AI Presentation Prompts That Actually Work (With 12 Examples)

By the Kinsy team · Published 27 July 2026

Every AI presentation tool — chatbot or deck generator — sits downstream of the same input: a few sentences typed into a box. Type "make me a deck about X" and all of them hand back the same polite, generic result. This guide is tool-agnostic craft: the anatomy of a prompt that works anywhere, twelve worked examples to copy, and the failure patterns that quietly cost you a good deck.

What should an AI presentation prompt include?

A strong AI presentation prompt states six things: the topic and the point you're making about it, the audience, the rough length, the outcome you want from the room, any constraints, and — where one exists — the source material the deck should be built from. People hunting for AI presentation prompt examples usually expect secret phrasing, and there isn't any: every good prompt in this post is just those six parts in plain sentences. What varies by tool is how much of the formula goes in the text box and how much lives in settings — more on that later.

Why "make me a deck about X" fails everywhere

Because a topic isn't a presentation. "AI in retail" could be an investor teaser, a board risk briefing, or a hundred-slide training course — and a tool given only the topic can't know which, so it produces the average of all of them: ten headings anyone could have written, bullets that are true and useless, an ending that stops rather than concludes. That isn't a product flaw — no generator can recover information you never supplied. The fix costs about sixty seconds of typing.

The six-part anatomy of a strong prompt

Every example below is assembled from these parts.

1. Topic — and the point. Name the claim, not just the subject. "Our hiring process" is a subject; "we're losing senior candidates at the take-home stage, and here's the fix" is a presentation.
2. Audience. Who is in the room, how much they know, and what they're disposed to believe. This line changes vocabulary, depth, ordering and tone more than any other.
3. Length. A slide count or a talk time — whichever your calendar actually constrains. See our guide to how many slides a presentation needs for a given time slot.
4. Outcome. What the room should do after the last slide: approve, invest, book. Decks with a destination end somewhere; decks without one just stop.
5. Constraints. The instructions a colleague would need: "one idea per slide", "misses up front", "nothing I can't defend". Two or three is plenty; ten is micromanagement.
6. Source material. If the deck comes from a report, a paper or rough notes, supply it rather than summarising from memory. Working from notes is its own craft — see our guide to turning notes into a presentation.

Twelve prompt examples that actually work

Copy the one nearest your job and swap in your facts; all twelve work in a chatbot or a deck generator alike.

1. The investor pitch

"A 12-slide seed pitch for our carbon-accounting software for mid-size manufacturers. Audience: generalist VCs skimming their fortieth deck this week. Lead with traction — £40k MRR in nine months — and end with the ask: £1.2m. Confident, not breathless."

The audience line forces the narrative to earn attention early, and the stated ask gives the deck somewhere to land.

2. The sales deck

"An 8-slide sales deck for our payroll service, aimed at finance managers at UK firms of 50–200 staff. Order: their problem, the cost of doing nothing, how it works, proof, pricing, next step. The next step is a 20-minute demo — make it unmissable."

Stating the outcome lets the deck be planned backwards from the booking; the fixed order stops feature-list drift.

3. The quarterly business review

"A 10-slide Q2 review for our exec team, built from the attached report. Revenue against plan, pipeline health, the two accounts we lost and why, priorities for Q3. Candid — the misses go up front, not in an appendix. Every figure must match the report."

"Candid — misses up front" is tone and structure in five words; pinning figures to a source is the best defence against invented numbers.

4. The board update

"A 6-slide quarterly board update for a seed-stage fintech. Non-executive directors, thirty minutes on the agenda. One slide each: headline metrics, cash and runway, product, hiring, risks, asks. No filler — they read fast and interrupt."

The slide-per-topic constraint matches how boards consume decks, and "they read fast and interrupt" makes every slide self-sufficient.

5. The university lecture

"A 45-minute undergraduate lecture introducing photosynthesis, for first-year biology students with no prior knowledge. Build up from everyday examples, define every term at first use, and give me two natural points to stop and ask the room a question."

A talk time hands the tool the pacing problem, and "no prior knowledge" sets the floor expert prompts forget to set.

6. The conference talk

"A 20-minute conference talk on what we learned from running 200 A/B tests on onboarding, for senior product managers. Be honest about the failures, one idea per slide, end with five rules they can apply on Monday."

"One idea per slide" is a constraint any generator can obey, and the takeaways ending gives the narrative a destination.

7. The training session

"A 90-minute onboarding training deck teaching new support agents our refund policy. They'll refer back to it for months, so make each slide self-contained, include three worked edge-case examples, and end with a ten-question quiz."

Naming the deck's afterlife — a reference document, not a one-off performance — changes every density decision the tool makes.

8. The project kickoff

"A 9-slide kickoff deck for our website migration project. Mixed room: engineers, marketing, one sceptical director. Cover the why, scope, timeline, owners, risks — and a non-scope slide saying plainly what we are not doing. That slide matters most."

Flagging the highest-stakes slide directs effort where the meeting will be won; naming the sceptic shapes how the "why" gets argued.

9. The internal launch briefing

"A 7-slide internal briefing on our new mobile app for the whole company at Friday's all-hands. Assume no context. What it is, who it's for, what changed since beta, the launch date, and how each team helps. Upbeat but no hype."

"Assume no context" resets the baseline internal decks chronically overestimate — the launch team has lived this for months; the room hasn't.

10. The webinar

"A 30-minute webinar deck on GDPR basics for owners of small e-commerce shops. They're not lawyers: plain English, real examples of fines, one checklist slide worth screenshotting, and a soft close inviting them to book a consultation."

Defining the audience by what they aren't sets the register, and the screenshot slide matches how webinar audiences behave.

11. The research talk

"A 15-minute talk presenting my paper on microplastics in freshwater lakes, for an academic audience outside my subfield. Method, findings, limitations — in that order. Nothing goes on a slide I can't defend; keep every claim to the paper, which is attached."

The stated order mirrors how academic audiences expect to be argued at, and the defensibility constraint stops the tool embellishing your results.

12. The client proposal

"A 12-slide proposal for redesigning a regional airline's booking flow, for their head of digital and their CFO. Their problem in their words first, then our approach, timeline, team, two case studies, the £85k price, next steps. Formal, no agency jargon."

Opening with the client's problem in their words is consulting's oldest trick, and stating the price stops the deck dancing around it.

Common failure patterns

Four mistakes account for most disappointing results, whatever the tool. Topic-only prompts are the classic. Over-specified prompts are the opposite failure: dictating all twelve slide titles does the narrative planning yourself, badly, and leaves the tool filling boxes. Two audiences in one prompt ("for the board, but also usable by sales") is two commissions — brief them separately. And vibe words without substance — "engaging", "professional", "modern" — give a generator nothing to act on; every constraint should be checkable by a stranger.

Chatbots vs deck generators: same craft, different delivery

The six parts are universal, but where you put them isn't. In a general chatbot — ChatGPT, Gemini — everything must live in the prompt itself, because the model knows nothing you don't type, and you must ask explicitly for a downloadable file, not a written outline. Chatbots are genuinely good at the thinking layer, and we've written honestly about what a ChatGPT PowerPoint actually looks like. Dedicated deck generators move parts of the formula out of the text box: slide counts become a picker, visual style a theme choice, and the prompt slims down to the parts only you can supply: point, audience, outcome, constraints. Neither approach is wrong; the mistake is writing a topic-only prompt in either and blaming the tool for the average that comes back.

Where purpose-built tools change the prompt

A one-shot prompt has to be complete — you get whatever your sentences deserve. Some purpose-built tools close the gap from the other side: they treat the prompt as a brief, ask clarifying questions before building — what the room should do at the end, which number must not be wrong — and take source documents as attachments rather than pasted fragments. Kinsy, which we build, works this way: a plain-language brief, analyst-grade follow-up questions, attachments in PDF, markdown or plain text anchoring a fact-check pass, speaker notes on every slide, and a native, fully editable .pptx at the end. The craft here still applies, but an imperfect prompt becomes recoverable rather than fatal. We keep a library of worked Kinsy brief examples as the companion to this guide.

Keep reading

• How to make a presentation with AI, end to end
• How to turn a Word document into a presentation
• How to build a board deck that survives the boardroom