At a glance
Podcast production has 6 chokepoints: planning an episode worth recording, researching the guest deeply enough to ask the question nobody else asks, getting the right guest to say yes, writing show notes that earn the click, turning each episode into clips that travel, and closing the post-mortem loop so episode 50 sounds better than episode 5. The 7 prompts below cover each one. AI handles the synthesis. You handle the curiosity, the conversation, and the taste.
Why do most podcasts plateau around episode 12?
A podcast that survives past episode 12 has solved a different problem than a podcast that launches. The launch problem is the show concept. The plateau problem is production load. Each episode is roughly 8-12 hours of work (guest research, booking, recording, editing, show notes, distribution, social), and hosts who do not systematize the load burn out around episode 12. The 7 prompts below are designed to compress the parts AI handles well (the research, the structuring, the show-notes writing, the clip-finding, the distribution drafts) so the host can spend energy on the parts that matter (the conversation, the taste, the next guest).
One ground rule: AI should never go on the show in your voice. Hosts who use AI to generate intro monologues, ad reads, or guest questions get caught quickly; the listener can hear the AI cadence, and the trust that podcasting earns over hours collapses in seconds. The prompts here are for behind-the-scenes work; the on-mic work is yours.
⚠️ Two things AI cannot do for your show
AI cannot ask the follow-up question that surfaces what the guest has not said before; that comes from your curiosity, your listening, and your willingness to be surprised. AI cannot tell you which 15 seconds of a 90-minute conversation are the moment that matters; that comes from your taste. Everything else (research, structuring, drafting, repurposing) is fair game. For the pre-publish editing pass on any written material from these prompts: How to Edit AI Out of Your Writing.
What are the seven podcasting prompts?
The 7-stage production cycle
- Episode brief — what this episode is about + who it serves
- Guest research — deep dive into who they are + the question only you would ask
- The booking email — the cold reach to the desired guest
- Show notes draft — the post-episode structure that earns the click
- Clip extraction — surfacing the moments worth turning into shorts
- Distribution stack — one episode → five platform formats
- Post-mortem loop — what to do differently next time
1. Episode brief
The audience this show serves: [paste your real audience read: not "founders" but "first-time founders in B2B SaaS who are stuck on positioning"].
Why this episode now (timing context, recent industry events, audience demand signal): [paste].
What I want the listener to walk away with that they did not have at minute zero: [paste in one sentence].
Build the brief. Output:
– The 1-sentence headline (what the show notes title will be)
– The 3-5 questions the episode HAS to answer (not what the guest will be asked: what the audience will leave knowing)
– The single moment of tension or surprise I should try to create
– The 2-3 things this episode should NOT try to also do (scope discipline)
– The 1 follow-up resource the listener should leave with
Hard constraint: do not generate guest questions yet. The brief is what the episode is FOR, not what I will ask. Questions come in prompt 2 after I have done the research.
Why this works: Most podcast episodes fail because the host walked in without a brief. The conversation drifts; the listener cannot remember what the episode was about by the end of it. The "single moment of tension or surprise" output is the structural anchor that distinguishes a memorable episode from a forgettable one. The "questions come AFTER research, not now" constraint prevents the most common AI-misuse pattern, which is generating generic interview questions before the host knows what they actually want to learn.
2. Guest research + interview prep
Their public work I have access to (paste links): [LinkedIn, recent talks, podcasts they have been on, papers, posts].
The 3 most recent things they have publicly said: [paste].
The episode brief: [paste from prompt 1].
Build the prep document. Output:
– The 3-sentence biography I will read into the intro
– The 5 questions the guest has answered too many times already (avoid these)
– The 3 questions nobody else has asked that fit this guest's body of work
– The 1 question that might make them think for a second before answering (the goal-state question)
– The 2-3 backup directions if the conversation stalls
– The single fact-check item I should verify before recording (do not let me cite a metric I cannot defend)
Hard constraint: do not invent details about the guest. If I have not given you a source for a claim, do not include it.
Why this works: The "questions they have answered too many times" output is the political-judgment move that turns an interview into a conversation. The "question that might make them think for a second" output is the structural anchor for an episode that the guest will reference later; if the guest had to think, the listener noticed, and so did the guest. The "do not invent details" constraint is what stops AI from confidently citing a paper that does not exist.
3. The guest-booking email
The show in 2 sentences (audience, what it is for, why a smart guest should care): [paste].
Why I want THIS guest specifically (not a generic compliment: the actual reason): [paste].
The episode angle (so they know what they would be invited to talk about): [paste from prompt 1].
A reason they should believe the conversation will be good (recent guest, audience signal, format detail): [paste].
Draft the booking email:
– Opens with the specific real reason I want them (not flattery)
– States the episode angle in 1 sentence so they know what they would be talking about
– Names the format: length, recording date window, whether it is video or audio only
– Offers 2-3 dates rather than asking for theirs
– Includes one piece of social proof that is true (audience signal, past guest, where the show has been featured) without overstating
– 5-7 sentences total, professional warm, no exclamation marks
– Closes with a clear yes / no / suggest-another-time response path
Mark anywhere the email reads as templated. Suggest the human-edit that would tip it the other way.
Why this works: Booking emails fail when they generic-compliment the guest. The "actual reason I want them" constraint is the structural move that signals to the guest you have actually engaged with their work. The "offer 2-3 dates rather than ask for theirs" line is the friction-reducer that real bookers know; the guest is more likely to say yes when the answer is "Tuesday at 2" vs "what dates work for you?".
4. Show notes draft
The cleaned transcript: [paste].
The episode brief: [paste from prompt 1].
The guest's preferred call-to-action (their newsletter, their book, their company): [paste].
Build the show notes. Output:
– The title (under 65 chars, specific claim or specific question, not the guest's name + a colon)
– The 2-sentence subhead that earns the click
– The 5-7 timestamped highlights (with the kind of moment, not a summary of what was said: "[12:34] The number that surprised me" not "[12:34] They talked about revenue")
– The 3-bullet "here is what you will learn"
– The guest's links and CTA, written in the show's voice
– The 2-3 related episodes from this show or others worth recommending
– The single quotable line from the conversation that could go on social (with the timestamp)
Hard constraint: do not invent moments that did not happen in the transcript. Every highlight cites the timestamp it came from.
Why this works: Show notes are how listeners decide whether to press play. The "kind of moment, not a summary" framing for timestamps is the move that earns the click; "[12:34] The number that surprised me" is curiosity-shaped; "[12:34] They talked about revenue" is a generic table of contents. The single quotable line output is the input to prompt 6's distribution stack.
5. Clip extraction
Show notes highlights: [paste from prompt 4].
The 3-5 outcomes I want from clips (drive subs to the show / drive clicks to the guest's thing / build the show's own social presence / etc): [paste].
Find me 5-8 candidate clip moments. For each:
– The exact start and end timestamp (precise to the second)
– A 1-sentence description of WHY this is a clip moment (the surprise, the specific story, the contrarian claim, the vulnerable admission, the data point, etc)
– The platform it best fits (vertical short for Reels/TikTok/Shorts vs horizontal for YouTube vs static-with-quote for LinkedIn vs audio-only for podcast feeds)
– The hook line that should appear as the first 2 seconds of text overlay
– The single change a human editor should make to maximize the clip (cut the throat-clearing, tighten the silence, etc)
Mark which 1-2 clips are most likely to break out (lowest competition + highest specific-curiosity).
Why this works: Clip extraction is the most leveraged repurposing move in podcasting; one 90-minute conversation becomes 5-8 short pieces of content that travel further than the full episode ever will. The "WHY this is a clip moment" output is the part that teaches the host their own taste over time; reading those why-lines back to back across a season tells you what your show actually does well. The "most likely to break out" mark is the prioritization that prevents producing 8 clips that all underperform.
6. The distribution stack
– 1 LinkedIn post (the show's voice, opens with a specific moment or claim from the conversation, ends with the link to the episode)
– 1 X / Threads thread (5-7 posts, each one stands alone, last post links to the episode)
– 1 newsletter blurb (150 words, send-to-subscribers voice, includes the timestamp link to the single most-quotable moment)
– 1 Instagram carousel outline (5-6 slides, the strongest insight from the conversation, the final slide drives to the show)
– 1 guest-side asset (the LinkedIn post the guest could repost, written in the guest's style as much as possible from their public writing, so they can copy-paste-edit and ship in 90 seconds)
For each, mark which clip from prompt 5 pairs with it. The clip + the post launched together outperforms either alone.
Hard constraint: no rhetorical-question openers. No "here's what I learned." No em-dashes in every paragraph.
Why this works: The single most underused move in podcasting is the guest-side asset: a piece of content the guest can ship from their own accounts. Hosts who provide this get 2-3x the audience overlap because the guest's audience sees the show in a context that respects their voice. The "clip + post launched together" pairing is the distribution structure that makes social actually convert to plays.
7. The post-mortem improvement loop
Performance data so far (1 week minimum): [downloads / plays, social signal, listen-through rate if available, guest's post performance, audience comments].
What surprised me about the data (positive or negative): [paste].
What I knew going in vs what I learned: [paste].
Run the analysis. Output:
– The 1-2 things that worked, with the structural reason WHY (so I can do it again)
– The 1-2 things that did not work, with the structural reason WHY
– The single change I should make to the NEXT episode based on this data
– The 1 prompt above (1-6) I should adjust based on what I learned
– The note for the running show-craft log (what to remember in 6 months)
Hard constraint: do not generalize from 1 episode. If the data does not support a confident conclusion, say so directly and recommend waiting for 2-3 more episodes of pattern.
Why this works: The post-mortem is what turns a podcast into a craft. Most hosts skip it because the episode is already done and the next one needs producing. The "1 prompt above I should adjust" output is the recursive improvement that makes episode 50 actually better than episode 5. The "do not generalize from 1 episode" constraint is the statistical discipline that prevents overreacting to single-episode noise.
What is the worst thing you can do with AI in podcasting?
- Use AI-generated questions in the actual interview. Generic questions get generic answers. The 3 questions nobody else has asked come from your research; AI surfaces patterns, you ask the question.
- Let AI write the show notes without referencing the transcript. Without the transcript, AI invents quotes and moments. Every highlight needs a timestamp.
- Skip the post-mortem. Prompt 7 is the recursive improvement. Without it, episode 50 sounds like episode 5.
- Use AI for ad reads or sponsorship copy. Listeners forgive a lot in a podcast; AI-cadence ad reads are not one of those things. Write your own ad copy in your voice.
- Produce clips that all look the same. Five clips with the same template, the same hook style, the same overlay font, the same captions all underperform because they pattern-match as content-farm output. Vary deliberately.
What if you publish weekly?
The 7 prompts are a strong ladder candidate for weekly publishers. Save each as a Claude skill. Bundle as podcast-stack. Wire the workflow: the brief prompt runs from a calendar event, the research prompt runs when you add a guest, the show-notes + clip prompts run from the transcript-upload, the distribution stack lands in your social-scheduler queue. Setup: a Saturday morning. Time saved per episode: 2-4 hours. After 5 episodes the system has paid for itself. The qualitative gain compounds: every episode benefits from prompt 7's recursive improvement.
📊 The Prompt-to-Workflow Ladder
Tier 1: the prompts (this post). Tier 2: the skill (one per production stage). Tier 3: the plugin (podcast-stack bundle). Tier 4: the workflow (full episode pipeline auto-triggered from the recording). When to climb →
What are common questions about AI in podcast production?
Should I disclose AI use in production?
For behind-the-scenes work (research, show notes, clip extraction, distribution drafts), no broad mandate. For anything that ends up on-mic in the show itself, disclose. The line is: listener-facing voice content needs disclosure; production scaffolding does not.
Which transcription tool should I use?
Otter.ai, Descript, Rev, Riverside, and Squadcast all produce usable transcripts. Descript and Riverside also handle the editing flow if you record in their tools. The quality is roughly comparable; the differentiator is the editing workflow + whether you want speaker diarization. Whatever you pick, run a cleanup pass (similar to prompt 2 of our Meeting Notes guide) before the downstream prompts.
Can AI host the show for me?
Technically yes; practically no. AI-hosted shows in 2026 still get caught quickly by listeners. The trust podcasting earns is voice trust; AI cannot replicate that yet. AI as a co-pilot for production: excellent. AI as the host: trust-collapse waiting to happen.
What about AI voices for sponsorship reads?
Don't. Sponsorship reads in your real voice are part of how listeners feel they know you. Outsourcing them to an AI voice clone in 2026 is the move that costs you the trust podcasting was earning. Some networks are testing it; the data so far suggests engagement drops on AI-read sponsor segments.
Where do these prompts come from?
They are the podcast section of the larger AI Prompt Library. The Library has over 500 prompts across 33+ categories with three difficulty levels, including the full content-creation stack: podcasts, blog posts, newsletters, video scripts, social media, and the longer-form pieces like book chapters and course modules.
Sources to read next?
- Edison Research Podcast Consumer Report (the standard industry data)
- Podnews (daily industry briefing)
- Descript blog (production craft + AI tools)
- Riverside blog (host research + production)
- Podcast (Grokipedia)
✏️ Before show notes / clips ship
The editing pass on show-notes copy and clip captions is the difference between a post that earns the click and a post that pattern-matches as AI-generated to your listeners' eyes: How to Edit AI Out of Your Writing → The full 29-pattern catalog is documented there, along with the open-source Humanizer skill that runs the list on show notes, clip captions, and social drafts.
The AI Prompt Library · $39
over 500 tested prompts including the full content-creation stack
The seven podcast prompts above are a free preview. The Library has the content-creation stack: blog posts, newsletters, video scripts, social drafts, book chapters, course modules, and the conference-keynote framework.
1-on-1 Deep Work Session with James · $175
Build the podcast-stack on your actual show
A focused 2-hour session. We run prompts 1, 2, and 4 on your next episode and save the system as a reusable skill set. You walk out with the production workflow that takes a 12-hour episode to 4 hours.
Get Smarter About AI Every Morning
Free daily newsletter. Built for people who want to use AI well, not chase every model.
Free forever. Unsubscribe anytime.
You may also like
- Best AI Prompts for Blog Writing — long-form that earns the click
- Best AI Prompts for Marketing — campaigns from voice to distribution
- Best AI Prompts for LinkedIn — the surface that fills your inbox
- How to Edit AI Out of Your Writing — strip AI patterns before you publish
- Prompt to Workflow: The AI Ladder — when to climb from prompt to workflow