AI-Assisted Podcast Show Notes: Timestamps, Quotes, and the Human Pass
A workflow: structured outline from the transcript, pull quotes, then a single read-aloud for voice match — before the episode ships.
Show notes are a discovery surface for Google and a retention surface for YouTube. Pure AI paste reads flat; pure manual does not scale. The hybrid: generate chapters and 3–5 must-hear pull quotes, then a human tightens the top paragraph so it sounds like the host, not a press release.
The outline prompt that works
Feed the model the transcript and ask for: one-sentence cold open, 5–8 timestamped H2s with 2-sentence takeaways, and 3 "signature lines" in the guest's voice. Ban superlatives unless the transcript supports them.
What to hand-edit
Names, book titles, numbers, and anything that could be wrong legally. The model is a fast drafter, not a fact checker.