One podcast episode produces 30 LinkedIn posts using a 4-step AI workflow: transcribe, extract, structure, voice-match. Full walkthrough with prompts, tool recommendations, and the discipline that makes it work. Total time: 90 minutes per episode.
Four steps. Transcribe the episode (5-10 min). Extract 30 distinct ideas with Claude (15-20 min). Structure each as a LinkedIn post using your voice prompt (30-45 min). Voice-match edit + schedule across 6-10 weeks (15-30 min). 90 minutes total. Requires a transcription tool plus a properly-built voice prompt — without the voice prompt, the output is generic and not worth shipping.
One hour of recorded conversation typically contains 30-50 distinct insights, framings, statistics, stories, and contrarian takes. The host doesn't realise this in real time — episodes feel like one continuous discussion. But when you transcribe and extract, the ideas separate cleanly.
That's the structural opportunity. Instead of writing 30 LinkedIn posts from scratch (10-30 hours of work), you spend 90 minutes per episode and ship 30 posts across 6-10 weeks. The math gets ridiculous: a weekly podcast produces enough LinkedIn content for the year by month three.
The reason most podcasters don't do this is the workflow. They've tried generic AI tools, gotten generic-sounding posts, and concluded the approach doesn't work. The approach works — the failure mode is missing the voice prompt layer.
STEP 1 · 5-10 MIN
Upload the audio to your transcription tool. Wait for the transcript. Most tools produce timestamped transcripts in under 10 minutes for a 60-minute episode.
Output you want: a clean text transcript, ideally with speaker labels if it's an interview podcast. Don't bother with the AI-generated summaries the tools sometimes provide — they're optimised for podcast show notes, not for content extraction.
STEP 2 · 15-20 MIN
Open your Claude Project (with voice prompt loaded). Paste the full transcript. Then run this extraction prompt:
Claude returns 30 ideas. Read through them. Cut any that don't feel strong enough — typically 5-8 will get cut on first review. You'll end up with 22-25 strong ideas, which is fine. Quality beats hitting the 30 number.
STEP 3 · 30-45 MIN
For each strong idea, run this prompt in your Claude Project:
Claude produces a draft. Read it. If voice match is weak, run a follow-up: "Tighten this draft against my voice rules. Specifically check: sentence length variation, banned words, opening with reader's pain not solution."
Move to the next idea. Each post takes 1-2 minutes including the iteration. 25 strong ideas × 1.5 min = 37 minutes total.
STEP 4 · 15-30 MIN
Read each draft sentence by sentence. Tighten anything that drifted from voice. Common edits:
Load the cleaned drafts into Buffer or your scheduling tool. Spread across 6-10 weeks at 3-5 posts per week. Mix podcast-derived posts with original observations and other content types.
Failure mode 1: skipping the voice prompt layer. Most podcasters who try AI repurposing skip building a voice prompt and run the extraction + structuring with default AI prompts. Output reads as generic "podcast clip" content. Underperforms. They conclude the approach doesn't work. The voice prompt is non-negotiable.
Failure mode 2: using Castmagic or Descript's built-in social outputs without voice match. These tools produce social-ready content automatically, but voice match is bounded by the templated brand voice features. The output is faster but worse. Use these tools for transcription, not for the social output stage.
Failure mode 3: posting all 30 in one week. LinkedIn's algorithm penalises accounts whose recent posts share the same source pattern. 30 podcast-derived posts in 7 days triggers low-quality signals and reach drops. Spread across 6-10 weeks.
Failure mode 4: not editing. Even with a strong voice prompt, the first draft of any AI post needs a 2-minute human edit. Posts that ship without the edit have a higher rate of voice slips, awkward phrasings, and missing specificity.
Failure mode 5: trying to extract 30 ideas from a 20-minute episode. The extraction is bounded by source material density. A 20-minute solo monologue might yield 8-12 strong ideas; a 90-minute interview yields 40-60. Match the extraction target to the episode length.
For a host running a weekly 60-minute podcast:
Practical implication: the bottleneck isn't source material. It's the discipline to actually run the workflow weekly. Most podcasters intend to do this and then don't because it feels like work. 90 minutes per week, scheduled like the recording itself, makes it stick.
This workflow also works on adjacent source material:
DFY Voice System builds the voice prompt + Claude Project setup that makes this workflow produce voice-matched output. £497 founder pricing. Without a proper voice prompt, the repurposing workflow produces generic posts.
See The Voice BuildFour steps: transcribe, extract 30 ideas with Claude, structure each as a LinkedIn post in your voice, edit and schedule. About 90 minutes per episode.
Yes. A 60-minute episode contains 30-50 distinct insights. Quality drops if you push past the source material's density.
Dedicated transcription (Castmagic, Descript) plus Claude Projects for the extraction and structuring with your voice prompt loaded.
90 minutes per episode at steady state. First time: 2-3 hours while you tune the extraction prompt.
No. Spread across 6-10 weeks. LinkedIn penalises same-source pattern repetition.