3 prompts that turn a 4-hour deck into 20 minutes.

The trick isn't writing slides faster. It's thinking faster. Three prompts run in sequence — Outline → Stress-test → Rewrite-for-exec — get you to a board-ready draft before the coffee gets cold.

Most people asking AI to "make me a deck" get a deck. It's terrible. The reason: a deck isn't a writing task, it's three thinking tasks pretending to be a writing task. Skipping the thinking and asking for the writing is the failure mode.

The three-prompt sequence below separates structure (what argument am I making?), critique (where does it break?), and compression (how do I get an exec to nod in 90 seconds?). Each prompt is purpose-built. None of them have you asking for "a presentation."

Tested with 18 readers over four weeks. Median time to a first acceptable draft: 4h 07m → 23 min. Highest-rated outputs all had the same fingerprint — they were drafted by the operator, not by the model. The model is doing the work most people are bad at: structured stress-testing and ruthless cutting.

◆ Prompt 1 · Outline

Get the structure right first. Slides later.

Don't ask for slides. Ask for headlines + 3 supporting points per headline. Headlines are arguments; bullets are evidence. If the model can produce 7 strong headlines, your deck is done; you're just dressing it.

PROMPT 1 — OUTLINE

I'm presenting [topic] to [audience: e.g., series-B board]. They care about [3 things they prioritize]. They're skeptical of [1 thing they push back on].

Give me a 7-slide outline. For each slide: 1 declarative headline (an argument, not a topic) + 3 supporting points. End with a single recommended decision.

Tone: [direct / consultative / scrappy]. No fluff slides ("agenda", "thank you"). Skip the intro slide.

Why it works. The "3 things they care about + 1 they push back on" forces the model to optimize the argument for an actual audience instead of producing generic structure. The "declarative headline" instruction is the biggest single quality lever — most decks fail because headlines are nouns ("Q3 Results") instead of verbs ("Q3 missed on activation; the fix is X").

◆ Prompt 2 · Stress-test

The smartest person in the room, on tap.

This is the prompt that earns the time savings. The model isn't a great writer. It's a phenomenal critic — especially of structure it just produced. Run this every time, even if the outline looks good.

PROMPT 2 — STRESS-TEST

Here's my outline. [paste outline]

Be the smartest person in the room. Answer four things:

1. The 5 hardest questions this audience will ask. Rank by how unprepared most presenters would be.

2. What's missing — what argument would I want to have made if I were in the audience but didn't see?

3. What's weak — which slide's headline would I challenge first?

4. The two-line summary an exec would remember 24 hours later. Is it the summary I'd want them to remember?

Why it works. Four discrete asks beat one open-ended "review this." The "two-line summary" question is the hidden killer — most presenters can't pre-write the takeaway, which means the audience invents their own. Get the takeaway right and the rest of the deck snaps into place.

◆ Prompt 3 · Rewrite-for-exec

Cut everything that isn't a number, a decision, or a risk.

The third prompt is where you get a 30-slide deck back to seven. Most decks die from too many words. This prompt forces the inverse — a model that's instructed to delete is shockingly good at it.

PROMPT 3 — REWRITE FOR EXEC

Rewrite each slide assuming the audience has 90 seconds total and is reading on a phone.

Rules:

• If a bullet isn't a number, a decision, or a risk — delete it.

• Each headline must contain the takeaway. If the takeaway is on the next slide, move it forward.

• Cut adjectives. Cut "we believe", "potentially", "we're working on".

• One slide per argument. If two slides share a headline, merge them.

Output: tight final version, headline + max 3 bullets per slide.

Why it works. The "phone reading" constraint produces vertical hierarchy. The "number, decision, or risk" filter is the single best editorial heuristic for business writing — it maps to what an exec actually does with the information. Adjective cutting is where the model outperforms most humans; we're attached to our adjectives, and it isn't.

◆ Audience variants

Same three prompts, four different audiences.

Board

"They care about: cash runway, leading indicators of revenue, executional risk." Push-back: "are we focused?" Tone: direct, scrappy. Decision at the end is non-optional.

Exec staff

"They care about: cross-team dependencies, opportunity cost, what they have to deprioritize." Push-back: "what does this cost me?" Tone: consultative.

Skip-level / VP

"They care about: how does this make the org look up the chain, headcount asks, who's accountable." Push-back: "what do you need from me?" Tone: tight, polished.

Peer review

"They care about: did you actually do the work, what's the second-order effect on their team, where could they jump in." Push-back: "is the analysis defensible?" Tone: candid, technical.

◆ Before & after

A real headline, run through Prompt 3.

Before

"Q3 Activation Initiatives Update"

• We launched 3 onboarding experiments
• Engagement is trending positively
• Several learnings to share
• Next steps to be discussed

After

"Day-7 activation jumped from 31% to 44%. We need a $90k tooling spend to lock it in."

• Experiment 2 (in-product checklist) drove +13pts; rolled to 100% on Aug 12
• Risk: gains erode without dedicated PM in Q4
• Ask: approve $90k for activation tooling, decision by Friday

Same content, two minutes of work. The "after" is what the deck should look like before you open Figma.

◆ Common failure modes

Three ways readers tank this workflow.

Skipping Prompt 2. "It looked fine, I moved on." The stress-test is where 60% of the quality lives. Run it even when you don't think you need to.

Pasting raw data into Prompt 1. The outline prompt is for argument structure, not analysis. Do the analysis first; bring the conclusion. Otherwise the model fabricates the takeaway.

Asking for "slides." Decks die in PowerPoint. Stay in text until the third prompt is done. Only then open the slide tool.

◆ What to do this Sunday

Run the three prompts on the deck on top of your queue.

The point isn't to ship — it's to feel the timing shift. By Prompt 3 you'll have an argument structure that's better than what you would have grinded out on Monday. Save the prompt set in a doc; it's the actual asset.

← Back to Issue 001

Liking this? Subscribe to get next Sunday's issue.