A podcast show notes ai prompt should produce four things: a summary someone can read instead of listening, chapter timestamps that match what was actually said, quotes worth clipping, and links worth keeping. Most generated show notes produce one of those four well, usually the summary, and quietly invent the other three.
That matters more than it looks. Timestamps are navigation your listeners depend on. A quote that does not appear in the episode damages trust with the guest, and a link to something nobody said in the conversation is a small fiction in a document that is supposed to be a record. Below are prompts that produce each part separately, so you can verify them independently.
The full show notes prompt
Paste the transcript, not a summary of the transcript. The model can only be accurate about what is in front of it.
Write show notes for [show name], episode [number and title]. Guest: [name and role, as they should be credited]. Episode summary: [two or three sentences describing the conversation, no adjectives I did not use]. Topics covered: [list]. Write: (1) a 120 word summary that makes sense to someone who never heard the episode, (2) five to eight chapter timestamps with a short descriptive label for each, (3) three pull quotes of no more than 30 words each, (4) a resources section listing the tools, books, papers or links the guest actually mentioned, and (5) three episode questions for the listener. Use the guest's own words from the transcript for anything you attribute to them. Do not invent statistics, study names, job titles, book titles, or URLs. Do not paraphrase a quote and present it as a quote. Where the transcript is unclear, write [CONFIRM FROM TRANSCRIPT] instead of guessing. Transcript: [paste]
Generate timestamps from a sectioned transcript
Timestamps generated from a summary will not match the audio. Ask for them from the transcript and from a structure you have already identified.
Read this transcript and produce chapter timestamps in the format [00:00 Chapter name]. Aim for between five and eight chapters, each covering between [minimum minutes] and [maximum minutes]. Name each chapter after the question it answers, not after the topic area. The first timestamp must be 00:00. Do not invent a timestamp for anything the transcript does not contain. If a section boundary is ambiguous, give the timestamp with a marker [VERIFY TIMING] rather than a confident value. Transcript: [paste]
Then play the file and correct the boundaries. This is a two minute job that removes the most common complaint you get about a show.
Pull quotes worth clipping
A pull quote has to survive being read on its own, out of context, on a card with no other information. Most models return summarising sentences instead.
From this transcript, extract six candidate pull quotes of 15 to 30 words each. A candidate must: stand as a complete thought without the surrounding conversation, express something the guest would recognise as their own view, and be quotable without misleading the listener about the episode's subject. Rank them and give the reason for the top three. Do not smooth grammar, merge two separate moments, or drop a qualifier such as "in that case" or "at the time". If the transcript does not support a candidate, do not produce it. Transcript: [paste]
The qualifier rule is the important one. Stripping "at the time" from a claim turns a dated observation into a general truth, which is how a clip ends up being screenshotted as the guest being something they never said.
Match the notes to where they will be read
Podcast host pages
Most hosts accept markdown or a formatted text field with a short description and a longer notes section. Give the model the character or word limits your host actually enforces, since exceeding them can silently truncate your links.
Social clips
The quote prompt doubles as a clip script. Cut the quote to a card of a few words, keep the full version in the notes, and attribute it exactly as the guest said it.
Search and discovery
Episode titles and descriptions are what listeners search on. Use the guest's own frame of reference for their expertise rather than a generic label, and keep the description readable before any keywords.
Worked example: notes for a 45 minute interview
A host interviews a freight forwarding company owner about how small importers handle customs delays. The transcript runs 8,400 words with several tangents into a family business and a sailing trip.
| Section | Result | Check needed |
|---|---|---|
| Summary | 120 words on the customs delay problem and the three-step approach the guest described | Names the guest's role correctly |
| Timestamps | Seven chapters, first at 00:00, longest at 11 minutes | Two boundaries marked [VERIFY TIMING] |
| Quotes | Three ranked, top one retains the "at the time" qualifier | Quote matches audio word for word |
| Resources | Two items the guest named on tape, no URLs | Guest credited both by name on tape |
The two marked timestamps fell inside a long answer where the transcript has overlapping timestamps, which is exactly the case the model should flag rather than resolve. The summary also dropped the sailing tangent entirely, which is correct for a summary and worth knowing before the guest asks.
Check the notes before you publish
1. Verify every quote against the audio
Play each quote. If a word was changed, fixed, or moved across a pause, correct it. Guests read their own show notes, and they check.
2. Confirm the timestamps by playing the boundaries
Start playback five seconds before each chapter and listen for the handoff. Correct anything off by more than a few seconds, since the value of a timestamp is that it is exact.
3. Check every link and attribution
Open each link. Confirm the title of each book, tool or paper is written the way the author or publisher writes it. A mis-titled source in notes travels to other people's posts.
4. Read the summary as a non-listener
Give it to someone who did not hear the episode. If they cannot tell what was discussed and why it matters, the summary is too abstract.
5. Confirm the guest credit
Check the spelling, the role, and the company exactly as the guest asked. This is the cheapest thing to get right and the most visible when it is wrong.
Common show notes prompt mistakes
- Feeding a summary instead of a transcript. Timestamps and quotes cannot be accurate from a summary.
- Asking for everything in one pass. Split the summary, timestamps, quotes and resources so each can be checked on its own.
- Accepting quotes without the qualifier. Removing "at the time" changes what the guest said.
- Ignoring the host's limits. Most hosts truncate long fields and drop the links first.
- Listing resources the guest only alluded to. If it was not named on tape, either leave it out or mark it as your own recommendation.
- Writing the description last. The first two lines of the description are read in most interfaces, so write them deliberately.
Frequently asked questions
Can AI generate accurate podcast timestamps?
From a clean transcript, it can get close enough to correct by ear. Overlapping or low-quality timestamps are where it goes wrong, so flag anything ambiguous and verify each boundary yourself before publishing. Some hosts also generate chapter markers natively, which is worth checking before building your own workflow.
How long should podcast show notes be?
Long enough that a non-listener understands the episode, and no longer. A 100 to 150 word summary plus timestamps, quotes and links covers most interviews. If you are writing for a guest who will share the episode, add a short list of the specific questions you covered.
Should I write quotes into the show notes at all?
Only if you are accurate with them. Quotes are the most-quoted and most-misquoted part of any episode. Use them if you check every word against the audio, and skip them entirely if you cannot.
Publish notes with the episode, not after
Run the summary, timestamps and quote prompts on a transcript from your most recent episode, then play the boundaries and check the quotes before you press publish. Doing it on an episode you already have beats drafting a template you have not tested. For more episode production and promotion prompts, browse the podcast success prompt collection.