Generate Show Notes for Your Podcast Inside ChatGPT (Whipscribe Workflow)
One episode mp3 becomes show notes with chapter markers, a tweet thread, and a blog post draft — without leaving ChatGPT. The Whipscribe Custom GPT transcribes the episode with speaker diarization; ChatGPT does the rest. Save the prompt as a Recipe and you'll run it on every future episode in one short message.
Photo: Will Francis · Unsplash
Why ChatGPT is the right place to do this
The post-production stack for an indie podcast is usually a recording tool, an editing tool, a hosting tool, and a separate writer-or-LLM-tool for the show notes. The writing stage is the one that consistently slips because it's the one without a deadline forcing function — the episode is published the moment the file uploads, but the show notes can always be "I'll write them tomorrow."
Doing the writing inside ChatGPT, against a real diarized transcript of the episode, collapses the loop. Drop the file in the Whipscribe GPT, ask for the artifacts, leave with everything you'd otherwise spend an hour on. The transcript is also saved to your Whipscribe library, so a month later when someone asks "what episode did you talk about X?" you can search it from the same chat.
Setup — about 90 seconds, one time
If you haven't connected Whipscribe to ChatGPT yet, the setup guide covers both paths. The short version:
- Open the Whipscribe Custom GPT in ChatGPT (works on every plan including free).
- The first time you ask for a transcript, sign in to Whipscribe with the email you'll use long-term.
- Done. Drop a file and ask.
ChatGPT Plus and Pro users have the option to add Whipscribe as an MCP Connector at https://whipscribe.com/mcp — same backend, available in every chat without switching to a specific GPT.
The Recipe — write once, run weekly
The prompt that produces all four artifacts in one pass. Save this verbatim:
Transcribe this podcast episode with speaker labels and word-level
timestamps. Speaker 1 is the host; Speaker 2 is the guest unless
otherwise stated. Map names from the intro if they're given.
Then produce four artifacts from the transcript:
1. SHOW NOTES (250–400 words)
- Two-sentence summary at the top
- 3–4 paragraphs covering what the episode is about, why it
matters, and the most memorable moment
- 1 pull-quote from the guest, with attribution
- "Links mentioned" section if anything was named
2. CHAPTER MARKERS
- 5–8 chapters
- Format: "0:00 Chapter title" (one per line)
- Use plain "0:00" / "12:34" timestamp format — YouTube will
parse these into chapters automatically
- Title each chapter in 3–6 words, the topic not the activity
3. TWEET THREAD (5–8 tweets)
- First tweet hooks with a concrete claim or surprising line
- Each subsequent tweet is one self-contained thought
- Last tweet is the call: link to the episode + your handle
- Tweets max 280 chars each; no thread numbering ("1/8") inside
4. BLOG POST DRAFT (1,200–1,500 words)
- H1: a question-shaped title the episode answers
- 4–6 H2 sub-questions
- Quotes from the guest are block-quoted and attributed
- Write as an original take that uses the conversation as
evidence — not a "the guest said X" summary
- Include a "Listen to the full conversation" CTA at the end
Then save the transcript to my "Episodes" Knowledge folder, named
with the episode date and title.To save it as a Recipe, ask in chat:
Save as Recipe
Save the prompt I just used as a Recipe.
Name it "Podcast → show notes + chapters + thread + blog".
Tag it: podcast, show-notes.From the next episode forward, the workflow becomes one message:
Run the Recipe on a new episode
Run my "Podcast → show notes + chapters + thread + blog" recipe
on this file.
[attach episode-074.mp3]First 30 minutes of audio per day are free. No card required.
Open the Whipscribe GPT →What the output actually looks like
A representative response on a 62-minute interview-style episode. Sample data, not from a real episode:
Ep 074 · How marketplaces find their second side
Show notes
Marketplaces almost always launch one side before the other — sellers before buyers, or vice versa. Getting the second side to show up is the actual hard problem. This episode pulls apart what worked at three companies that made the second-side leap in under 12 months.
The guest, a Series B operator who has shipped two marketplaces, walks through the patterns that compounded and the ones that looked smart in retrospect but were noise. The most memorable line in the episode is about choosing concentration over breadth — “you don't need 50 cities, you need three really hot ones.”
"Liquidity is not a feature. It's not a campaign. It's a property of the geometry of your supply and demand. If the geometry's wrong, no amount of growth marketing fixes it." — Guest, 00:38:14
Links mentioned: the marketplace-liquidity essay (Andreessen Horowitz, 2018); Tom Tunguz on cold-start problems; the guest's substack.
Chapter markers (paste into your description)
0:00Cold open — what we got wrong about supply3:42The three-city hypothesis14:18Why subsidies usually backfire22:55How retention told us what to scale35:10The geometry of liquidity47:30Pricing as the second-side magnet56:05What we'd do differently this time
Tweet thread (first tweet shown, full thread in your chat)
Most marketplace failures aren't supply or demand failures. They're geometry failures — the supply and demand never line up in the same time, place, and price. New episode unpacks how three companies fixed the geometry inside 12 months. (continues for 6 more tweets)
Blog post draft
(1,400 words returned in the chat — header structure, block-quoted guest moments, original take. Copy into your CMS to edit.)
Why each artifact is shaped the way it is
Show notes — written for the listener-deciding-to-play
The show-notes block on Apple Podcasts and Spotify is for someone hovering over the play button, deciding whether the episode is worth their next hour. Two-sentence summary at the top + the most memorable moment + one pull-quote does that work. Long show notes are a misuse of the surface — listeners don't read 800 words about an episode they haven't started.
Chapter markers — formatted exactly the way YouTube and Apple parse
YouTube parses 0:00 Chapter title in the description into in-player chapters automatically. Apple Podcasts and Spotify both surface chapter markers in their players when the episode embeds them. Either way, the format the Recipe outputs is what these platforms expect. Don't reformat. Paste as-is.
Tweet thread — the hook is the only part that matters
Most podcast tweet threads die because the first tweet describes the episode instead of starting a conversation. The Recipe forces the first tweet to be a concrete claim or a surprising line, not "new episode out." That single change carries threads further than any amount of formatting.
Blog post draft — useful starting point, not a publishable artifact
Be honest with yourself about this one. A 1,500-word post generated by ChatGPT from a transcript is a useful first draft, but Google has been actively demoting machine-rewritten transcript-based posts for years. Plan to spend 60-90 minutes editing — adding your own framing, cutting padding, sharpening the H2s — before publishing. Treat the draft as a structured outline that already has the quotes ready to drop in. The pattern's covered in our podcast SEO repurposing post if you want the deeper dive.
Quality knobs the Recipe leaves you
- Speaker mapping. If your episode has a co-host, change "Speaker 1 is the host; Speaker 2 is the guest" to your real names so attributions land correctly.
- Pull-quote count. Default is 1 in the show notes. For an interview-heavy episode, bump to 2-3. Edit the Recipe and re-save.
- Chapter granularity. 5-8 chapters is right for a 60-minute episode. For 90+ minute episodes bump to 8-12; for 30-minute episodes drop to 3-5.
- Blog post tone. Add a sentence to the Recipe like "Write in a conversational, second-person voice" or "Use a serious, reported tone" — the LLM defaults to a flat middle without it.
- Title style. Add "Use a question-shaped H1 the episode answers" (already in the Recipe) or swap to "Use a one-sentence declarative H1 stating the central insight."
What about a video podcast?
The Recipe works the same way for video files (mp4, mov, mkv). Whipscribe transcribes the audio track, ChatGPT produces the same four artifacts. The chapter markers are particularly useful for video — YouTube turns the timestamp list directly into in-player chapters, which improves session length and retention.
One nuance: when uploading a video file directly to ChatGPT, the upload time is the long pole. For long videos (45+ min), upload to a hosted location you control and pass the URL to the Whipscribe GPT instead — it's faster than the chat upload path.
Don't run this on episodes you don't own
The Recipe is built for podcasters processing their own episodes. Don't paste a third-party podcast feed URL or a competitor's episode and ask for the artifacts — that's a different category of question with platform-specific terms attached. Stay on episodes where you own the audio.
If you're not a podcaster yet
This same Recipe pattern works for any long-form audio you produce — webinars, conference talks, video interviews. The four-artifact shape (notes + chapters + thread + draft) maps to those formats too. The thing that's specific to podcasting is the "0:00 chapter" format and the show-notes shape; the rest is general.
Frequently asked
Can ChatGPT really write show notes from a podcast episode?
Yes — when you give it a real diarized transcript. The Whipscribe GPT transcribes the episode with speaker labels and word-level timestamps; ChatGPT writes the show notes from the structured turns. The chapter markers and pull-quotes both come from the timestamped transcript.
How accurate are the chapter markers?
Chapter markers come from word-level timestamps — the timing is exact. The choice of where to break is a judgment call, and it's usually within 30 seconds of where you'd put one yourself. Audit the first three episodes; from then on most podcasters trust the picks.
What about a blog post draft from the same episode?
The Recipe produces all four artifacts in one pass. The blog draft is a starting point, not a publishable piece — Google demotes machine-written transcript rewrites, so plan to do a 60-90 minute editorial pass before publishing.
Can I do this for every weekly episode?
That's exactly what Recipes are built for. From the next episode onward, the workflow is one short message — "run my podcast-recap recipe on this" with the new mp3 attached.
Does this work for video podcasts?
Yes — Whipscribe transcribes video files (mp4, mov, etc.) as well as audio. Chapter markers are useful for YouTube descriptions, since YouTube parses the 0:00 timestamp format directly into chapters.
What if my episode has a guest, not a co-host?
Diarization handles two-speaker setups well. Tell the GPT "Speaker 1 is the host, Speaker 2 is the guest" (or use names from the intro) and the show notes will attribute pull-quotes correctly.
Run the workflow on your next episode
Open the Whipscribe Custom GPT, drop your latest episode, paste the Recipe above, save it. From episode 2 forward, you send one short message and the four artifacts come back in the same shape.