck. Then stop. Don't immediately publish it. Don't assume that a polished deck contains polished thinking. Instead, critique it.

Does every slide have a job? If removing the slide changes nothing, why is it there?

Is the hierarchy obvious? Can the audience understand the main point quickly?

Is the deck saying too much? A slide containing six ideas may technically contain more information, but it often communicates less.

Are the visuals actually helping? A chart should clarify. A diagram should explain. An image should add context or emotion. A decorative visual should not distract from the message.

Does the evidence support the claim? A beautiful chart built on questionable data is still a bad slide.

AI has made it dramatically cheaper to produce a first draft. That means we can afford to become more critical of first drafts, not less.

6. Treat feedback as an input

There's another common mistake with AI: when the output is bad, people simply regenerate it. Again. And again. Eventually they have five versions of the same mediocre presentation.

A better approach is to treat feedback as new information.

Suppose a colleague says I don't understand why this matters to our business. Don't respond with "make this slide better." Instead, give the feedback to the process:

A reviewer said the connection between this finding and our business objective is unclear. Rewrite the narrative so that the business implication is explicit. Identify which slide should change and explain why.

Now feedback has become an input. The workflow becomes Generate → Review → Diagnose → Revise → Review again.

This is much more powerful than simply regenerating. And it changes the role of AI. AI isn't the party responsible for deciding whether the presentation works. It becomes the fastest collaborator you've ever had for producing and revising possible solutions.

7. Verify what AI gets wrong

There is one step that should never disappear from an AI-assisted presentation workflow: verification.

Numbers. Dates. Names. Percentages. Research findings. Market sizes. Quotes. Citations. Claims that could influence a decision. These deserve verification.

The reason is simple. AI can produce language that sounds extremely confident even when the underlying claim needs checking. And the more polished the presentation looks, the easier it is to forget that the content still needs scrutiny.

A practical workflow is: AI proposes → source is checked → claim is confirmed or corrected → presentation is updated.

The goal isn't to remove AI from the process. It's to make AI-assisted communication trustworthy enough to matter.

8. Turn data into visual explanations

Data creates another interesting problem. A spreadsheet is not a story. And a chart is not automatically a story either.

Imagine showing an audience twelve numbers and expecting them to figure out what matters. You've presented information. You haven't necessarily communicated an insight.

Instead, start with the question: what should the audience understand after seeing this data? Then choose the visual that makes that conclusion easiest to see.

Suppose your sales report contains revenue numbers for five regions. The important discovery might be that one region is growing three times faster than the others. The goal of the slide isn't to show all five numbers equally. The goal is to make the important relationship impossible to miss.

This is where human judgment becomes critical. AI can help organize the data. It can generate charts. It can suggest visual structures. But someone still has to decide which relationship is the story.

9. The workflow that actually works

The entire approach reduces to a repeatable system.

1. Research. Collect the information and evidence before generating slides.
2. Define the audience. What do they already know? What do they care about? What decision are they trying to make?
3. Define the objective. Write the desired outcome in one sentence.
4. Build the narrative. Decide what the audience needs to understand and in what order.
5. Generate. Give AI the audience, objective, source material, tone, and structure.
6. Critique. Review the narrative, slide titles, hierarchy, evidence, and unnecessary content.
7. Verify. Check important claims, numbers, sources, and citations.
8. Get feedback. Ask someone: what was the main point? What confused you? What do you remember? What would you remove?
9. Iterate. Use those answers to improve specific weaknesses.
10. Finalize. Only after the thinking is right should you obsess over the final polish.

This workflow turns AI from a slide generator into something much more useful: a system for accelerating the entire presentation process. It applies whether you are building a classroom deck, a nonprofit fundraising pitch, or an investor ESG report.

10. The new presentation skill is judgment

AI has lowered the cost of producing slides. That is fantastic. But it creates an unexpected consequence.

When everyone can produce a polished presentation in minutes, polish stops being a competitive advantage. The scarce skill becomes judgment.

Knowing what to remove. Knowing which evidence matters. Knowing when a chart is misleading. Knowing when the narrative is weak. Knowing what the audience actually needs. Knowing when an AI-generated claim needs verification. And knowing when the presentation is finally finished.

The future of AI-assisted presentations isn't humans versus AI. The strongest workflow is likely to be simpler than that: AI handles the speed, humans handle the judgment.

AI can research. AI can summarize. AI can structure. AI can generate. AI can redesign. AI can iterate.

But someone still needs to look at the finished presentation and ask: if my audience remembers only one thing from this, what should it be?

That isn't a formatting question. It isn't a prompting question. And it isn't a slide-generation question.

It's a judgment question. And as AI makes creating slides easier, judgment may become the most valuable presentation skill of all.