TL;DR
Feed an AI model a one-paragraph pitch plus your guest’s real bio, and it returns a timed episode skeleton in about a minute: a 60-second cold-open hook, five segments with time markers, transition lines, and ad-break placements. Use the prompt below in ChatGPT (GPT-5.5), Claude (Sonnet 4.6), or Gemini (3.1 Pro) — any of the three handle this well. Then do the two things AI cannot: verify every guest fact, and add the two or three questions sharp enough that you are slightly nervous to ask them.
Why an outline protects your numbers
A 35-minute conversation that wanders into 65 minutes does not just bore people — it gets punished by the algorithm. Apple and Spotify both rank on completion rate, the share of an episode a listener actually finishes. Two numbers explain why the opening matters most:
- Podcasts lose 20-35% of listeners in the first five minutes of an episode (industry retention data, 2026).
- Over 40% of listeners decide within the first 15 minutes whether they stay or leave.
Episode length tracks the same way. Shows in the 20-40 minute band post the highest completion rates (around 85%), while episodes over 60 minutes fall to roughly 60%. The average episode in 2026 runs about 41 minutes, and over 70% of listeners finish most or all of a typical episode — but only when the structure earns it. An outline with a hard-hitting cold open and timed segments is the cheapest way to defend those numbers before you record a single word.
When AI is the right tool here
- You have a confirmed topic and a guest (or a solo episode you’ll carry yourself).
- You know the audience: their knowledge level and what they tune in for.
- You publish on a cadence and want a repeatable template, not a blank page each week.
AI is a structuring tool, not a research source. It will not know your guest, and it writes polite, low-friction questions by default. The two sections below are the human work that the model cannot do for you.
When not to rely on AI alone
AI does not know your guest. Without source material it will invent biographical detail and inflate credentials. Paste the guest’s actual bio and a link or two to recent work, and never publish an AI-written guest description without the guest’s sign-off.
It also defaults to safe questions. The questions that make an episode worth listening to are usually the ones that make the host slightly nervous to ask. Write at least two of those yourself; the model is there to build the scaffold, not the spine.
What to feed the model
- Episode topic in two sentences
- Guest bio, current role, and one or two recent things they shipped or said (paste the real text)
- Audience description: knowledge level, and what they come to your show for
- Target episode length (aim for the 30-40 minute sweet spot unless your genre runs long)
- House style: tone, signature segments, and where ads go
Copy-ready prompt
Paste this into ChatGPT, Claude, or Gemini and fill the bracketed fields. Keep the brackets — they tell the model exactly what to substitute.
Outline a podcast episode. Be specific; do not invent facts about the guest.
Topic: [two-sentence topic]
Guest: [paste guest bio]
Recent guest work: [paste links or notes]
Audience: [knowledge level + what they come for]
Target length: [minutes] minutes
House style notes: [tone, signature segments, ad placement]
Output the outline as:
1. Cold-open hook (30-60 seconds): one provocative line plus the
promise of the episode. Make it land in the first 30 seconds.
2. Intro and setup (90 seconds - 2 minutes): why this guest, why now.
3. Five segments with time markers totaling the target length. Each:
- Segment title
- Time range (e.g. 8:00-15:00)
- 3-5 questions, increasing in depth
- One transition line into the next segment
4. Wrap (last 2-3 minutes): one reflection question tied to the
episode's argument, plus a listener call to action.
5. Ad-break suggestions: place one mid-roll after a natural segment
transition (mid-rolls earn the highest engagement and CPM), with
a one-line rationale for each break.
Then list 3 "uncomfortable but worth asking" questions I should
consider adding.Which model to use
All three flagship chat models produce a usable outline from this prompt. The differences are small but real, as of June 2026:
| Tool | Model | Plan / price | Best at | On-site guide | | --- | --- | --- | --- | | ChatGPT | GPT-5.5 | Free $0 / Plus $20/mo | Fast ideation, alternate angles | YouTube script with AI | | Claude | Sonnet 4.6 | Free (limited) / Pro $20/mo | Natural prose, tighter questions | Blog outline with AI | | Gemini | 3.1 Pro | AI Pro $19.99/mo | Ingesting long transcripts / prep docs | Livestream script with AI |
If you record video and have hours of prior episodes or a long prep transcript, Gemini 3.1 Pro’s 1M-token context is the practical pick for “read all of this, then outline.” For the cleanest question phrasing on the first try, Claude Sonnet 4.6 tends to win. For most weekly shows, whichever one you already pay for is fine.
The arc the model should produce
- Cold-open hook
- Intro / setup
- Segment 1: context and credibility
- Segment 2: the central topic
- Segment 3: the counterintuitive angle
- Segment 4: tactical / how-to detail
- Segment 5: forward-looking
- Wrap and call to action
This is a documentary arc: establish, deepen, complicate, ground, project. If your draft outline does not move through those five beats, it is five versions of the same conversation.
How to check the output
- Read the segment titles aloud in order. Do they tell a story? If not, reorder.
- Cut the weakest segment. If the outline still holds, kill that segment for real.
- Check time markers — most segments should land at 6-10 minutes, and the five should total your target length.
- Confirm a mid-roll sits after a segment transition, not mid-thought. Mid-rolls command 30-40% higher CPMs than pre-roll, so place them where attention is highest but the flow has a natural seam.
- Verify every guest fact against their public bio. Treat anything the model wrote about the guest as a claim to check, not a fact.
Common mistakes
- No segment timings, so the conversation drifts past the 60-minute completion cliff.
- No transition lines, so cuts feel jarring in post.
- Five segments that restate the same idea in different words.
- A generic wrap (“any final thoughts?”) instead of one question tied to the episode’s argument — wasting the most attentive minutes of the show.
- An ad break dropped mid-sentence, which spikes the drop-off you were trying to avoid.
Iterate from your retention curve
After 3-5 episodes, open the drop-off (retention) curve in Apple Podcasts Connect or Spotify for Creators. The spot where listeners consistently exit is usually where a segment was weak or where two segments overlapped. Move that beat earlier, tighten it, or cut it — then update your template so the next outline starts from what already works.
FAQ
- How long should the episode be? Target 30-40 minutes for most shows; that band posts the highest completion rates (~85%). Go longer only if your genre and audience expect it — episodes over 60 minutes drop to roughly 60% completion.
- How many questions per segment? Three to five. More turns an interview into an interrogation and blows your time markers.
- Should I share the outline with the guest? Share the segment titles and themes, not the exact questions. A little surprise produces sharper answers.
- Can I reuse one outline across guests? Reuse the structure; never the questions. Questions must be guest-specific or they sound canned.
- Where should the ad break go? One mid-roll after a natural segment transition, ideally near the middle of the episode. Mid-rolls earn the highest engagement and CPM; never drop one mid-thought.
- Which AI should I use? Any of ChatGPT (GPT-5.5), Claude (Sonnet 4.6), or Gemini (3.1 Pro). Use Gemini when you want it to read a long prep transcript first; use Claude for the tightest question wording.
Related
- Prompt library: PPT outline prompts
- Prompt library: tutorial outline prompts
- Prompt library: blog outline prompts
- YouTube script with AI
- Livestream script with AI
Tags: #AI writing #Content creation