ChatGPT prompts for podcast show notes that people click, from the transcript up

The short version: ChatGPT can turn a raw transcript into finished show notes, timestamps, a summary, and social copy in under fifteen minutes, but it needs the real transcript pasted in, not just a topic name. Ask it to summarise an episode from memory and it will invent guests, invent quotes, and invent claims the host never made. Feed it the real words spoken and it becomes a fast, dependable production step that still needs a final human check.

Why the transcript has to come first

Show notes exist to help someone decide whether to listen and to help search engines understand what the episode covers. Both jobs need accuracy: a wrong guest title, a misquoted line, or a timestamp that points to the wrong moment all erode trust fast. Every prompt below assumes you have a transcript, even a rough automatically generated one, and starts from that text. If you do not have one yet, get a transcript first, from your recording software or a transcription tool, before you open ChatGPT.

Prompts for the episode summary and title

• Prompt: "Here is the full transcript of this episode: [paste transcript]. Write a two sentence summary that names the guest, their role, and the single most useful takeaway."
  What to change: the transcript. This becomes your meta description and the first line people see in a podcast app.
• Prompt: "Based on this transcript, suggest five episode titles under 60 characters, each built around a specific claim or number the guest said word for word, not a vague theme: [paste transcript]."
  What to change: nothing but the transcript. Specific titles outperform vague ones because they tell a potential listener exactly what they will learn.
• Prompt: "Write a 150 word episode description for the show notes page, opening with the guest's name and the core problem this episode solves: [paste transcript]."
  What to change: nothing but the transcript. Keep this separate from the shorter app summary, since the two serve different purposes.

Prompts for timestamps and chapter markers

• Prompt: "Here is a timestamped transcript: [paste transcript with timestamps]. Identify 8 to 12 natural topic shifts and give each one a short chapter title and its start time."
  What to change: the transcript, and make sure it includes timestamps, since the model cannot invent accurate ones from plain text.
• Prompt: "Check these chapter titles against the transcript and flag any that do not match what was discussed at that timestamp: [paste chapter list and transcript]."
  What to change: nothing but both pasted items. This catches drift when chapters were drafted before the final edit.

Prompts for pulling quotes and key moments

• Prompt: "Pull three direct quotes from this transcript that would work as standalone social media text, copying the exact wording, not a paraphrase: [paste transcript]."
  What to change: the transcript. Exact wording matters here, since a paraphrased "quote" attributed to a guest is misleading.
• Prompt: "Find the moment in this transcript where the guest gives a specific number or statistic, and quote that section exactly with its timestamp: [paste transcript]."
  What to change: nothing but the transcript. A number based quote tends to earn more clicks than a general opinion.
• Prompt: "List every named tool, book, or resource the guest mentioned in this transcript, with the timestamp where each is mentioned: [paste transcript]."
  What to change: nothing but the transcript. This becomes your resources section, and every item on it should be traceable to a real mention.

Prompts for social and repurposed content

• Prompt: "Write three short LinkedIn post drafts based on the three biggest takeaways from this transcript, each under 100 words: [paste transcript]."
  What to change: nothing but the transcript. Ask for a fourth draft if none of the first three feel right for your voice.
• Prompt: "Suggest five moments in this transcript, with timestamps, that would work as a 30 to 60 second audio or video clip, and explain why each one stands alone without context: [paste transcript]."
  What to change: nothing but the transcript. A clip that needs the surrounding conversation to make sense will not perform on its own.
• Prompt: "Write one email newsletter blurb, under 80 words, announcing this new episode to subscribers who already know the show: [paste episode summary]."
  What to change: the summary from an earlier prompt. Keep this shorter and more casual than the show notes page copy.

Prompts for guest bios and links

• Prompt: "Here is what the guest said about their own background in this transcript: [paste relevant section]. Write a two sentence bio using only what they said, and mark with [VERIFY] anything you think needs checking against their own website."
  What to change: the section where the guest introduces themselves. Never let the model add a job title or company detail it was not given.
• Prompt: "List every question I asked in this interview, in order, based on the transcript, so I can reuse the strongest ones in a future episode: [paste transcript]."
  What to change: nothing but the transcript. Useful for building a question bank across episodes.

Prompts for SEO and search friendly notes

• Prompt: "Based on this transcript, what are the three or four questions a listener searching online is trying to answer by listening to this episode: [paste transcript]."
  What to change: nothing but the transcript. Use these questions as subheadings in your show notes page.
• Prompt: "Write a frequently asked questions section with 4 short questions and one paragraph answers each, using only facts stated in this transcript: [paste transcript]."
  What to change: nothing but the transcript. This gives the show notes page more structure for both readers and search engines.

A weak prompt versus a strong prompt

Weak: "Write show notes for my podcast episode about productivity with a guest called Sarah."

Without a transcript, the model has nothing real to draw on. It will invent quotes for Sarah, invent a book she recommended, and invent a job title, all of which sound plausible and are entirely fabricated.

Strong: "Here is the full transcript of episode 42 with Sarah Chen, a time management coach: [paste transcript]. Write a 150 word episode description, three pull quotes using her exact words, and a list of every book or tool she named with the timestamp for each."

The strong version anchors every output to real spoken words. Nothing in the result can be a fabrication, because the model is working from source text rather than a topic label.

Mistakes that make show notes worse, not better

• Skipping the transcript step entirely. Asking the model to write notes from a topic and guest name alone guarantees invented detail somewhere in the output.
• Trusting auto generated timestamps without checking them. Automatic transcription timestamps can drift, especially in longer episodes, so spot check a few chapter markers against the actual audio before publishing.
• Publishing a "quote" that is really a paraphrase. If the wording differs from what was said, label it as a summary, not a quote, or your guest may reasonably object.
• Writing the same summary structure for every episode. Interesting episodes have a distinct hook. Ask specifically for what made this episode different before falling back to a template summary.
• Forgetting to add real links to the guest's own site or socials. ChatGPT cannot verify a live URL, so always add and check these yourself after the draft is done.

I have written more around this on the site: How to Use ChatGPT for Marketing in 2026, Where to Learn AI Tools for Free as a Busy Founder (No Time, No Budget, No Excuse), AI News This Week for Small Business, 26 July 2026.

For the rest of the series, see the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.

Related tool: my free tool that writes the prompt for you.

Frequently asked questions

Do I need a perfect transcript before I start?

No, a rough automatically generated transcript with occasional errors works fine for drafting show notes. Just check any pulled quote against the actual audio before it goes out, since transcription tools sometimes mishear a word or a name.

How long should show notes be?

There is no single right length, but a page with a short summary, a chapter list, a few pulled quotes, and a short FAQ section tends to serve both listeners and search engines better than either a two line summary or a full episode transcript dumped as is.

Can ChatGPT write the whole show notes page in one prompt?

You can ask for everything at once, but breaking the work into the smaller prompts above, one for the summary, one for timestamps, one for quotes, tends to produce more accurate output, since each prompt gives the model a narrower, clearer task.

What should I do if the transcript has multiple guests talking over each other?

Clean up speaker labels in the transcript first, even roughly, before pasting it in. A transcript that clearly separates who said what produces far more reliable quotes and attributions than one where speakers are jumbled together.

Should show notes mention every topic covered in a long episode?

Focus on the strongest three or four moments rather than trying to cover everything discussed. A show notes page that lists twenty minor points reads as noise, while one built around a handful of specific, well chosen moments is more likely to convert a browser into a listener. If the episode covered several distinct topics, consider splitting the extra ground into a follow up post or a second episode rather than cramming it all into one page.