Presentation Outline and Speaker Notes for ChatGPT
Turn a topic and an audience into a slide-by-slide outline with one message per slide, suggested visuals, speaker notes timed to your slot, and the three questions you should prepare for.
Fill in the variables
Prompt
<role>
You are a presentation coach and speechwriter who has prepared talks for conferences, board meetings and sales pitches. You build decks around one idea per slide, cut everything that does not serve the audience's decision, and write speaker notes a nervous presenter can actually say.
</role>
<context>
Topic and the one message the audience should leave with: {{message}}
Audience (who they are, what they already know, what they care about): {{audience}}
Setting and time slot: {{setting}} — {{minutes}} minutes including questions
What I want the audience to do afterwards: {{desired_action}}
Material I have (data, results, stories, demos): {{material}}
Constraints (template, slide count limit, things to avoid): {{constraints}}
</context>
<constraints>
- One message per slide, written as a full sentence that would make sense alone (the slide title is the takeaway, not a topic label like "Results").
- Slide count fits the time: about one content slide per 1.5–2 minutes of talking, leaving 20% of the slot for questions.
- Each slide: title (the takeaway), what is on the slide (max 3 bullets or one visual), suggested visual, speaker notes of 60–120 spoken words.
- The opening slide states the message and why it matters to this audience within the first 30 seconds; no agenda slide unless the talk is over 30 minutes.
- Use only the material I gave. Where a slide needs evidence I did not provide, write [need: …] instead of inventing it.
- Speaker notes are spoken language, first person, no jargon the audience does not share.
- The closing slide repeats the message and names the action.
</constraints>
<goals>
- A deck an audience can follow from the titles alone.
- Notes I can rehearse from and finish on time.
- Readiness for the questions most likely to come.
</goals>
<instructions>
1. Confirm the one message in a single sentence and the time budget (content minutes vs question minutes).
2. Write the slide-by-slide outline in order.
3. Add "Likely questions": the three hardest questions this audience will ask, with a two-sentence answer for each.
4. Add "Cut list": anything in my material you deliberately left out and why.
</instructions>
<output_format>
Numbered slides, each with Title, On the slide, Visual, Notes. Then Likely questions and Cut list. Plain text.
</output_format>
Example input
Message: Our support team should own help-center content, because it cut tickets by 20% in the quarter we tried it and it costs nothing extra. Audience: our leadership team (CEO, COO, head of product, head of marketing); they know support is overloaded and that marketing currently owns the help center but has no time for it. Setting: monthly leadership meeting — 15 minutes including questions Desired action: approve moving help-center ownership to support, with 20% of one senior agent's time allocated. Material: Q3 pilot: support rewrote 60 articles; ticket deflection rose from 11% to 31% (tickets per 100 active users fell from 8.2 to 6.6); CSAT unchanged at 94%; marketing published 4 articles in the same quarter; one agent spent about 6 hours a week on it; customer quote: "I found the answer before I finished typing the ticket." Constraints: company template, max 7 slides, avoid blaming marketing
Example output generated with Claude Sonnet 4.6 from the example input above
Message: Support should own the help center because the Q3 pilot cut tickets by 20% at no extra cost.
Time budget: 15 minutes = 12 minutes of content (6 content slides at about 2 minutes) + 3 minutes for questions.
1. Title: Support owning the help center cut tickets by 20% last quarter, at no extra cost.
On the slide: the one line above; "Q3 pilot, 60 articles, one agent, 6 hours a week".
Visual: a single large number, "−20% tickets", with the quarter under it.
Notes: I want to ask for one decision today, and I will get to it on the last slide. In Q3 we ran a small experiment: our support team took over writing help-center articles for one quarter. Tickets per hundred active users dropped by twenty percent, and it cost us about six hours a week of one senior agent's time. The next five slides show what happened and what I am asking for.
2. Title: Customers were asking us questions the help center should have answered.
On the slide: deflection rate before the pilot, 11%; four articles published in the previous quarter.
Visual: a short bar showing 11% deflection.
Notes: Before the pilot, only eleven percent of people who opened the help center found their answer there and did not file a ticket. The team that owned the content had other priorities, and that is understandable; four articles went out that quarter. Support, meanwhile, was answering the same questions by hand.
3. Title: Support rewrote 60 articles in one quarter, from the questions they already answered every day.
On the slide: 60 articles; sourced from the top ticket categories; one senior agent, ~6 hours a week.
Visual: a simple before/after of one article's title, old and new.
Notes: We did not hire anyone. One senior agent spent roughly six hours a week turning the questions we already answered in tickets into articles. Sixty articles in a quarter, because the raw material was sitting in our inbox.
4. Title: Deflection nearly tripled and tickets per user fell by a fifth.
On the slide: deflection 11% → 31%; tickets per 100 active users 8.2 → 6.6.
Visual: two bars for deflection, two for ticket rate, labelled with the numbers.
Notes: The result: thirty-one percent of help-center visitors now found their answer without a ticket, up from eleven. Tickets per hundred active users went from eight point two to six point six. That is the twenty percent. [need: confirm active-user base stayed comparable quarter over quarter, so the ratio is fair]
5. Title: Quality held: CSAT stayed at 94% while the team spent less time on repeat questions.
On the slide: CSAT 94% before and after; customer quote: "I found the answer before I finished typing the ticket."
Visual: the quote, large.
Notes: One worry with deflection is that you are just hiding tickets. Satisfaction stayed at ninety-four percent, and this quote came in unprompted. Fewer repeat questions also meant more time for the complex tickets that actually need a person.
6. Title: Decision: move help-center ownership to support, with 20% of one senior agent's time.
On the slide: the ask in one line; what marketing keeps (brand voice review, product launch pages); start date next sprint.
Visual: a two-column "support owns / marketing keeps" box.
Notes: So the ask: make support the owner of help-center content, formally, and protect twenty percent of one senior agent's time for it. Marketing keeps the brand voice review and launch pages; this frees their time rather than adding to it. If we start next sprint, we can report the first full quarter in January. I would like a decision today.
Likely questions
- "Why not just give marketing more time instead?" Because the content comes from tickets; the people answering them can write the articles in a fraction of the time, and the pilot showed it. Marketing's time is better spent on launches.
- "Is 20% of an agent sustainable when ticket volume spikes?" The 20% is the floor; during spikes the agent returns to the queue and articles pause, which is exactly what happened in week 9 of the pilot. [need: confirm the week-9 detail or remove]
- "Did deflection rise for other reasons, like a quieter quarter?" Tickets per active user, not raw tickets, is the metric, which controls for growth. If leadership wants, we can compare against Q3 last year.
Cut list
- The number of articles marketing published (4) is on slide 2 as context only, not as a comparison, per the "avoid blaming" constraint.
- Individual article view counts: interesting but not needed for the decision.
- The agent's name: keep it a team result in a leadership meeting.
Why this works & how to customize
Why it works: the rule that every slide title is a full-sentence takeaway produces a deck that can be understood from the titles alone, which is how most executives read decks. Sizing the slide count to the time slot, with questions budgeted, is what keeps you from running over. The [need: …] placeholders stop the model from inventing data and give you a prep checklist.
How to customize: for a conference talk, set the audience to "mixed, mostly unfamiliar with the topic" and the opening slide becomes a story rather than a number. For Gamma or other AI slide tools, paste the outline directly; the "On the slide" lines are what they render. If you have a strict template, say how many bullets a slide allows. For longer talks, allow an agenda slide by raising the time above 30 minutes.
How to customize: for a conference talk, set the audience to "mixed, mostly unfamiliar with the topic" and the opening slide becomes a story rather than a number. For Gamma or other AI slide tools, paste the outline directly; the "On the slide" lines are what they render. If you have a strict template, say how many bullets a slide allows. For longer talks, allow an agenda slide by raising the time above 30 minutes.
Comments (0)
Sign in to comment.
No comments yet.