AI prompts for podcast show notes people actually use
Nobody reads a summary of an episode they just heard. These five prompts write podcast show notes that work as a reference and as search bait.
Most podcast show notes are a paragraph summarising the episode, a guest bio, and three links. The summary is read by nobody — people who have listened do not need it, and people who have not are not persuaded by it.
Show notes earn their keep in two other ways. They are a reference for the listener who is trying to find the thing that was mentioned at minute thirty-four, and they are the only text a search engine has to work with for an audio file. These five prompts write for both: a core notes builder from a transcript, a variant for interview episodes, a variant for the clips and social cuts, a review prompt that catches misattribution, and a batch prompt for a back catalogue.
Why do podcast show notes go unread?
Because they answer a question nobody has. The listener arriving at your notes has already heard the episode and wants one specific thing: the name of the book, the URL of the tool, the study that was quoted. A prose summary buries all three.
The second reason is that notes are written from memory or from a rough outline rather than from the transcript, so the specifics — which are the whole value — are exactly what goes missing. The producer remembers the argument and forgets the name of the paper.
Third, podcast show notes are usually the only indexable text for the episode, and a 90-word summary gives a search engine almost nothing. An episode that discusses six substantive topics for an hour is represented online by a paragraph that mentions two of them.
What should podcast show notes contain?
Timestamps with substance attached. Not “12:04 — discussion of pricing” but “12:04 — why they moved from per-seat to usage pricing and what it did to churn”. The timestamp is a navigation aid; the description is what makes someone use it.
Every proper noun mentioned: people, companies, books, papers, tools, places. This is the reference layer, and it is the thing listeners come back for.
The claims worth quoting, in the guest’s own words, because those are what get shared. And a short section on what the episode is actually about — written for someone who has not listened and is deciding whether to, which is a different job from summarising.
The core prompt for podcast show notes
Click any highlighted blank to fill it in before you copy.
You are writing podcast show notes from a transcript. Work only
from the transcript. Do not add context, background or
explanation that is not in it.
Podcast: {{name, what it covers, who listens}}
Episode: {{number, title if set}}
Guest: {{name, role, organisation — as stated in the episode}}
Duration: {{length}}
Transcript with timestamps: {{paste}}
Produce:
1. WHAT THIS EPISODE IS ABOUT — 40-60 words for someone
deciding whether to listen. Lead with the most specific
thing discussed, not the theme.
2. TIMESTAMPS — every substantive shift in topic. Each entry
is the timestamp plus what was actually said, specifically
enough to be useful without listening. Aim for one entry
per 3-5 minutes. Never write "discussion of X" — write the
position taken.
3. MENTIONED IN THIS EPISODE — every proper noun: people,
companies, books, papers, tools, products, places. Group
by type. Mark anything you could not identify confidently
from the transcript as [CHECK].
4. QUOTES — the three or four most quotable passages,
verbatim, with timestamps. Do not clean them up beyond
removing filler words.
5. THE ARGUMENT — if the episode makes a case, state it in
three bullets. If it does not, say so and skip.The accuracy rules
Click any highlighted blank to fill it in before you copy.
Rules:
- Never invent a surname, title, company or URL. If the
transcript says "Sarah from the Cambridge study", write
"Sarah [CHECK surname], Cambridge study [CHECK reference]".
- Never correct or improve a guest's claim. If they said
something factually wrong, the notes record what they said.
- Quotes must be verbatim. Removing "um" is allowed;
tightening a sentence is not.
- Do not attribute a host's statement to the guest or the
reverse. Where the transcript is ambiguous about who is
speaking, mark it [SPEAKER UNCLEAR].
- No promotional language about the episode. No "fascinating
conversation", no "you won't want to miss".
- Timestamps must come from the transcript, never estimated.
- Flag anything that sounds like a legal, medical or financial
claim that a listener might act on.A worked example: the timestamp difference
The same five minutes of an episode, written two ways. Conventional:
18:20 — Discussion of hiring
23:45 — Thoughts on remote work
And what this prompt produces:
18:20 — Why they stopped hiring senior engineers for eighteen months, and what it did to the mid-level team’s progression
21:10 — The interview question they use instead of a take-home, and the two things it reveals
23:45 — Why they went back to two office days after fully remote, and the specific meeting type that forced it
The second version is useful without the audio, findable in a search, and shareable. The first is a table of contents for a book nobody has.
The prompt for interview episodes
Click any highlighted blank to fill it in before you copy.
You are writing show notes for an interview episode, where the
value is in the guest's specific experience rather than in a
thesis.
Guest, exactly as introduced: {{name, role, organisation}}
What they are known for: {{as stated in the episode}}
Transcript: {{paste}}
Produce the standard sections, plus:
- WHAT THIS GUEST KNOWS THAT OTHERS DO NOT — two or three
specific things from their direct experience. Not opinions
they hold; things they have seen.
- NUMBERS THEY GAVE — every figure the guest quoted, with the
timestamp and enough context to interpret it. Mark any
figure they hedged as approximate.
- WHERE THEY DISAGREED WITH THE HOST — the points of genuine
disagreement, stated fairly from both sides. These are
usually the most interesting part of the episode and are
almost never in the notes.
- WHAT THEY WOULD DO DIFFERENTLY — if they said.
Rules:
- Represent disagreement neutrally. Do not resolve it in the
host's favour.
- Do not summarise the guest's background beyond what the
episode states. Their bio is not show notes.
- If the guest made a claim they could not support when
pressed, note the exchange rather than the claim alone.The prompt for clips and social cuts
Click any highlighted blank to fill it in before you copy.
You are identifying clip candidates from a podcast transcript.
Transcript with timestamps: {{paste}}
Platforms: {{where clips will go, and length limits}}
Audience: {{who follows us}}
What has performed well before: {{examples}}
For each candidate, output:
- Start and end timestamp.
- Duration.
- The verbatim opening line.
- Why it works as a clip in one sentence.
- Whether it stands alone without the surrounding context.
- Suggested on-screen text, under 10 words.
Find candidates that are:
1. A complete thought in under {{n}} seconds.
2. Opening with a claim or a question, not with throat-clearing.
3. Comprehensible with no prior context.
Rules:
- Never suggest a clip whose meaning changes when removed from
context. Flag any that are close.
- Do not propose a clip that makes a guest look worse than the
full passage does. Check what came before and after.
- Verbatim opening lines only. Do not write a better first
sentence than the one that exists.
- Rank by standalone clarity, not by how provocative they are.That second rule matters more than it looks. A clip that is technically accurate and misrepresents a guest costs you the guest, and eventually the guests who hear about it.
The review prompt: catching misattribution
Click any highlighted blank to fill it in before you copy.
Audit these podcast show notes against the transcript.
Output, quoting both the notes and the transcript line:
1. INVENTED DETAIL — any name, title, company, figure, date or
URL in the notes that is not in the transcript.
2. MISATTRIBUTION — anything credited to the wrong speaker.
3. ALTERED QUOTES — any quoted passage that differs from the
transcript beyond filler removal.
4. IMPROVED CLAIMS — anywhere the notes state something more
precisely or more defensibly than the guest actually did.
This is the most common and most dangerous error.
5. LOST HEDGING — where the guest said "roughly", "I think",
"about", or "this was a few years ago" and the notes state
it flatly.
6. UNRESOLVED [CHECK] MARKERS still present.
7. TIMESTAMP DRIFT — any timestamp that does not match.
TRANSCRIPT: {{paste}}
NOTES: {{paste}}Where an episode is transcribed automatically, remember the transcript is a draft too — accuracy guidance applies to what you publish about what someone said, not to what the software heard. Category five is the quiet one. A guest who said “I think it was around 40% but don’t quote me” appearing in the notes as “40% of users churned” is a misrepresentation you published, and it is the kind that ends up screenshotted.
Writing show notes for a back catalogue
Click any highlighted blank to fill it in before you copy.
You are producing show notes for {{n}} existing episodes that
have none, or have only a summary paragraph.
Podcast: {{name, topic, audience}}
Format reference: {{paste one set of approved notes}}
Episodes: {{table — number, title, guest, duration, transcript}}
For each episode, produce the standard notes.
Rules for the batch:
- Build a cumulative index: every proper noun mentioned across
all episodes, with the episodes and timestamps where it
appears. This is the most valuable artefact in the batch and
the reason to do them together.
- Identify episodes that cover the same topic, so they can be
cross-linked. Output: TOPIC | EPISODES.
- Flag episodes where the transcript quality is too poor for
reliable notes rather than producing unreliable ones.
- Flag any guest whose role or company has since changed, so
the notes can say "at the time of recording".
- Note any episode containing advice that may have dated,
particularly anything regulatory, technical or pricing
related.The cumulative index is the thing worth doing this for. A podcast with sixty episodes has a substantial reference work buried in it, and nobody can search audio.
Common mistakes with podcast show notes
Working from memory instead of the transcript. The specifics are the entire value, and they are exactly what memory drops. If the transcript is bad, fix the transcript first.
Cleaning up quotes to read better. It feels like editing and it is misquotation. The listener who heard the episode will notice, and the guest certainly will.
Writing the “about” section as marketing. Someone deciding whether to spend an hour wants to know what was specifically discussed, not that it was a fascinating conversation.
Ignoring the [CHECK] markers. They are the model correctly telling you it is unsure, and publishing over them converts a well-designed safeguard into a confident error.
What to check before podcast show notes are published
Resolve every [CHECK] marker against a real source. Names and papers are where errors embarrass you, and a wrong surname in show notes is permanently indexed.
Spot-check three timestamps by listening. Transcript timing drifts, especially where the audio was edited after transcription, and notes with wrong timestamps are worse than none.
Verify that anything a listener might act on — legal, medical, financial or safety-related — is attributed to the speaker rather than stated as fact in your own voice.
Finally, send the notes to the guest before publishing. It costs a day, it catches misattribution you cannot see, and it is the single cheapest way to keep guests willing to come back.
Frequently asked
Questions this article answers
Why do podcast show notes go unread?
Because they answer a question nobody has. The listener arriving at your notes has already heard the episode and wants one specific thing: the name of the book, the URL of the tool, the study that was quoted. A prose summary buries all three. The second reason is that notes are written from memory or from a rough outline rather than from the transcript, so the specifics — which are…
What should podcast show notes contain?
Timestamps with substance attached. Not "12:04 — discussion of pricing" but "12:04 — why they moved from per-seat to usage pricing and what it did to churn". The timestamp is a navigation aid; the description is what makes someone use it. Every proper noun mentioned: people, companies, books, papers, tools, places. This is the reference layer, and it is the thing listeners come back for. The claims worth quoting, in…
What to check before podcast show notes are published?
Resolve every [CHECK] marker against a real source. Names and papers are where errors embarrass you, and a wrong surname in show notes is permanently indexed. Spot-check three timestamps by listening. Transcript timing drifts, especially where the audio was edited after transcription, and notes with wrong timestamps are worse than none. Verify that anything a listener might act on — legal, medical, financial or safety-related — is attributed to the…