A tool can turn a topic into twelve tidy slides in a minute, and nobody in the room knows what you want them to do. Here's the thinking to do first, and a checklist for the review at the end.
Write the decision and the slide titles before any tool builds slides. Name who decides, what they believe now and what they’ll object to. Write each title as a claim, read them alone to check the argument, then have the tool build slides around them and flag what it needs. You own the argument and every number; score the finished deck before it goes out.
Most AI-built decks go wrong before the first slide exists. The prompt names a topic, the tool returns that topic with good formatting, and the audience gets a run of slides titled “Market Overview” and “Key Challenges” that nobody can disagree with or act on. The deck looks finished. Nobody wrote the argument.
A marketing manager preparing a budget request opens Copilot or Gamma, types the subject and the audience, and gets a clean deck back in about a minute. Each slide is reasonable in a general way. None says what she wants the finance director to approve, so the finance director spends the meeting working that out aloud.
The fix is a change of order. Tools are fast at turning words into slides, so the thinking comes first, in three steps, and the tool gets the third one.
The three-step order this guide follows. It is a Future Factors working method based on our own judgment, not a research-validated process.
If you want ChatGPT-specific prompts for outlines, bullets and speaker notes, the ChatGPT presentation walkthrough goes through that end in detail. This one is about the decisions that sit on either side of the tool.
A topic is what you’re talking about. A decision is what you want a named person to do by the end of the meeting. “Our retention programme” is a topic. “Approve a pilot of lifecycle email for one customer segment” is a decision, and it changes everything downstream: what you show first, which numbers you need, what you leave out.
Before any slide, fill in the six-field brief below. The example is a budget request, and the values are specific enough that someone could disagree with them.
| Field | Example: a pilot budget request |
|---|---|
| Who decides | The finance director, with the CMO present. One holds the budget, the other decides if it’s worth arguing for. |
| What they believe now | Email is a cost line that runs itself, not a growth lever. |
| The one decision | Approve a pilot of lifecycle email for one customer segment, with a review date. |
| Their likely objection | We tried more email before and unsubscribes went up. |
| Proof they will accept | Repeat-purchase data split by acquisition and retention, from the finance partner, not marketing. |
| Setting and length | A twenty-minute slot in a leadership meeting, read in advance by the finance director. |
A completed brief. People tend to leave the objection and the proof blank, and those two decide most of what goes on the slides.
Finish the sentence “By the end, I want ___ to decide ___” before you name the first slide.
Not every deck has a decision in it. A status update’s nearest equivalent is “nothing needs your attention” or “I need one thing from you.” A training session’s is what attendees will do differently on Monday. If you can’t find any version of the sentence, a short written note may serve better than a deck.
The decision also differs by role, which is why generic advice about “knowing your audience” doesn’t help much. Each of these calls for a different first slide and a different proof:
You can use AI for this step, as an interviewer and not as an author. If your brief is thin, ask the tool to question you until it isn’t. Copilot in PowerPoint may ask who the audience is and what style you want before it builds an outline, according to Microsoft’s support page.[2] Answer them with your brief. What you shouldn’t hand over is the decision itself, because a tool left to pick the point of the deck picks one that sounds sensible to everyone and commits to nothing.
Once you know the decision, write the argument as slide titles. The test for a good title is that it makes a claim: a full sentence that someone in the audience could agree or disagree with. “Customer retention” labels a slide. “Most of our growth spend goes to winning customers we then have to win again” argues something, and the slide underneath only has to prove it.
There’s some evidence that this habit helps people take things in. In a 2006 conference study, roughly 800 students in a large geoscience course were taught from slides with either sentence headlines or phrase headlines. Students on sentence-headline slides averaged 79% correct on recall questions against 69%.[1] That was a lecture course where students kept the slides afterwards, so I’d treat it as support for the habit, not proof about a leadership meeting.
Here’s what a finished storyline looks like for the budget request above. Read down the second column only and you should be able to follow the whole argument.
| # | Message title | Evidence it needs | Who confirms the number |
|---|---|---|---|
| 1 | We’re asking you to approve a pilot of lifecycle email for one segment. | None. This is the ask, stated up front. | Presenter |
| 2 | Most of our growth spend goes to winning customers we then have to win again. | Spend split between acquisition and retention. | Finance partner |
| 3 | Repeat customers are the cheapest revenue we have. | Repeat-order rate next to the cost of a new customer. | Analyst |
| 4 | Our current emails treat every customer the same. | Three real emails from the current programme. | Presenter |
| 5 | A pilot on one segment tests the idea without the full commitment. | Pilot scope, owner and review date. | Marketing operations lead |
| 6 | The risk is unsubscribes, and we’ll watch it weekly and stop if it moves. | The guardrail metric and the stop rule. | Marketing operations lead |
| 7 | Approve the pilot today, and we’ll bring you the results at the review. | The ask again, with the date. | Presenter |
A seven-slide storyline for a twenty-minute slot. The last two columns are what you gather before building and who confirms each figure.
Read the titles alone, in order. If the argument doesn’t hold, no layout will rescue it.
Here’s how I’d write the titles, in this order:
AI is useful here as a drafter you argue with. The first pass usually gets the order about right and the wording slightly generic, and your rewrite is where it starts to sound like you.
| I’m a [role] presenting to [who decides]. They currently believe [belief]. The one decision I want is [decision]. Their likely objection is [objection], and the proof they’ll accept is [proof]. The slot is [length and setting]. Draft a storyline of [number] slides as message titles. Each title must be one full sentence that makes a single claim, under 15 words. Make the last slide the ask. Place a slide answering the objection before it. For each title, list the evidence it would need and whether I’d have to supply it. Do not invent statistics, customers or quotes. If my brief is too thin to write a title, ask me up to three questions before drafting. Then offer one alternative order and say what it would change. |
Use it after the brief, not instead of it. Edit the titles by hand before moving on, and keep the evidence column, because step three depends on it.
The tool you choose matters less than whether your titles survive the trip into it. Tools that build slides from a prompt tend to rewrite what you give them, which is fine for a rough pass and a problem once the titles are right.
Every number on a slide gets traced by you, to a source you’ve opened, not by the tool that put it there.
Both major tools give you a way to keep your words. In Copilot in PowerPoint, Microsoft’s support page describes a flow where you describe the deck, answer clarifying questions, review a generated outline, and only then ask for slides.[2] For a Word document, Microsoft recommends using Word styles so Copilot can see the structure, and starting from your organisation’s template so its layouts are reused. When you create from a file, you can’t add extra context in the same prompt, so the document has to carry everything. A storyline written as a Word outline with headings is exactly that document.
Gamma’s developer documentation describes three text modes: one that rewrites and expands your input, one that condenses a large amount of text, and one that “will retain the exact text” you provide, sometimes adding structure such as headings.[3] That documentation covers Gamma’s generation interface, so check the options on your own screen. The decision rule travels: when your titles are final, pick the setting that keeps your text. With only rough ideas, pick the one that expands, and bring your own titles back afterwards.
| Your situation | What to do in the tool | What to check afterwards |
|---|---|---|
| Final titles and notes in a Word outline, working in PowerPoint | Create from the document using heading styles, starting from your organisation’s template. | Slide titles match your storyline word for word. |
| Final titles, no document | Paste the storyline into the chat, answer its questions from your brief, and approve the outline before slides. | The outline follows your order, with no added slides. |
| Mostly written text, working in Gamma or similar | Choose the setting that preserves your text and only structures it. | Any headings the tool added. |
| Only rough ideas, exploring | Let the tool expand freely, then treat the result as a sketch. | Replace every title with your own, and remove claims you can’t source. |
A decision table for the build step. Microsoft notes some Copilot options reach Insiders and Frontier users first, so confirm against what your account shows.
| Build a presentation from the storyline below. Use each message title exactly as written as the slide title. One idea per slide. Do not add slides, statistics, quotes or claims that aren’t in my storyline. For each slide, suggest one visual that proves the title (a chart, table, screenshot or short quote) and say what data it needs from me. Where you need something I haven’t given you, write [NEEDS: what] on the slide instead of filling it in. Keep body text to a few short lines. [Paste the storyline table here: title, evidence needed.] |
The [NEEDS] marker is the useful part. It turns the tool’s habit of plausible filler into a visible to-do list for you.
Which tool to use is mostly a question of where your files and templates already live. If it’s Microsoft 365, Copilot in PowerPoint keeps you inside them. The review step below doesn’t change either way.
Where AI does real work on a deck, be exact about who owns what. The third column matters most, because each human responsibility needs a named way of being checked.
| What AI does | What you still own | How it gets checked |
|---|---|---|
| Interviews you to fill gaps in the brief | The audience, the one decision and the objection, which come from knowing the room. | A colleague who knows the audience reads the brief and says whether it matches the meeting. |
| Drafts message titles and alternative orders | The argument: which claims, in what order, and what you leave out. | A titles-only read: someone outside the project says what you want from the audience. |
| Builds slides, suggests layouts and visuals | What each slide must prove, and whether the visual shows it or decorates. | Compare each slide to its title and delete what doesn’t prove it. |
| Writes slide text and speaker notes from your storyline | Every number, quote and claim. A tool can misquote a figure or invent one. | Each figure traced to a source you opened and confirmed by the named owner. |
| Plays a sceptical audience member | Which objections are real, and your answers. | You write your own answers to the three hardest questions. |
| Nothing. This part is yours. | Delivery: pacing, and the judgment to skip a slide when the room moves on. | A timed rehearsal out loud, without reading the slides. |
A human-ownership table for a deck built with AI. The checks are Future Factors working practice, not a standard.
For a team, this works best as a shared habit. Training builds the skill of writing message titles and briefing a tool. Workflow design gives that skill somewhere to be used, such as a one-page brief that comes with every deck request and a standard review step. Reinforcement keeps it going: a manager who asks for the titles first in each deck review. Adoption is what you see when people do it without being asked.
A team can begin on its own with the brief and the titles-only read. A workshop built around your team’s real decks shortens the learning curve, because people practise on their own proposals and templates and leave with one shared standard instead of one enthusiast’s habits.
To check whether the habit is working, the number of decks made with AI is the weakest signal, since it only shows use. Better signals are whether the decision got made at the meeting and, a month later, whether decks need fewer rewrites than before.
Before a deck goes out, score it. A checklist beats a general impression because tired people skim, and a deck built quickly with AI tends to look more finished than it is. Score each check 0 if it fails, 1 if it partly holds and 2 if it fully holds, then add them up.
| # | Check | How to test it |
|---|---|---|
| 1 | One decision | Someone outside the planning can say what you want after slide one. |
| 2 | Titles-only read | The titles alone, in order, make the argument and end on the ask. |
| 3 | Numbers sourced | Every figure is traced to a source you opened yourself, not to the tool that wrote it. |
| 4 | Nothing invented | No quote, customer or statistic you can’t point to. Every [NEEDS] marker is filled or deleted. |
| 5 | Objection answered | The hardest likely question has its own slide, placed before the ask. |
| 6 | One idea per slide | Each slide proves its title and nothing else. Extra material is in an appendix. |
| 7 | Visual proof | Key slides show the claim with a chart, table, screenshot or quote, not decoration. |
| 8 | Your voice | Terms and names are consistent, and filler such as “robust” or “streamlined” is rewritten. |
| 9 | Delivery | You can say each point without reading it, and a timed run fits the slot. |
| 10 | File hygiene | Nothing confidential went into an unapproved tool, and speaker notes are cleaned before sharing. |
Ten checks, 20 points. Bands: 17 to 20, send it. 13 to 16, fix the lowest-scoring checks and send. 9 to 12, go back to the storyline table. Below 9, rebuild from the brief. The bands are Future Factors working guidance, not a validated scale.
Check 3 is where AI-built decks tend to fail, because a plausible number on a clean slide gets treated as true. Our guide to fact-checking ChatGPT walks through a workflow you can reuse on any figure the tool produced. Check 10 has a tool-specific detail: Microsoft says slides with images generated in Copilot carry an “AI-generated” source note in the speaker notes,[2] so decide whether you want that note travelling with the file before you share it.
A second reader helps with the scoring, and AI can be that reader if you point it at the person who will challenge the deck.
| Act as [the finance director] at [type of company]. You currently believe [belief]. I’m going to present the slide titles and speaker notes below, and I want you to decide whether to [decision]. List the three titles you would challenge most and the exact question you’d ask about each. Tell me which titles depend on a number or claim I haven’t evidenced. Tell me what you’d expect to see on the slides that isn’t there. Don’t rewrite my deck and don’t reassure me. If the argument holds, say which title carries it. [Paste titles and notes.] |
Run it in a fresh conversation, and answer the three questions yourself before you look for any rewrite. The questions are the useful output, not the reassurance.
You don’t need a new tool to start. Pick the next deck on your calendar, because a real deadline makes the habit stick better than a practice run, and follow the sequence below. The first two steps are the ones people skip.
Fill in the six-field brief and finish the One-Decision sentence.
Draft titles with the prompt, rewrite them by hand, and fill in evidence and owners.
Send only the titles to a colleague and ask what you want from the audience.
Hand over the storyline with the keep-my-titles prompt and resolve every [NEEDS] marker yourself.
Score the deck, run the sceptical pass in a fresh chat, and rehearse once, timed.
The sequence in order. For a team, make the titles-only read a standing step in deck reviews.
If it goes well, people stop asking what the deck is trying to say and start arguing with the claims, which is the conversation you wanted. Keep this deck’s brief and storyline as a template and change only the values next time.
It can draft one, but only well from your brief, and you decide the argument. Ask for the storyline plus one alternative order, then pick and say why. If you can’t explain why one order beats the other, the brief is still too thin.
One sentence you can say in a single breath. Under about 15 words is a working limit we use, not a research finding. A title that needs two clauses is usually carrying two claims and wants to be two slides.
Name the nearest equivalent. For a status update it’s ‘nothing needs your attention’ or ‘I need one thing from you.’ For training it’s what people do differently afterwards. If neither fits, a short written note may serve better than a deck.
Check your organisation’s policy first and use the tools it has approved. For anything confidential, describe the situation in general terms and add the real figures yourself afterwards. The method allows for that, because the titles carry the argument and the numbers come last.
Yes, as a sketch when you’re still working out what you think. Treat the result as brainstorming material, replace every title with your own once you’ve decided, and don’t send it. A deck generated from a topic alone shows what a typical deck on that topic says, which rarely matches what your audience needs.
The research cited here, a 2006 conference study of sentence headlines in teaching slides, was read through its ASEE paper page on October 6, 2026; the journal version was not readable, so only figures on that page are used. Copilot in PowerPoint details come from Microsoft’s support page for creating a new presentation, and the Gamma text-mode definitions from Gamma’s developer documentation, both read the same day. Menus, plans and rollout stages change often, so confirm the options your account shows. The three-step order, the three rules, the checklist scoring and the ownership table are Future Factors working methods based on our own judgment, not research-validated scales. The scenes and the budget-request example are illustrations, not client stories.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.