Best ChatGPT Prompts for Podcast Hosts in 2026

**The prompt:** ``` You are a senior podcast researcher prepping me for a 45-minute interview. Guest: [name + role + company] My show: [1-sentence positioning] My audience: [who they are + what they care about] I'm pasting in (1) the guest's LinkedIn About + last 10 posts, (2) the relevant 10-K section if public-co, (3) their last 3 podcast appearances summarized. Output 4 things in order: 1. The 3 contrarian opinions this guest is on record holding (with source line). 2. The 5 questions they've answered 10+ times already — DO NOT ASK THESE. 3. The 3 unasked questions a sharp host would ask given (1) and (2). 4. One concrete moment in their career that no other host has dug into — and the question that opens it. ```

**Why it works:** Most guest-research prompts produce a Wikipedia summary. This one frames the model as adversarial to obvious questions and forces it to flag what's been asked-to-death. The 'with source line' constraint anchors contrarian opinions to evidence, cutting hallucination materially per OpenAI's prompt engineering guide.

**Sample output:** *'Contrarian opinion #1: Guest argues B2B SaaS valuations are 3-5× too high — source: LinkedIn post 2026-04-12. Already-asked: founding story (My First Million 2025-09), Series B (20VC 2025-11). Unasked: their position change on remote work — the 10-K shows lease costs jumped 40%, but they still say 'remote-first.''*

**The prompt:** ``` Build me a 45-minute interview outline for [guest] on [topic]. Constraints: - 4 acts of ~11 minutes each. - Act 1: a hook moment, not a bio recap. Open with a specific question that drops the listener mid-story. - Act 2: the guest's actual expertise — 3 questions with follow-ups designed to extract a counterintuitive insight. - Act 3: a friction point. One question the guest probably doesn't want to answer. Frame it kindly but ask it. - Act 4: a forward-looking question + concrete CTA. For each question, give me: the question itself, the ONE follow-up if they give a short answer, and the pivot if they go off-topic. ```

**Why it works:** Standard outline prompts return a flat list of 15 questions. This imposes an act structure (Terry Gross, Tim Ferriss, Lex Fridman all use variants) that prevents minute-25 fade. The follow-up and pivot columns are what hosts forget to prep and improvise badly.

**Sample output:** *'Act 1 hook: 'In 2023 you fired your entire sales team in a week. What were you seeing in the pipeline data nobody else saw?' Follow-up if short: 'Walk me through the conversation with your VP of Sales the morning you decided.' Pivot if off-topic: 'Hold that — let's come back to macro after we get what triggered the decision.'*

**The prompt:** ``` Write me 8 opening hooks (15-30 seconds spoken, so ~40-80 words each) for an episode where [guest] discusses [topic]. The 8 variants must use these distinct openings: 1. Cold open with a tape clip from later in the interview (you write what the clip says) 2. A specific data point that contradicts conventional wisdom 3. A personal story from me (the host) about why this guest matters 4. A question I'll answer in the episode (not a rhetorical one) 5. A 'this episode is for you if...' segmentation hook 6. A pop-culture or news-event tie-in (timely as of June 2026) 7. A controversial statement the guest will defend in the episode 8. A 'here's what you'll learn' three-bullet promise Keep each under 80 words. Make them sound spoken, not written. ```

**Why it works:** Hosts default to the same opening every time. Forcing 8 structurally distinct openings exposes options they wouldn't have considered. Spotify for Podcasters' guide notes the first 30 seconds drive most of the drop-off decision.

**Sample output:** *'Variant #2 (data contradiction): 'Conventional wisdom says you need three years of runway to survive a recession. My guest today did it on six weeks of cash. He'll explain how — and why the three-year rule is wrong.''*

**The prompt:** ``` You're cleaning a podcast transcript for publication. Below is raw ASR output from [tool — Descript / Riverside / Otter]. Clean it as follows: 1. Label speakers as HOST and GUEST based on conversational context (introductions, who asks vs. answers). 2. Remove filler: 'um', 'uh', 'like' (when used as filler), 'you know' (when used as filler). 3. Keep filler when it's load-bearing (e.g., 'you know' before a vulnerable moment). 4. Fix obvious ASR errors when you have ≥80% confidence (mark <80% confidence as [unclear: best-guess]). 5. Break paragraphs every 3-5 sentences for web readability. 6. Preserve every substantive word — DO NOT summarize or paraphrase. Output the cleaned transcript. Then output a separate 'ASR errors I corrected' list so I can verify. ```

**Why it works:** 'Clean this up' silently paraphrases — destroying SEO value and accuracy. Explicit 'DO NOT summarize' + 'output the errors I corrected' creates an audit trail. The [unclear] marker catches hallucinated cleanup.

**Sample output:** *'HOST: Walk me through March. GUEST: March was the inflection. We had — I think it was like — [unclear: $4.2M / $4.6M] in the bank, burning $800K a month. Errors corrected: 'ate hundred thousand' → '$800K' (high confidence).'*

**The prompt:** ``` Draft show notes from this transcript. Format: [Top: 2-sentence episode summary, written to make someone click 'play' — not summarize what happened.] [Body: 6-10 timestamped chapter markers. Format: '[MM:SS] Chapter title (5-8 words). One-line description of what's discussed.' Title each chapter as a SPECIFIC THING (not 'Background').] [Bottom: 5-7 bullet points of the most quotable / shareable lines from the guest, with timestamps.] [Then: a 'Mentioned in this episode' list of every book, company, person, and URL referenced — with the spelling correct.] Do not invent timestamps. If you're unsure of a timestamp, write [~MM:SS]. ```

**Why it works:** Timestamp accuracy is where most AI show notes fall apart. The '~' uncertainty marker + 'Mentioned in this episode' with correct spelling solves the two complaints listeners have. Podtrac's 2025 industry report shows episodes with chapter markers get higher completion rates — show notes aren't just discovery, they're retention.

**Sample output:** *'[02:14] How a Series A board meeting went sideways. [12:47] The pricing change that doubled MRR. Quotable: '[15:33] We were optimizing for the wrong metric for 18 months.''*

**The prompt:** ``` Generate 12 podcast episode title variants for an episode where [guest] discusses [topic]. Constraints: - Variants 1-3: curiosity-gap titles (something the listener wants to know that the title doesn't answer) - Variants 4-6: specific-promise titles ('How [guest] [did specific thing] in [specific time]') - Variants 7-9: contrarian titles (state a position the guest holds that contradicts conventional wisdom) - Variants 10-12: number/list titles ('3 [things] [guest] [verb]ed before [outcome]') Rules: - 55-70 characters per title (fits Spotify + Apple Podcasts display) - Guest's name in titles 4-12 only (1-3 use curiosity, no name) - No clickbait that the episode doesn't deliver on - Output as a table with: variant #, title, character count, hook type, predicted click-through reasoning (1 sentence) ```

**Why it works:** Most title prompts return 10 similar variants. Forcing 4 distinct hook types produces real A/B tests, not synonym permutations. Character counts matter — Apple Podcasts truncates titles around 60-70 chars in iOS.

**Sample output:** *'#3 (curiosity): 'The Email That Cost a SaaS Founder $4M' (40 chars). #7 (contrarian): 'Why Maria Chen Thinks Product-Market Fit Is a Myth' (52 chars).'*

**The prompt:** ``` Find 10 social clips inside this transcript. Each clip must be: - 45-90 seconds when read at conversational pace (so ~110-220 words of guest dialogue) - A self-contained moment — listener doesn't need context outside the clip to get it - Ranked by one of these hooks: surprising stat, contrarian take, specific story moment, vulnerable admission, useful framework, useful tactic Format each clip as: CLIP #N — [hook type] [Timestamp range MM:SS – MM:SS] Full verbatim text of clip WHY IT'S A CLIP: 1 sentence on what makes it work SUGGESTED PLATFORMS: Twitter / LinkedIn / TikTok / Instagram Reels — and which 1-2 are best fit SUGGESTED CAPTION: 1 sentence written to make the clip click Do not invent dialogue. If you can't find 10 strong clips, output fewer with a note. ```

**Why it works:** Generic 'find clips' returns surface highlights. The 6 hook types force structurally different clips for different platforms — vulnerable admissions for TikTok, contrarian takes for Twitter, frameworks for LinkedIn. 'Do not invent dialogue' stops the model from synthesizing clips that didn't happen.

**Sample output:** *'CLIP #2 — vulnerable admission [18:42 – 19:51]. 'I almost shut the company down in November 2024...' WHY: First public admission of how close they came. PLATFORMS: LinkedIn, TikTok. CAPTION: 'The founder who almost gave up — and what changed his mind.''*

**The prompt:** ``` Write 3 variants of a 60-second host-read ad for [sponsor]. My voice: [paste 3 paragraphs from my actual show — opening monologues, not interview questions] Sponsor's product: [1-paragraph description from their brief] Sponsor's required talking points: [list, verbatim] The one thing the sponsor most wants the listener to do: [single action] My audience's biggest objection to this category of product: [from your knowledge of your audience] Write 3 distinct variants: 1. Story-led (open with a personal anecdote that lands the product) 2. Problem-led (open with the audience's pain, then bridge to the product) 3. Curiosity-led (open with a question or counterintuitive claim) Each 60 seconds spoken = ~150-180 words. Each must include all required talking points. Each must sound like ME, not like a CMO wrote it. ```

**Why it works:** Naive sponsor reads sound robotic because the model doesn't have the host's voice. Pasting actual show monologues + naming the audience's objection produces ads that match the show's tone. The 3 variants give the sponsor an A/B test.

**Sample output:** *'Variant 1 (story-led): 'A few weeks back I was prepping for an episode at midnight, panicking because the guest had moved our slot. I opened [sponsor] and...''*

**The prompt:** ``` Below are [N] listener questions submitted for the next Q&A episode. Triage them into 4 buckets: BUCKET A (answer in episode): 5-7 questions that are (a) specific enough to give a real answer, (b) interesting enough to be valuable to listeners beyond the asker, (c) something I can genuinely answer with expertise. BUCKET B (answer in newsletter or short-form): 5-10 questions that have shorter answers and don't need a podcast format. BUCKET C (decline politely): questions that are out of my expertise, repeats of past episode topics (flag which past episode), or too vague to answer. BUCKET D (great follow-up question we should DM the asker about): questions that would be a full guest interview if explored properly. For each Bucket A question, draft a 1-sentence note on the angle I should take when answering — not the answer itself, just the framing. ```

**Why it works:** Hosts either answer every question (boring) or randomly pick favorites (misses gems). The 4-bucket triage makes curation visible. Bucket D is the secret — turning a listener question into a future guest interview is the highest-leverage inbound move.

**Sample output:** *'Bucket A #3: 'LinkedIn vs. TikTok clip decision?' — Angle: walk through your 2-question filter. Bucket D #1: 'How did you negotiate your distribution deal?' — DM Jess about coming on the show; this is a full episode.'*

**The prompt:** ``` Design a 12-episode podcast season with a clear thematic arc. Show positioning: [1 sentence] Season theme: [1 sentence — the question this season answers] My target audience for this season: [who you most want to reach] Guests I've already booked / can probably book: [list] What I want listeners to believe / do by the end of the season: [the transformation] Output: 1. A season-arc story (3 acts across 12 episodes — what shifts in the listener's understanding from episode 1 → 4 → 8 → 12). 2. For each of the 12 episodes: episode title, guest profile (or solo), the question that episode answers, and how it advances the arc. 3. The 'connective tissue' — 3-4 callbacks or running threads I should reference across episodes to make the season feel like one story. 4. The 2 episodes most at risk of feeling out-of-place — and how to reframe them to fit. ```

**Why it works:** Most podcasts are episodic with no season-level architecture. A thematic arc creates binge behavior and makes finales land. Connective-tissue callbacks are what showrunners use; the 'episodes at risk' diagnostic catches problems before recording.

**Sample output:** *'Act 1 (eps 1-4): listener believes 'PMF is a phase you finish.' Act 2 (eps 5-8): you complicate that with founders whose PMF degraded. Act 3 (eps 9-12): resolve into 'PMF is a continuous practice.' Connective tissue: every guest answers 'when did you know you were losing fit?''*

**The prompt:** ``` I'm pasting in (1) the drop-off curve from [Spotify for Podcasters / Apple Podcasts Connect / Chartable] for my last 6 episodes — minute-by-minute retention %, (2) the chapter markers / show notes for each episode. Do this: 1. For each episode, identify the 1-2 minutes with the steepest drop-off relative to the episode's baseline. 2. For each drop-off point, look at the chapter / topic at that minute and propose 3 hypotheses for why listeners dropped (e.g., topic shift felt unrelated, host stopped asking questions, energy dipped, ad placement, etc.). 3. Rank the hypotheses by plausibility given the pattern across all 6 episodes. 4. Propose 3 testable changes for the next 6 episodes — specific edits I can make, not vague advice like 'be more engaging.' 5. Flag if any drop-off is normal (e.g., the natural drop-off curve from Spotify's average podcast retention data) vs. a real signal. ```

**Why it works:** Hosts look at drop-off curves and feel bad. Structured hypothesis + testable changes turns analytics into edits. The 'normal vs. real signal' flag prevents fixing a drop-off that's just baseline platform retention.

**Sample output:** *'Ep 47 dropped 11% between 14:00 and 15:30. Hypothesis: topic shifted from personal story to industry trend — listeners came for the story and bailed when it generalized. Testable change: bridge specific → general with a one-sentence callback to the story you just told.'*

**The prompt:** ``` Write a cross-promo pitch email to [target podcast]. My show: [name, 1-sentence positioning, audience size, audience description] Their show: [name, what you actually like about it — specific episode you'd reference] The cross-promo I'm proposing: [trailer swap / guest swap / co-promoted episode / joint live event] What's in it for them: [specific — don't write 'mutual benefit'] Write a 5-paragraph email: 1. Opening that proves I actually listen (reference a specific episode + a specific moment in it) 2. 1-sentence positioning of my show with audience size 3. The specific cross-promo proposal — what we'd each do 4. Why my audience is good for them (specifics: demographics, behavior, why they'd like the target's show) 5. Low-friction CTA — propose 1 specific next step, not 'let me know if you're interested' Keep total under 200 words. No flattery beyond the specific reference in paragraph 1. ```

**Why it works:** Cross-promo pitches fail because they're generic. Proving you listen + a specific moment + a low-friction CTA converts. Trailer swaps with similarly-sized shows remain the highest-ROI organic growth tactic per Edison Research's listener-acquisition findings.

**Sample output:** *'Hey Jamie — your June 4 episode with Priya, minute 22 where you pushed back on her 'do less' framing — cleanest interview move I've heard this year. Most hosts would have agreed and moved on. I run [show], 18K weekly listeners, B2B SaaS founders mostly...'*

**Transcript source:** Descript and Riverside produce ASR transcripts you can paste into Prompts 4-7 and 11. Hindenburg PRO added transcription in 2025.

**ChatGPT tier:** Free works for Prompts 1-3, 6, 8, 9, 12. For Prompts 4, 5, 7, 10, 11 you'll want ChatGPT Plus or API access — transcript + analytics context exceeds free-tier limits. Per OpenAI's pricing page, GPT-5 API costs are low enough that running these at scale is functionally free relative to time saved.