Five Of These Seven Never Mention The Thing You Say Out Loud
Seven prompts for building a presentation, posted in full in the comments: a story arc, an investor pitch, a beginner explainer, a research-backed deck, an executive briefing, a complete deck, and slides with a speaker script.
They are good prompts. Prompt five even contains the line “distinguish verified data from assumptions”, which most decks of this kind skip. But read all seven together and something is missing from five of them, and it is the half the audience actually receives.
The source post — “ChatGPT can turn a simple idea into a professional presentation”, BANI Academy. Seven presentation prompts, each posted as its own comment.
A presentation is two artefacts, not one
There is what is on the screen, and there is what comes out of your mouth. They are made of different material, they do different jobs, and the audience gets both at once. A deck is only half the thing.
Now count. Of the seven prompts, five never mention the spoken half at all. The investor pitch, the beginner explainer, the research-backed deck, the executive presentation and — this is the one — “Build a Complete Presentation” all describe slides and only slides.
The sixth, the story arc, adds it at the end: “Add speaker notes with storytelling cues, pacing, emphasis, and delivery guidance.” Notes, appended.
And the seventh treats the two as a pair, which is why it is the best prompt on the card.
The best sentence here is in prompt seven
“Ensure the spoken explanation complements the slides rather than repeating them.”
That single clause is the difference between a talk people remember and the one everybody has sat through: a person reading their own bullet points aloud while forty adults read ahead and get there first.
It is also the hardest thing on the card to actually get, and the reason is mechanical rather than a question of effort. When one model writes the slides and the script in the same pass from the same source, the script is a natural expansion of the slide. Nothing in the process pulls them apart. The slide says “Three reasons retention collapses at 30 seconds”, and the script says “There are three reasons retention collapses at thirty seconds…” Asking for complementarity is right. Getting it requires changing the order of work.
Diagram: what belongs on the screen and what belongs in your mouth
| The screen | Your mouth | |
|---|---|---|
| Good at | Numbers, shapes, comparisons, anything with a position — a chart, a photograph, a before and after | Causation, emphasis, the aside, the story, why this matters to these people today |
| Bad at | Nuance, sequence of reasoning, anything that needs a tone of voice | Six figures in a row, a spatial relationship, a shape |
| How fast it is read | Instantly, and ahead of you | At the speed you speak, in the order you choose |
| If both carry the same words | The audience finishes reading before you finish the sentence | You become a narrator of a document they already read |
| The test | Could I say this out loud in one breath? Then it does not need a slide | Could somebody get this from the slide alone? Then do not say it |
All seven, and who each one is for
The other useful cut is audience. Six of the seven name one, and the quality of the output tracks it closely.
| Prompt | Who it is for | Does it build the spoken half? | The line worth keeping |
|---|---|---|---|
| Turn a topic into a story | A general audience — a keynote, a talk | Appended — “add speaker notes… pacing, emphasis” | The arc is named: Hook → Problem → Journey → Discovery → Transformation → Call to Action |
| Build an investor pitch deck | Investors | No | Thirteen named sections, and “realistic and evidence-based” |
| Explain any topic simply | Beginners, no prior knowledge | No | “One clear takeaway per slide” and “minimal text” |
| Create a research-backed deck | A sceptical or academic room | No | “Slide-level citations and a well-organized references slide” |
| Create an executive presentation | Senior leaders, a board | No | “Distinguish verified data from assumptions” — and structure it around the decision |
| Build a complete presentation | Nobody in particular | No | “Not just an outline” — an instruction, not an audience |
| Create slides + speaker script | You, in front of a room | Yes, as a pair | “Complements the slides rather than repeating them” |
That blank is worth a moment. “Build a Complete Presentation” is the prompt most people will reach for, because it sounds like the one that does everything. It is also the only one that does not say who is in the room — and a presentation with no audience in mind is the definition of the generic deck the whole card is trying to avoid. The other six beat it precisely because they each narrow to somebody.
And notice the tension between prompt three and prompt six. Three asks for minimal text and one takeaway per slide. Six asks for “polished content… not just an outline”, which pushes words back onto the screen. Both instructions are reasonable; they cannot both govern the same deck. If you only take one, take three’s.
Build it in the other order
Here is the practical change, and it costs nothing. Write the script first. Then ask what the words cannot carry on their own, and put only that on a slide.
That sequence makes complementarity the default instead of something you have to request. A slide earns its place by doing a job speech is bad at — showing six numbers at once, a shape, a before and after, a photograph of the thing. Everything else stays in your mouth, where it belongs, and the deck gets quieter and better.
Concretely, replace prompt six with two passes:
Pass one. “Write what I will say out loud about [TOPIC] to [AUDIENCE], for [N] minutes. Conversational, first person, no bullet points. Mark the moments where I would want to point at something.”
Pass two. “Here is the script. For each marked moment, tell me what belongs on the screen — and only what speech cannot carry. If a slide would restate a sentence I am already saying, say ‘no slide’ instead. Give me the slide count that results.”
The permission to return no slide is the whole trick. It is the same device we keep finding in well-built prompt decks: give the model a legal way to produce less, and it will. Without it, you asked for a deck, so you get a deck — one slide per paragraph, forever.
The underlying point is one we have made about video too: the knowledge in a tutorial is usually in the part nobody said out loud, and the picture and the narration carry different halves of it.
What each deck still needs from you
Three of the seven ask for material the model cannot supply, and it is worth naming which, because the consequence differs each time.
| Prompt | What it asks for | Who has to provide it | What happens if nobody does |
|---|---|---|---|
| Investor pitch | Market opportunity, business model, financial projections, metrics, team, milestones, funding request | You. None of those numbers exist in a model | A plausible deck shown to people whose profession is checking the numbers in it |
| Research-backed deck | “Credible, verifiable sources”, slide-level citations, a references slide | You, or a search tool with the papers in front of it | Citations that look correct and point at nothing — on a slide, in front of a room |
| Executive presentation | Performance metrics, forecasts, SWOT, risks | You. It has none of your performance data | A SWOT of a generic company in your industry |
| The other four | Structure, language, sequencing, analogies, delivery cues | The model, genuinely well | These are the four to use as written |
Prompt five already half-solves its own problem with that excellent line — “distinguish verified data from assumptions”. Add the same clause to prompts two and four and all three become honest: anything the model supplied gets labelled as an assumption until you replace it with a real figure and a real source.
One format note, briefly, because it decides how much time this actually saves. What comes back is text — slide-by-slide content, notes, and recommendations for visuals. Four of the seven prompts ask the model to recommend charts and visuals rather than make them, which is the author being realistic. Turning that into an actual file is a separate step, and what your account offers for it is worth checking before you plan a deadline around it.
Four honest caveats
- We did not run the seven prompts. This is a reading of what each one asks for and what it leaves out. The four that need only a topic may well produce excellent structure; nothing here says otherwise.
- The audience column is our reading. The prompts do not label themselves that way. “Build a complete presentation” could reasonably be called general-purpose rather than audience-less — our argument is that for a presentation those are the same thing.
- Script-first is a working preference, not a law. Some people think better in slides and write the talk from the storyboard. The point is only that the two artefacts should differ, and that whichever you write second should be checked against the first for duplication.
- Export behaviour changes. What a given assistant hands you — text, a canvas, a file — depends on the product and the plan, and it moves. Check yours rather than ours.
The one-line version
Seven prompts, and five of them build only the thing on the wall. The audience receives two performances at once, and the deck is the lesser one. Prompt seven knows it — complement, do not repeat — and that instruction is hard to honour when the same model writes both halves in the same breath.
So reverse the order. Write what you will say, then ask what the words cannot carry, then put only that on a screen, and let the model answer no slide when the answer is no slide. You will finish with fewer slides than you expected, which is the reliable sign that it worked.
If the talk is going to be recorded rather than given live, the same split governs the voice track, and most of what makes narration sound robotic is fixed in the script rather than the voice. And if you teach for a living, the classroom version of this problem is the same one with a whiteboard.
The thirty-second test, on a deck you already have. Open it and read only the slides, out loud, as if you were the audience reading ahead. Every sentence you would have said anyway is a sentence that should not be on the screen. Delete those and see what is left.
One Comment