Where we left the deck
Session 1 gave you one number and one lever. The number was 27 percent, the share of a room that can state your ask thirty seconds after a normal corporate slide goes up. The lever was the sentence headline, worth 21 points on its own and the biggest single move on all six slides in the golden set.
★ What your deck gains tonight. A one-sentence thesis you can say out loud without reading it, three claims that carry it, and an outline that is roughly half the length of what you walked in with. Nothing gets designed tonight. That is the point.
LiveWhy the outline comes before the slides3 min▶
Here is the failure I see most often, and I have shipped it myself. Somebody opens the tool, types a topic, gets twelve handsome slides back in ninety seconds, and spends the next four hours making those twelve slides prettier. Nobody ever asks whether twelve was the right number, or whether slide 7 is arguing against slide 4.
Design work on a broken argument is the most expensive kind of work, because it feels like progress. An outline is cheap to change. A designed slide is not, and a designed deck is emotionally expensive to cut, which is exactly why people stop cutting.
A twenty-two slide vendor readout, beautifully built, presented to a steering group. Slide 3 said the pilot was working. Slide 17, added later by a different author, said the pilot had not yet started. Nobody caught it in the build because nobody ever read the deck as a single argument. The steering group caught it in four minutes.
Self-studyThe forty-minute rule I actually use3 min▶
For any deck under twenty slides, I give the outline forty minutes and refuse to open a slide tool until the timer goes. It sounds arbitrary. What it really is, is a commitment device: forty minutes is long enough that you have to confront a weak thesis, and short enough that you cannot hide in it.
The test for whether the forty minutes worked: can you send the outline, as plain text, to one colleague and have them tell you what you are asking for? If yes, the deck is now a formatting job. If no, more slides will not help.
A deck is an argument, not a document
6 min liveDocuments are complete. Arguments are persuasive. Those are different jobs, and most corporate decks fail because they were built as documents and then presented as arguments. An argument has a shape: one thing you claim, a small number of reasons it is true, and one thing you want as a result.
LiveThe one-sentence thesis, and how to know it is real5 min▶
A thesis is one sentence that a reasonable person could disagree with. That last part is the whole test. "Q3 performance was mixed" is not a thesis, because nobody can argue with it. "We should stop selling into EMEA mid-market until we fix onboarding" is a thesis, because half the room wants to fight it.
Three quick diagnostics, in order of how often they catch me:
- The disagreement test. Can someone in the room hold the opposite view honestly? If not, you have written a summary.
- The "and" test. If your sentence needs an "and" joining two unrelated halves, you have two decks. Pick the one this meeting is for.
- The no-notes test. Say it out loud, looking at nothing. If you have to read it, the room will not remember it.
A long thesis is usually two theses. Cut at the "and" and see which half the meeting is actually for. The other half becomes session 4's problem or somebody else's deck.
The outline gate, and why the good tools keep it
7 min liveHere is a thing you can check yourself in ten minutes, and it changed how I judge these tools. Every AI deck tool that produces decks worth editing puts an editable outline in front of you before it designs anything. That gate is not a limitation of the technology. It is the product decision that separates a useful tool from a slide-shaped random number generator.
LiveThree tools, three versions of the same gate6 min▶
These rows are verified against the vendors' own documentation, and the catches are theirs rather than mine. Re-check them before you present this: Copilot and Gemini surfaces move monthly.
| Tool | Where the outline gate sits | The catch that bites |
|---|---|---|
| Beautiful.ai | Outline-first "Create with AI". You get a low-fidelity, text-first outline and you edit it before anything is designed. Per-slide regenerate preserves your copy. | Paid tiers only. Pro is 14.50 dollars a month billed annually, Team is 40 dollars per user per month annual. |
| Google Slides + Gemini | Generates a full editable deck from a prompt or from your Drive files, and shows you an editable outline to approve first. Can match an existing deck's style. | Workspace Experiments, trusted tester only. Desktop only, English only. Google's own page mentions no PPTX export at all. |
| Copilot in PowerPoint | Its best mode is not a prompt, it is a Word document. Your Word file with real heading Styles becomes the outline, so the gate is a document you already know how to edit. | Generate-a-slide and create-from-topic need a Designer licence, and nothing tells you why they are missing. 2,000 character prompt ceiling. It defaults to the first layout in the Slide Master unless you guide it. |
| Claude for PowerPoint | Not an outline gate. It builds on your existing template and makes pinpoint single-slide edits, so your deck is the outline. | Pro, Max, Team or Enterprise only. Not on 2016 or 2019 perpetual, not on iPad or Android. Documented prompt-injection risk, so only use it with files you trust. |
The Copilot move worth stealing even if you never touch Copilot. Write the outline in Word, using real Styles rather than manually bolded text. Heading 1 for each slide's assertion, Heading 2 for its evidence. That file is a better outline than any bullet list, it survives handoff to a human, and if you do have Copilot it is the single input that most improves what comes back. Apply your org template first, before you generate.
Self-studyWhat to actually put in the brief5 min▶
Whatever tool you use, the outline you get back is a function of what you put in. Topic in, generic out. Argument in, editable out.
Two details in there earn their keep. "Outline only, no design yet" holds the gate open even in tools that would rather jump ahead. And "do not fill it in with a plausible number" is the line that saves you in session 4, because a confident invented figure is the failure mode that actually ends careers.
Now cut half of it
5 min liveEvery outline you generate is too long, including the good ones. Not because the tool is verbose, but because you asked for completeness and it gave you completeness. Completeness is a document virtue. In a room with thirty minutes, every extra slide is taxing the slides that matter.
LiveThe kill criteria, and the appendix as a real place4 min▶
One question per slide: does this carry a claim the thesis needs? Not "is it interesting", not "did I do the work", not "will someone notice it is missing". Does the argument need it. Most decks lose a third of their slides to that question on the first pass.
The reason people cannot cut is that deleting feels like throwing away work. So give the work somewhere to go, and take the appendix seriously:
- An appendix is built, not dumped. Same template, real headlines, in order. If it looks like a scrap heap you will not open it in front of anyone.
- It is navigable. One divider slide listing what is back there, so you can jump rather than scroll while six people watch.
- It has an owner in the room. "The regional detail is in the appendix, Priya, I know you will want it" is the sentence that makes an appendix land.
And the honest half of the rule: if you cannot name the person who will ask for a slide, it is not appendix material, it is finished. Delete it. An appendix that holds everything is just the long deck with extra steps.
A board pack went from 34 slides to 11 with a 23-slide appendix. The meeting ran 18 minutes instead of the scheduled 45, and two appendix slides got opened. The other 21 were never missed, and the chair asked whether the next pack could be shorter still.
The cut is easier with someone else's hand on it. Read your outline to a colleague and have them say "needed" or "nice to have" on each line, fast, without discussion. You will argue with three of their calls and agree with the rest, and that is a better result than an hour alone with it.
Thesis, outline, then delete half of it
★ 22 min · everyone buildsOpen the real deck you chose in session 1. We are not touching a single slide tonight. Four moves, twenty-two minutes, and you walk out with an outline that is shorter than what you walked in with and says more.
Thesis. One sentence someone could honestly disagree with. Say it out loud with your eyes off the screen. If you have to read it, it is not finished. Six minutes.
Claims. Three, and only three, each one a reason the thesis is true. Write the evidence you already have next to each. Gaps stay marked as gaps. Five minutes.
Outline. Now let a tool expand it, or expand it yourself in Word with real heading Styles. Outline only, no design. Five minutes.
Cut. Run the kill question down the list. Appendix, delete, keep. Aim to lose half. Six minutes, and it is the six that count.
Watch the longest-headline warning. Anything over twelve words is a paragraph that has not admitted it yet, and session 3 is going to make you rewrite it anyway.
Agenda · Market Context · Q3 Revenue Overview · Regional Breakdown · Key Highlights · Challenges · Opportunities · Next Steps · Thank You
Nine slides, nine filing labels. Every one is defensible and not one of them says anything. The room can follow it and still leave without deciding, because nothing on any slide was ever contested. This is what a topic prompt returns, and it is why people think AI decks are generic.
Q3 missed by 8 percent, EMEA mid-market is the whole gap · Churn tripled after the March onboarding change · The gap costs 1.4 million a quarter · A one-quarter pause is cheaper than fixing it live · Approve the pause today
Five slides, five claims, one ask. Somebody will fight slide 4, which is the point: a room that argues is a room that decides. The other four slides exist to make that argument survivable.
Write the thesis before you open anything. Plain text, one sentence. Then run the disagreement test on it out loud. If nobody in your head objects, it is a summary and you need to be braver.
Name the decision, separately. One sentence: what you want the room to approve, fund, stop or choose. If the thesis and the ask are the same sentence, good. If you cannot write the ask at all, that is tonight's real finding.
Write the three claims, with their evidence beside them. Where you do not have the number, write OPEN QUESTION and move on. Do not let yourself, or a tool, fill that in with something plausible.
Expand to an outline, and hold the gate. Word with real Styles, or a tool's outline mode, or a text file. Say "outline only, no design" explicitly. If your tool jumps to slides, you have learned something about your tool.
Paste it into the estimator above. Read the slide count and the minutes at 90 seconds a slide. Compare that against the slot you actually have. Most people are 40 percent over and had no idea.
Run the kill question down every line. Keep, appendix, delete. Name the person for every appendix line. Anything with no name gets deleted, tonight, while it is still cheap.
Send the plain-text outline to one colleague. One question: "what am I asking for?" If they get it from text alone, session 3 is a formatting job. If not, better to know now than after you have designed twelve slides.
If you are presenting tomorrow. Do the thesis sentence and the cut, and skip everything else. Twelve minutes of writing and six of deleting will do more for that meeting than a night of redesign.
Before session 3
25 min- Finish the cut. Get your outline to a number you can present in your actual slot at 90 seconds a slide. Write down what you started at and what you ended at.
- Build the appendix properly. Same template, real headlines, one divider slide listing what is in there. Put a name next to every appendix item.
- Test the outline on a human. Plain text, one colleague, one question: "what am I asking for?" Write down their exact words. If they are wrong, the thesis is wrong.
- Run the gate on your own tool. Whatever you have a licence for, ask it for an outline only and see whether it obeys. Note what it did. Session 5 is where that note gets used.
- Bring three slides to session 3. Pick the three that survived the cut and matter most. We rebuild those three as assertion plus evidence, live.
- Optional, and it pays for itself: rewrite your outline in Word using real Heading 1 and Heading 2 Styles rather than bold text. It is the input Copilot works best from, and it makes the outline survive being handed to a person.
Three questions
Clear all three and session 2 gets stamped in your passport.
1 · You cannot state your deck's thesis in one sentence. What does that tell you?
A thesis that will not fit in one sentence is usually two theses joined by "and". Cut at the "and" and decide which half this meeting is for. More slides cannot rescue an argument that has not been made yet.
2 · What do Beautiful.ai, Google Slides with Gemini, and Copilot from a Word document have in common?
Beautiful.ai is outline-first with a low-fidelity text outline you edit. Gemini in Slides shows an editable outline to approve. Copilot works best from a Word doc with real Styles, which is your outline. Export and licensing are exactly where all three differ, and that is session 5.
3 · A slide does not carry any claim your thesis needs, and you cannot name one person who would ask for it. What happens to it?
The appendix is a real destination with an owner, not a graveyard. If nobody will ask for it, moving it back there just gives you a longer deck to maintain and one more thing to scroll past while six people watch.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map. Certificates, videos and assessments stay with the official providers.
Session 2 cheat sheet · pin this
You now have half as many slides and a reason for every one of them. Every one of those slides is still built the old way. Session 3 rebuilds them as one claim plus the evidence that proves it, and this is the session where the golden-set average stops crawling: 27 to 73 in forty-five minutes, three of the five levers, in one sitting.