27 May 2025

How I Create Expert-Level Presentations Using AI in 20 Minutes (2 Prompts)

1Frameworks

7Insights

Frameworks in this episode

Communication8 steps

Expert-Framework Presentation Prompt Chain

Turn a presentation goal into an expert-informed deck with two prompts

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 1

Hot Take19:00

Presentation AI Is Fast, Impressive, and Still Flaky

The live demonstration produces polished visuals but also exposes practical defects, including typos, excessive animation, failed edits, and tool errors. The speed advantage is substantial, but reliable delivery still requires inspection, retries, and manual correction.

Expect occasional rendering and editing failures
Inspect slides for typos and unwanted effects
Specify static slides when animations add no value
Preserve time for correction despite rapid generation

“Now, the problem, we do have limitations, right, with AI. It is kind of still flaky.”
Host · 19:00

“And we have an error. We cannot get this working.”
Host · 19:00

#ai limitations#genspark#quality control#reliability

Explainer· 1

Explainer18:30

Why Tiny GenSpark Slide Edits Can Take So Long

GenSpark slides are implemented as web pages, so asking the system to change one small element may cause it to rebuild the entire slide. Direct selection editing is faster when available, but some rollover or animation effects still require full regeneration.

Each generated slide is effectively a web page
Small requested changes may trigger full reconstruction
Use direct element selection for faster edits
Complex visual effects may not support targeted editing

“what it's really doing is rebuilding an entire web page.”
Host · 18:30

“And that slide is actually just code. It's a it's a web page.”
Host · 18:30

#genspark#slide editing#web pages#limitations

Tool· 2

Tool09:30

Use Deep Research to Build the Deck's Factual Foundation

The host demonstrates using a deep-research report to gather the substance required for a product presentation. The report covers the ideal customer, customer problem, value proposition, differentiation, category context, proof, messaging architecture, and narrative spine before any slides are rendered.

Research the product before generating slides
Cover positioning, differentiation, proof, and narrative
Use internal positioning assets when available
Treat the research report as source material for the outline

“You obviously need some content. I'm going to give it a deep research project I did on ramp.”
Host · 09:30

“So basically it's going to create all of that for for me from a deep research project because I'm going to use that to create…”
Host · 10:00

#deep research#positioning#source material#product marketing

Tool10:30

Aggregate Executive Updates Across Internal Systems with AI

An enterprise AI assistant can search connected calendars, email, cloud documents, and project-management systems for presentation evidence. The host proposes organizing the results into milestones, active projects, performance, open questions, and follow-ups while separately flagging inconsistent updates.

Search multiple authorized internal systems
Constrain collection to a relevant date range
Retain source, date, and participant details
Flag conflicting updates for human review
Convert the consolidated evidence into an executive presentation

“I can quickly get the information I need here for certain updates and actually craft that into a presentation as well.”
Host · 12:00

“Cloud basically can integrate with any tools now through MCP.”
Host · 22:00

#claude#mcp#internal data#executive updates#research

Takeaway· 3

Takeaway13:00

Why AI Presentation Drafts Still Need Human Editing

The host pushes back against expecting one prompt to produce a perfect industry-specific deck. AI should create a contextual starting point, while the marketer reshapes the narrative, language, evidence, and slide choices for the real audience.

Expect a strong draft rather than perfection
Adapt the output to the specific industry
Edit the outline before rendering slides
Build on AI output instead of copying it unchanged

“Yes of course because you're not a lazy marketer and just copy and pasting.”
Host · 13:30

“You are using AI as a partner and then you are building on top of it.”
Host · 13:30

#human in the loop#editing#ai workflow#presentations

Takeaway16:30

Validate Every Statistic and Customer Claim Before Presenting

The generated Ramp deck included statistics, customer stories, and claims pulled from research and case studies. Although the model may select legitimate material, the host emphasizes that presenters must verify the evidence rather than relying on a polished slide as proof of accuracy.

Check statistics against original sources
Verify customer stories and case studies
Replace placeholders with real company data
Do not confuse visual polish with factual reliability

“You have to validate that it's picking the right data.”
Host · 16:30

“But you want to just check here that it's true.”
Host · 16:30

#fact checking#data validation#case studies#ai risk

Takeaway18:00

Two Prompts Can Produce an 89%-Complete Presentation Draft

After reviewing the generated Ramp presentation, the host estimates that a knowledgeable presenter could reach a nearly finished deck from two prompts. He suggests that an hour of editing and refinement could turn the generated material into a strong collaborative first draft.

Use generation to compress first-draft time
Budget time for factual and visual edits
Share the draft early for team feedback
Measure value by speed to a useful draft, not zero-touch completion

“you are already 89% of the way there from two prompts.”
Host · 18:00

“I suspect it would take you an hour to do something phenomenal.”
Host · 21:30

#speed#first draft#productivity#presentations