When creating presentations from notes or source material, I don't recommend first asking ChatGPT or Claude to generate a prompt that you then feed back into the model to create the deck. In my experience, this introduces an unnecessary translation step. The model often omits important details from the original notes and invents slide structures or frameworks that don't translate well into the final presentation.

Think of it like explaining your material to a colleague, who sketches out the deck structure and guesses at what visuals might work, then hands those notes to a second colleague who actually builds the slides. By the time the presentation is finished, you've gone through two rounds of translation, and many of the details that mattered have been lost.

Instead, provide the original notes directly and ask the model to generate a deck outline. Review the outline, make any necessary edits or additions, and then provide both the revised outline and the original notes back to the model to generate the presentation. The outline acts as a high-level guide while the original notes preserve the context and nuance needed to produce a stronger final deck.