Show notes are the chore nobody books time for. The episode's done, you're tired, and now you owe a summary, timestamps, links, and a guest bio — for a page most listeners skim. So it gets rushed, or skipped, or handed to a chatbot that returns "In this captivating episode, we dive deep into the fascinating world of..."
That opening is the tell. A podcast audience — and the host re-reading it — clocks it instantly. It reads like nobody listened to the episode, because nobody did.
I maintain a free, MIT-licensed library of Claude skills for media producers. The show-notes skill is one of the most-used, because it removes the chore without leaving the fingerprint.
What it does
You paste the transcript or a rough outline. It returns:
- A summary that sounds like the show — written in the register of your podcast, not the generic "join us as we explore" template.
- Chapter timestamps pulled from the actual structure of the conversation, not invented.
- A links/resources block for anything named in the episode.
- A short guest bio when there's a guest, drawn from what was actually said — flagged for you to confirm, never fabricated.
Why it doesn't sound like AI
The skill runs against the same quality bar as everything in the library: a seven-dimension rubric whose hard floor is whether the output reads like a human who works in the medium. The banned-phrase list — "dive deep," "captivating," "in this episode we explore" — is enforced, not suggested. What you get back is a draft you can post, not one you have to launder first.
The honest limit
It's only as good as what you feed it. A clean transcript gives clean notes; a vague three-line outline gives a thinner summary and more placeholders for you to fill. And it won't invent a guest's credentials or a stat that wasn't in the episode — if the source doesn't say it, the notes flag the gap instead of guessing. That restraint is the point: wrong show notes are worse than late ones.
Paste an episode in and see. No install — open the skill, copy it into Claude, drop your transcript below it.
Top comments (0)