Turn a blank deck into a talk that lands.

What you'll be able to do

• Shape any talk around a clear narrative arc instead of a slide dump
• Name the one message your audience should walk away with
• Turn raw content into headline-plus-bullet slides with speaker notes
• Tighten a rambling deck down to what actually serves the point
• Generate visual and layout ideas to brief a designer or build yourself

Inside the path

A focused set of five-minute lessons. Each one ends with a hands-on exercise, not a quiz you can guess.

1. Find the one message 4 min

   Before any slides, use AI to pin down the single thing your audience must remember.

2. Build the narrative arc 6 min

   Prompt for a structure — problem, stakes, approach, proof, the ask — sized to your time slot.

3. Content into slides 5 min

   Turn notes into headline-plus-bullet slides with speaker notes you can actually deliver.

4. Tighten and cut 5 min

   Use AI to spot the slides and sentences that don't earn their place.

5. Ideas for visuals 5 min

   Generate chart, diagram, and layout ideas to brief a designer or build yourself.

Try a sample exercise

This is the kind of card you'd practice inside Iro: you do the thinking, then get feedback.

Structure beats slides — and it's where AI helps most

Most weak presentations fail before design ever enters the picture: they're a list of everything the presenter knows, in no particular order, with no single point. Opening PowerPoint first makes this worse, because you start decorating slides instead of deciding what the talk is for.

This is exactly where AI is strongest. Before you make a single slide, use it to answer two questions: what is the one thing this audience must remember, and what arc carries them to it? A reliable arc is problem, stakes, approach, proof, and the ask. Feed the model your audience, your goal, and your time limit, and have it propose that structure first. Get the skeleton right and the slides almost build themselves.

From content to slides, notes, and visuals

Once the structure holds, AI is a fast drafting partner for the rest, as long as you keep the judgment.

• Hand off: turning your notes into headline-plus-bullet slides, drafting speaker notes, tightening wordy slides, and brainstorming chart or diagram ideas for each point.
• Keep: the final call on what stays, the actual design and brand, your delivery, and whether any claim on a slide is true.

A good move is to ask for one headline and no more than three bullets per slide: it forces the deck to stay skimmable and stops you from reading paragraphs at your audience. Then ask the model for a couple of visual options per key slide so you can brief a designer or build them yourself. The presentation is still yours; AI just gets you to a strong draft faster.

How do you get from raw notes to a talk you can deliver?

Work in passes, and don't ask for the deck first. Each pass hands the model something it couldn't guess, and hands you a decision to make while the talk is still cheap to change.

• Brief it. Paste the raw material: your notes, the doc it came from, the numbers. Then state the audience, the goal, and the time limit, and tell it explicitly not to write slides yet.
• Ask for three arcs, not one. "Give me three different ways to structure this talk. For each, name the opening move and what it asks the audience to believe." Choosing between three is a much better use of your judgment than accepting the first.
• Expand the arc you picked. Now you can ask for slides: a headline, bullets, and a speaker note each. A slide a minute is a sane ceiling, so a 10-minute talk runs roughly six to ten slides, not twenty.
• Stress-test it. "You're the CFO in this room. Ask the five hardest questions and name my weakest claim." Write your answers down. Most talks come apart in the questions afterwards, not in the deck.
• Cut with a timer running. Read the whole thing out loud against the clock. Anything you stumble over was written to be read, not said, so rewrite it the way you'd actually say it.

One pass belongs to you alone. Check every number and every named fact against its source before it goes on a slide, because a model will produce a confident figure it never actually saw, and a slide makes any figure look like evidence. A wrong number on screen costs you the room for the rest of the talk.

Pitch deck, internal update, conference talk: what changes in the prompt

The three inputs stay constant (audience, goal, time limit), but the arc, the evidence, and the ask shift with the room. Naming the format in your prompt is the cheapest way to get a draft that fits.

Investor pitch

The room is deciding two things: whether this is worth doing, and whether you're the one to do it. Name the arc you want: problem, why now, what you've built, evidence it's working, the raise. Tell the model to keep one claim per slide and push the detail into an appendix. Then ask it to attack your "why now," because that's the question you'll get.

Internal update or decision meeting

Your audience already has the context, so cut the setup. Lead with the recommendation and the ask, then the evidence behind it. A prompt that works: "Rewrite this so the decision I need appears in the first 30 seconds and the supporting detail comes after." Have it draft a one-paragraph spoken version too, because exec meetings run late and you'll sometimes get four minutes instead of twelve.

Conference talk or teaching session

Nobody is approving anything here, so the goal is one idea people can repeat later. Push for a single takeaway sentence, then three concrete moments (a story, a live demo, a before and after) that make it land. Slides carry fewer words in this format, and the speaker notes carry more.

One deck rarely survives a format change. If you're reusing a pitch as an internal update, tell the model which parts are now assumed knowledge and have it rebuild the opening, rather than trimming slides off the front.