TL;DR: These 25 prompts cover the parts of podcast production that are mostly typing: guest research briefs, cold outreach, interview question ladders, transcript cleanup, clip selection, episode titles, and repurposing one episode into a week of posts. None of them listen to audio, browse the web, or publish anything, so you paste in notes and transcripts and copy the output out.
What Does AI Actually Handle in a Podcast Production Pipeline?
Worth mapping out once, before the prompts, because the honest boundary is the same in every section below: AI drafts text from material you already have, and a person still does everything that requires listening, judgment, or hitting publish.
| Production stage | What a prompt handles | What still needs a human |
|---|---|---|
| Guest research | Turning notes into a brief | Finding the notes in the first place |
| Outreach | Drafting the pitch and follow-up | Deciding who to invite, and sending it |
| Interview prep | Building the question ladder | Reading the room live and adapting |
| Transcript cleanup | Removing filler, fixing speaker labels | Checking the fix against the actual audio |
| Clip selection | Pointing to strong lines with a reason | Cutting the actual clip |
| Titles | Generating and pressure-testing options | Picking the one that fits the show |
| Chapter markers and show notes | Drafting the list and summary | Uploading it in your host's specific format |
| Repurposing | Drafting captions and post outlines | Actually posting them |
Nothing in that right-hand column is optional busywork you can prompt your way out of. It's the actual production work; the prompts below just clear everything around it off your plate first.
How Do You Research a Podcast Guest Before You Ever Message Them?
The point of a research prompt is compression, not discovery: it takes the bio, articles, and posts you've already found and turns them into something you can actually use in five minutes before an interview, instead of a pile of open tabs. It doesn't go find that material for you — paste in what you've already gathered, because a model asked to research a guest with nothing to read will produce something plausible-sounding and wrong.
1. Turn scattered notes into one guest brief
Role: a podcast producer preparing a guest brief.
Task: turn the pasted material below into a one-page guest brief: who they are, what they're
known for, their current project, and 3-4 topics they'd likely want to discuss.
Material: [PASTE BIO, LINKEDIN TEXT, ARTICLE EXCERPTS, ETC.]
Format: short sections — background, current project, likely topics, one open question.
Constraints: use only the material provided; if something isn't covered in it, write
"not covered in the material provided" rather than filling the gap with a guess.
Tone: concise, briefing-document style.
What to change: the more specific the pasted material, the less generic the brief. A guest's own words from a recent interview or post beat a two-line company bio every time.
2. Find the angle nobody else asks this guest about
Role: a podcast producer looking for a fresh angle on a guest who's been interviewed before.
Task: given the material below, including any prior interview excerpts, identify 2-3 topics
or questions that seem under-explored compared to what they usually get asked.
Material: [PASTE BIO + ANY PRIOR INTERVIEW EXCERPTS OR ARTICLES YOU HAVE]
Format: bulleted list, each item naming the angle and why it looks under-explored.
Constraints: base "under-explored" only on the material provided; do not claim to know what
every other interviewer has asked if you weren't given that material.
Tone: curious, specific.
3. Spot potential landmines before you ask about them live
Role: a podcast producer doing a sensitivity pass before an interview.
Task: given the material below, flag any topic that seems sensitive, contested, or likely to
need careful framing if raised on air, and suggest a considerate way to phrase a question
about it.
Material: [PASTE BIO + ARTICLES]
Format: bulleted list — topic, why it's sensitive, one suggested phrasing.
Constraints: flag only what's actually indicated in the material; do not speculate about
anything not covered by what was pasted in.
Tone: careful, non-judgmental.
What to change: treat every flag as a framing exercise, not a reason to avoid the topic. Most sensitive subjects are entirely fair to raise on air; the prompt's job is finding a considerate way in, not talking you out of asking.
How Do You Turn a Guest Brief Into a Cold Outreach Email That Gets a Reply?
A guest brief is only useful once it's turned into a pitch that respects the reader's time. The prompts below assume you already know why this particular guest fits your show. The model's job is phrasing and structure, not deciding who to invite.
4. First cold outreach email
Role: a podcast host writing a cold outreach email to a potential guest.
Task: write a short pitch inviting [GUEST NAME] onto [PODCAST NAME], explaining in one line
why they specifically fit the show and what topic you'd want to cover.
Details: [PODCAST NAME, AUDIENCE SIZE OR NICHE, WHY THIS GUEST, PROPOSED TOPIC]
Format: subject line under 50 characters, body under 120 words, one clear ask (a reply, or
a link to book a time).
Constraints: no generic flattery ("huge fan of your work!") without a specific reason attached.
Tone: warm, direct, respectful of a busy inbox.
What to change: the "why they specifically fit" line is what separates this from a form letter. If you can't fill it in with something true and specific, that's worth noticing before you send, not after.
5. Follow-up when the first email gets no reply
Role: a podcast host sending one follow-up to an unanswered invitation.
Task: write a short, low-pressure follow-up to the email below, sent [NUMBER] days after
the original with no response.
Original email: [PASTE YOUR FIRST EMAIL]
Format: under 60 words, restates the ask in one line, easy to ignore without awkwardness.
Constraints: exactly one follow-up; do not suggest building a longer sequence.
Tone: light, no guilt, genuinely fine either way.
What to change: send this once, then move on either way. A guest who was going to say yes eventually rarely needs a third email, and a third email rarely changes a genuine no.
6. Warm-intro request to a mutual connection
Role: someone asking a mutual connection for an introduction to a potential guest.
Task: write a short message to [MUTUAL CONNECTION] asking if they'd be willing to introduce
me to [GUEST NAME] for a podcast interview, giving them an easy way to say no.
Details: [YOUR RELATIONSHIP TO THE MUTUAL CONNECTION, WHY YOU WANT THE INTRO]
Format: under 80 words, includes a one-line forwardable blurb they could use to make the intro.
Constraints: make declining easy and consequence-free; do not pressure a specific timeline.
Tone: casual, respectful of their relationship with the guest.
How Do You Build a Question Ladder Instead of a Flat List of Questions?
A flat list of ten questions treats them as interchangeable. A ladder orders them on purpose: a broad, easy opener that gets the guest talking, then more specific questions that use what they just said, then room for a harder or more personal question once there's enough trust in the conversation to ask it. Building the shape first, before the individual questions, is what makes an interview feel like a conversation instead of a survey.
7. Generate a full question ladder from a guest brief
Role: a podcast host building an interview question ladder.
Task: using the guest brief below, generate a question ladder: 2 broad opening questions,
3-4 specific questions building on likely answers, 1-2 harder or more personal questions for
later in the conversation, and 1 closing question.
Guest brief: [PASTE THE BRIEF FROM THE PROMPT ABOVE]
Format: four labeled sections — opener, specifics, harder questions, closer.
Constraints: each specific question should reference something concrete from the brief, not
a generic version that could apply to any guest.
Tone: curious, conversational, not interrogation-style.
What to change: treat the harder-questions section as a suggestion for where the conversation could go, not a script you're committed to following. The best moment to ask a hard question is whenever trust has actually built, which may not match where the ladder placed it.
8. Generate follow-up branches for a likely answer
Role: a podcast host preparing for how a conversation might branch.
Task: given the question and the likely answer direction below, generate 2-3 follow-up
questions for each plausible direction the guest's answer could take.
Question: [YOUR PLANNED QUESTION]
Likely answer directions: [YOUR BEST GUESS AT HOW THEY MIGHT ANSWER, 1-2 DIRECTIONS]
Format: one branch per direction, 2-3 follow-ups each.
Constraints: keep follow-ups specific to the stated direction, not generic "tell me more"
prompts that would fit any answer.
Tone: attentive, building on what's said rather than moving to a new topic.
What to change: this is prep, not a transcript to read from live. Skim the branches beforehand so the shape is in your head, then actually listen to what the guest says instead of scanning a printout for the matching line.
9. One deliberately harder question, framed considerately
Role: a podcast host preparing one harder question for later in an interview.
Task: given the topic below, phrase one direct, harder question about it in a way that's
still respectful and gives the guest room to answer honestly.
Topic: [WHAT YOU WANT TO ASK ABOUT, EVEN BLUNTLY, IN YOUR OWN WORDS]
Format: one question, plus one alternate phrasing if the first feels too blunt on the day.
Constraints: keep the substance of the question intact; soften the phrasing, not the ask.
Tone: direct but not adversarial.
How Do You Turn a Raw Transcript Into Something Usable?
Everything from here on assumes a transcript exists, because a model working from a transcript is describing something real, and a model working from a title alone is inventing an episode. Raw transcripts are messy in predictable ways: filler words, false starts, misattributed speakers. Cleaning them is a mechanical job worth doing before anything else touches the text.
10. Clean filler and false starts while keeping the voice
Role: a transcript editor cleaning a raw podcast transcript.
Task: remove filler words ("um," "like," false starts, repeated words) from the transcript
below without changing what either speaker actually said or how they said it.
Transcript: [PASTE THE RAW TRANSCRIPT]
Format: same structure as the input, cleaned in place.
Constraints: do not rephrase, summarize, or improve word choice; remove disfluency only,
keep every actual claim and word choice intact.
Tone: neutral, mechanical cleanup.
What to change: run this in chunks for a long episode. A model cleaning ninety minutes of dialogue in one pass tends to get looser about "only removing filler" toward the end, and a spot check on the last third catches that drift early.
11. Fix speaker labels in a transcript with mixed-up attribution
Role: a transcript editor fixing speaker attribution.
Task: given the transcript below, where speaker labels are inconsistent or missing in places,
use context (who's asking questions vs. answering, referenced names) to assign each line to
the correct speaker.
Transcript: [PASTE THE TRANSCRIPT]
Speakers: [LIST WHO IS IN THE CONVERSATION]
Format: same transcript, with a speaker label added or corrected on each line.
Constraints: mark a line "[UNCLEAR SPEAKER]" rather than guessing when context genuinely
doesn't indicate who's talking.
Tone: neutral, mechanical.
12. Pull the strongest quotes from a cleaned transcript
Role: an editor extracting pull-quotes from a podcast transcript.
Task: read the transcript below and extract 8-10 quotes that would work as standalone
pull-quotes: sharp, self-contained, understandable without the surrounding context.
Transcript: [PASTE THE CLEANED TRANSCRIPT]
Format: numbered list, each quote exact, with the speaker's name and an approximate location
in the conversation (early / middle / late) since the plain text has no timestamps.
Constraints: quote exactly; do not trim a sentence in a way that changes its meaning.
Tone: selective, not exhaustive.
What to change: run this on the cleaned transcript, not the raw one. A pull-quote lifted from an un-cleaned transcript comes with the filler words attached, and trimming those out afterward risks becoming the exact meaning-changing edit the prompt was told not to make.
How Do You Find the Clips Worth Publishing?
A model can't watch or listen to your recording, so a clip-selection prompt works from the transcript and points to specific lines. You still have to find those exact words in your editor and make the cut. What it's actually good for is triage: reading two hours of conversation and returning ten candidates with a reason each, instead of you re-listening to the whole thing hunting for a moment you half-remember.
13. Find clip candidates with a reason for each
Role: a social media editor scanning a podcast transcript for clip candidates.
Task: read the transcript below and identify 8-10 moments that would work as a standalone
short clip, quoting the exact lines and explaining briefly why each one works alone.
Transcript: [PASTE THE TRANSCRIPT]
Format: numbered list, quoted lines plus one-line reason (surprising claim, strong emotion,
concrete story, a clean hook-payoff pair).
Constraints: quote exact lines you can search for in the transcript; do not estimate a
timestamp you weren't given.
Tone: selective, editorial.
14. Rank clip candidates by hook strength
Role: a social media editor prioritizing which clip to cut first.
Task: given the clip candidates below, rank them by how strong the first line is as a
scroll-stopping hook, most compelling first.
Clip candidates: [PASTE THE LIST FROM THE PROMPT ABOVE]
Format: reordered list, one-line reason for each ranking.
Constraints: judge the opening line specifically, not the clip as a whole; a clip can be
substantively strong but open weakly.
Tone: blunt, hook-focused.
What to change: if a clip ranks low on hook strength but the content is genuinely strong, consider re-cutting where it starts rather than dropping it. Sometimes the fix is three seconds earlier or later, not a different clip entirely.
15. Write a caption and on-screen hook text for one clip
Role: a social media editor writing the caption and hook text for one clip.
Task: given the clip transcript below, write a short caption and an on-screen hook line for
the first second of the clip.
Clip transcript: [PASTE THE EXACT CLIP TEXT]
Format: caption under 100 characters, hook text under 8 words.
Constraints: the hook text must be something the speaker actually said or a close, honest
paraphrase, not an invented claim stronger than what's in the clip.
Tone: attention-grabbing, still accurate to the clip.
What to change: check the hook text against the clip one more time after editing. A caption written before the final cut sometimes references a line that got trimmed out of the version you actually publish.
How Do You Write an Episode Title That Isn't Generic?
A title has two audiences pulling in different directions: someone searching for the topic by name, and someone scrolling a feed who's never heard of it and needs a reason to click. Generating options for both, then picking deliberately, beats writing one title and hoping it does both jobs.
16. Generate title options for search and for curiosity
Role: a podcast producer titling an episode.
Task: given the transcript summary below, generate 5 titles optimized for someone searching
the topic by name, and 5 titles optimized for scroll-stopping curiosity in a feed.
Episode summary: [PASTE A SHORT SUMMARY OR THE TRANSCRIPT]
Format: two labeled groups of 5, each title under 70 characters.
Constraints: every title must describe something actually discussed in the episode; no
title implying a topic or claim that isn't in the material.
Tone: two distinct styles, clearly separated.
What to change: pick from both groups rather than defaulting to whichever one sounds more fun to write. A show that's mostly discovered through search needs the search-shaped title working, even when the curiosity-shaped one reads better out loud.
17. Pressure-test a title against the actual content
Role: an editor checking whether a proposed title overpromises.
Task: given the title and the episode summary below, state whether the title accurately
represents what's covered, and suggest an adjustment if it overpromises.
Title: [YOUR DRAFT TITLE]
Episode summary: [PASTE A SHORT SUMMARY]
Format: one line verdict (accurate / overpromises), one sentence explaining why, one
adjusted title if needed.
Constraints: judge only against the summary provided.
Tone: direct, skeptical.
18. Title a series or multi-part episode consistently
Role: a podcast producer titling a multi-part series.
Task: given the topics of each part below, generate a consistent title format that signals
they're part of one series while distinguishing each part's specific topic.
Parts: [LIST EACH PART'S TOPIC]
Format: one title per part, sharing a consistent structure (e.g. a series name plus a
per-episode subtitle).
Constraints: keep the shared structure identical across all parts; vary only the part-specific
piece.
Tone: consistent, series-branded.
What to change: decide the shared structure before generating titles for parts you haven't recorded yet. Naming part one before you know the shape of part three tends to produce a format that doesn't actually fit everything that comes after it.
What Do You Do With Chapter Markers?
A quick chapter list is a five-minute pass over a transcript. The fuller system, timestamped chapters formatted for the platform you actually publish to, plus a proper show-notes bundle around them, deserves more room than one section here. For that, see our dedicated podcast show notes guide, which covers timestamped chapters, summaries, and the SEO fields in full. What follows is the quick version.
19. Quick chapter list from a timestamped transcript
Role: an editor building a basic chapter list.
Task: given the timestamped transcript below, identify 6-10 natural topic breaks and write
a short chapter title for each.
Timestamped transcript: [PASTE TRANSCRIPT WITH TIMESTAMPS]
Format: timestamp, then a chapter title under 8 words, one per line.
Constraints: only use timestamps that actually appear in the transcript; if the transcript
has no timestamps, say so instead of inventing them.
Tone: brief, navigational.
20. Turn a chapter list into a one-line episode description
Role: an editor writing a short episode description from a chapter list.
Task: given the chapter list below, write a one-sentence description of the episode that
captures its overall arc, not just a list of the chapter titles restated.
Chapter list: [PASTE THE CHAPTER LIST]
Format: one sentence, under 40 words.
Constraints: describe the arc, don't just concatenate the chapter titles.
Tone: inviting, not a table of contents restated as prose.
What to change: write this last, after the chapter list exists, not before. It reads better as a summary of a real structure than as a prediction of one, and it's easier to write honestly once you can see the whole shape of the episode.
How Do You Turn One Episode Into a Week of Social Content?
The transcript, the pull-quotes, and the clip list from the sections above are the raw material. Repurposing prompts turn them into the specific formats each platform actually wants, so one recording produces a week of posts instead of one clip and silence. For the fuller version of this idea applied beyond podcasts specifically, see our content repurposing workflow.
21. Turn a pull-quote into a text-and-image quote card brief
Role: a social media editor briefing a quote card graphic.
Task: given the pull-quote below, write the exact text to place on a quote card image
(the quote itself, shortened if needed, plus attribution) and a one-line visual direction.
Pull-quote: [PASTE THE QUOTE]
Format: card text (under 30 words), attribution line, one-line visual direction (mood,
color, style).
Constraints: the card text must be the guest's actual words or a clearly marked close
paraphrase, not an invented stronger version of what they said.
Tone: punchy, platform-native.
What to change: keep the card text shorter than you think it needs to be. Text that fits comfortably in an email pitch tends to overflow a square image, and a quote card with cramped text is harder to read at a glance than one with a shorter, punchier fragment of the same quote.
Once you have that brief, our GPT Image 2 prompt generator covers how to turn text-and-image briefs like this one into an actual prompt for generating the card, including how to get the text rendered cleanly inside the image.
22. Turn an episode into a text-thread outline
Role: a social media editor turning a podcast episode into a text-post thread.
Task: given the transcript summary below, outline a 6-8 post thread: a hook post, 4-6 posts
each covering one idea from the episode, and a closing post linking to the full episode.
Episode summary: [PASTE A SHORT SUMMARY OR THE TRANSCRIPT]
Format: numbered posts, each under the platform's typical character limit for a single post.
Constraints: each post should stand alone if someone only reads that one; don't require the
whole thread to make sense of any single post.
Tone: hook-first, conversational.
23. Turn an episode into one long-form post for a professional network
Role: someone writing a long-form post about a podcast episode for a professional network.
Task: given the transcript summary below, write a post that shares the single most useful
takeaway from the episode, in my own voice, with a link to the full episode at the end.
Episode summary: [PASTE A SHORT SUMMARY]
My take on the topic: [YOUR OWN REACTION OR ADDITIONAL CONTEXT, OPTIONAL]
Format: under 200 words, one clear takeaway, one link.
Constraints: the takeaway must come from something actually said in the episode, not a
generic industry observation the episode happens to be adjacent to.
Tone: personal, first-person, not a press release.
What to change: fill in "my take" even briefly. A post that's entirely a paraphrase of the episode with no added perspective reads as a summary of someone else's work, not a post from you.
How Do You Write a Short Summary Without Writing an Entire Show-Notes Page?
Sometimes you need a two-sentence blurb for a newsletter or a podcast app's episode card, not the full show-notes bundle. This is the one prompt on this page that overlaps with our dedicated show notes guide: use this for the quick version, that guide for the full system.
24. Short episode summary for a directory listing
Role: an editor writing a short episode summary for a podcast directory or newsletter.
Task: given the transcript summary below, write a summary a reader would see before
clicking play, not a recap for someone who already listened.
Episode summary: [PASTE A SHORT SUMMARY OR THE TRANSCRIPT]
Format: 50-70 words, ends with a reason to click play, not a spoiler of the ending.
Constraints: describe only what's actually discussed; don't tease a claim the episode
doesn't make.
Tone: inviting, specific to this episode, not a generic podcast-blurb template.
What to change: write two versions if your directory and your newsletter have different length limits, rather than forcing one blurb to fit both. A summary trimmed to fit a shorter field usually loses its ending, which is the part doing the actual inviting.
How Do You Keep AI From Inventing Details About a Guest or an Episode?
Every prompt above depends on one habit: the model only knows what you paste in, and asking it to fill a gap with something plausible is exactly how a guest's job title gets misstated or a quote gets attributed to the wrong speaker. The fix is the same one that matters for any high-stakes AI output: build the check into the prompt itself, not into hoping you'll notice afterward.
25. Fact-check a generated brief or summary against its own source
Role: a reviewer checking a generated brief against its source material.
Task: given the brief below and the source material it was built from, mark any claim in
the brief that isn't directly supported by the source material as "unsupported."
Brief: [PASTE THE GENERATED BRIEF, SUMMARY, OR CAPTION]
Source material: [PASTE WHATEVER YOU ORIGINALLY GAVE THE MODEL]
Format: the brief broken into claims, each tagged [SUPPORTED] or [UNSUPPORTED] with a
one-line reason.
Constraints: be strict; a detail that sounds right but isn't actually in the source counts
as unsupported.
Tone: skeptical, checklist-style.
What to change: run this on the outputs you're about to publish, not on every draft along the way. It's a final gate, and running it on every intermediate version turns a two-minute check into busywork that gets skipped once you're behind schedule.
Does the Free Plan Cover This?
Mostly, yes, for a single episode at a time. Every prompt above is plain text that runs in ChatGPT, Claude, or Gemini exactly as written, with nothing to install. Prompt Architects is an optional layer on top: a Chrome extension that turns the role/task/format/constraints/tone structure each prompt uses into a sidebar inside the chat you already have open, so filling in brackets doesn't mean digging through a notes app between takes.
The Free plan includes 5 architected prompt generations a day, forever, no card required, per the FAQ page, which is enough to run several of the prompts above on one episode. What Free doesn't include: saving your own versions of these as reusable templates with tags and search, so you're not rebuilding the guest-brief or question-ladder prompt from scratch every week, or a personal context library where your show's name, format, and typical episode length would sit once. Those are Pro-plan features, priced at the time of writing on /pricing; check the current figure there, since it's running a launch discount that won't last. For a weekly show recording one to two episodes, Free may cover a fair amount of the pipeline above. A daily or multi-show operation will hit the cap fast.
Stop rewriting prompts. Start shipping.
Works with ChatGPT, Claude, Gemini, Grok, Midjourney, Ideogram, Veo3 & Kling. 4.8★ on the Chrome Web Store.
Create An AccountNone of this records, edits, or uploads anything. What it does is take the parts of production that are mostly typing (the brief, the pitch, the questions, the cleanup, the title, the caption) and get a usable first draft of each one out of a transcript and some notes, so the time you actually spend on an episode goes to the conversation and the edit, not the paperwork around them.