How Do I Build a Claude-Powered Content Repurposing Pipeline?

One well-produced podcast or keynote can become 40 pieces of content: a LinkedIn post, a Twitter thread, a newsletter, a three-part blog series, a YouTube Short pack, and a carousel. The work of cutting and rewriting is the bottleneck — not the ideas. A Claude-powered repurposing pipeline takes a transcript in and pushes out all of those formats in one pass, in your voice. Here's how to build it.
Why this matters
Content repurposing is where consultants and creators lose the most time with the least return. You interview someone, you record a talk, you draft a newsletter — and then you hand-chop the same material into four other shapes because "you should be on every platform." Most of it isn't better for being hand-chopped; it's just slower.
A structured Claude pipeline does three things you can't easily do manually:
- Hold your whole voice guide in context for every output, so every piece sounds like you.
- Stay consistent across formats — the LinkedIn post and the tweet thread reference the same central idea instead of drifting into different takes.
- Run overnight — you drop the transcript in at 10pm, you review seven drafts at breakfast.
Before you start
You need:
- A Claude Project set up with your voice guide and past published examples. If you don't have one, start with How Do I Build My First Claude Project.
- A real transcript to repurpose. Otter, Riverside, Descript, YouTube auto-captions — any of those. Bad transcripts produce bad output, so don't skimp on the source.
- One hour to run through the full setup and produce your first pipeline output.
Step 1: Decide on your output formats and their shapes
Don't "do every platform." Pick the three or four that actually move your business. For me:
- LinkedIn post — 1,200 chars, hook + three-beat argument + CTA. No hashtags except category tags.
- Newsletter section — 400 words, subhead + 3 takeaways + one story.
- Blog-length piece — 1,500 words, our Why/Before/Steps/Verify structure.
- X/Twitter thread — 8 tweets, numbered, hook tweet on the insight.
Each format gets a short spec: length, structure, voice traits, what to exclude. Write these down as markdown files — one per format — and drop them in your Project's knowledge base as format-linkedin.md, format-newsletter.md, etc.
Step 2: Build a Skill per format (or one master Skill)
Two paths. Both work.
Multi-Skill: one /repurpose-linkedin, one /repurpose-newsletter, etc. Easier to iterate on each format independently; more friction to run all of them.
Master Skill: /repurpose-transcript that takes a transcript and a list of target formats, and produces all of them in one run. More upfront work; one command for everything afterward.
I use the master Skill. SKILL.md:
---
name: repurpose-transcript
description: Turn a source transcript into multiple content formats in Jake's voice.
---
# Repurpose-transcript
You transform a transcript into content across multiple formats.
## Inputs
- Transcript: plain text or path.
- Target formats: list from {linkedin, newsletter, blog, thread}.
## Process
1. Read the transcript fully. Identify the ONE central insight.
2. Identify 3-5 supporting points.
3. Identify 1-2 specific stories or examples the speaker used.
4. For each requested format, follow the format spec in
format-<name>.md and produce one draft.
## Rules
- Preserve the speaker's specific words for quoted material.
- Use Jake's voice (see voice-guide.md).
- Do NOT invent facts the speaker didn't mention.
- If a format doesn't fit the material, say so rather than force it.
- Mark any ambiguous takeaways with [CONFIRM: ...] for human review.
## Output
One section per requested format, prefixed with "## Format: <name>".Skills in general are covered in How Do I Package a Team SOP as a Claude Skill.
Step 3: Feed it a real transcript
In Claude Code or Claude.ai (inside your Project):
/repurpose-transcript
[paste transcript]
Formats: linkedin, newsletter, threadClaude reads, identifies the central insight, and drafts each requested format. First run, read the output carefully. You'll see:
- Where your voice guide is landing and where it isn't.
- Which format specs need tightening ("too much hedging," "wrong hook style").
- Whether the central insight Claude picked is actually the right one.
Revise the format specs and voice guide based on what you see, and run it again.
Step 4: Add a review pass
Claude's first draft is 80% of the way there, not 100%. Build a review stage into the pipeline where Claude rereads its own output:
Add to SKILL.md:
## Self-review pass
After producing the initial drafts, re-read each one and:
1. Check for filler words or hedging. Remove them.
2. Check that the hook is in the first 10 words. Rewrite if not.
3. Check that every claim has a concrete example. Flag if not.
4. Check voice alignment against voice-guide.md. Rewrite if off.
Output the revised versions only.This single extra pass removes about 60% of the hand-editing I used to do. The hook check is the highest-ROI line in this whole guide.
Step 5: Schedule it for unattended runs
Once the Skill produces reliably good output, wire it into an overnight run so you never manually invoke it.
Typical workflow: transcripts land in a Dropbox folder → a cron job picks up new files → runs Claude Code with the Skill → writes output back into a drafts/ folder. Morning, you review drafts and publish.
The scheduling side is covered in How Do I Schedule Claude Code to Run Overnight Jobs?.
Step 6: Close the loop — feed published pieces back
Every piece you publish becomes a new voice-guide data point. After a month, drop the 10 best published pieces back into the Project knowledge base (in a published-examples/ subfolder). The pipeline's output quality compounds because Claude has more aligned reference material.
Treat the voice guide as living. Every quarter, prune weak examples, add strong ones.
Verify it worked
1. One transcript produces all target formats. Feed a real transcript through the Skill, confirm you get LinkedIn + newsletter + thread + whatever you configured.
2. Each format matches its spec. LinkedIn drafts within 1,200 chars, threads at exactly 8 tweets, blog pieces in your structure. Length and shape should be mechanical.
3. Voice feels like you. Run one transcript through the pipeline, edit the output the way you'd usually edit. If your edits are under 10% of the text, voice alignment is solid. If you're rewriting 40%, the voice guide or format specs need work.
Where this breaks
- Transcripts with heavy background noise or crosstalk. Garbage in, garbage out. Clean up obvious transcription errors before running. One minute of cleanup saves an hour of bad drafts.
- Content that doesn't survive format-switching. A tactical walkthrough might make a great blog post and a terrible tweet thread. If the Skill forces it anyway, you get mediocre output. Let the Skill say "this format doesn't fit" — better than a bad draft.
- Voice drift over time. Your voice evolves; your voice guide doesn't unless you maintain it. Review and refresh quarterly.
- Claude being too helpful with "filler ideas." If the transcript covers three topics, Claude may try to include all three in every format. Tell the Skill explicitly: "pick the ONE central insight and stay on it."
- Scheduled runs producing content you never review. The pipeline is a first-draft engine, not a publishing robot. Always review before posting, even when the output looks great.
What to try next
- How Do I Build My First Claude Project That Actually Knows My Business? — the voice-guide-and-examples foundation the pipeline assumes.
- How Do I Package a Team SOP as a Claude Skill Anyone Can Run? — the Skill pattern generalized beyond content ops.
- How Do I Schedule Claude Code to Run Overnight Jobs? — the cron wrapper that turns the pipeline into a background service.
Let's talk about your AI + SEO stack
If you'd rather skip the how-to and have it shipped for you, that's what I do. Start a conversation and we'll figure out the fastest path to results.
Let's Talk