Why 90% of AI Presentations Fail (do this instead)
About this video
Jeff Su argues AI presentation tools fail at insight generation, not design: they cannot tell you why a client should enter the US market, but can present that insight once you have it. He recreates a pitch deck from his time as a salesperson at Google using a three-step process for a budget skincare brand targeting young adult males 25 to 35 across Japan, Indonesia and the United Kingdom. Step 1 Research: use ChatGPT Deep Research with the latest reasoning model and a detailed prompt applying Porter’s Five Forces per market, asking for an executive summary, supporting data and a ranked market scorecard; in the demo it ran about 7 minutes and ranked Indonesia first, with tactics like formulating for tropical climates and oily skin, obtaining halal certification, and prioritising Shopee and Lazada. Step 2 Structure: in the same thread, prompt ChatGPT to turn the report into an 8-slide outline following the SCQA-style framework he names, bottom line up front — enter Indonesia, market size 9 billion growing at 8% vs Japan 4% and UK 2%, with one slide on why the other markets are deprioritised. Step 3 Design: paste the outline into Gamma (free plan works), choose condense not generate, detailed text, illustrations, then refine with human expertise, Edit with AI, layout and image changes. He notes Gamma’s spotlight presentation mode is a strength, while export to Google Slides introduces spacing/formatting issues.
Key takeaways
- Separate the workflow: Deep Research for insights, a text model for story structure, Gamma as the visualisation layer only — match the tool to the task.
- Deep Research prompts must be specific: objective, audience, framework (Porter’s Five Forces), deliverable format, and a ranked recommendation/scorecard, not just research.
- Structure before slides: one recommendation up front, three supporting points per slide, and a slide explaining why alternatives were rejected to reduce decision fatigue.
- In Gamma, paste your outline, select condense (not generate) and detailed text, then expect the result to be only about 80% there — the remaining edits, checking narrative flow and that no critical data was lost, are yours to do.
- Quote he closes with from Professor Ethan Mollick at Wharton: value comes from finding the right points of human-AI collaboration for each task, not surrendering or clinging to human-only workflows.
Episode details: published June 17, 2025 · 13:17 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Jeff Su’s own claims, shared for education — they are not typical or guaranteed.
← Previous: My Last Day at Google (after 9 years) · Next: The CORRECT way to use ChatGPT (in 2026) →
All Jeff Su videos · Jeff Su profile · Jeff Su (official site)
