How to build
How to build a content repurposing agent
Most teams publish something good and then let it die in one format. Repurposing is genuinely valuable and genuinely tedious, which is why it is always on the list and never done. The catch is that naive automation produces obvious slop — so the interesting part of building this is not the generation, it is the voice and the human checkpoint.
What you’re building
A content repurposing agent takes one source — a blog post, a recording, a long document — and produces the derived pieces you actually publish: a LinkedIn post, an X thread, a newsletter section, a set of pull quotes. It writes against examples of your own writing rather than a generic house style.
Who this is a good fit for
- Small marketing teams with more good material than distribution
- Founders who write well occasionally and publish inconsistently
- Anyone whose newsletter is mostly rewriting things they already wrote
How to build it
Step 1
Describe the outputs and give it your voice
The voice is the whole game, and it cannot be described adequately — "professional but friendly" produces exactly the output you are trying to avoid. Give it five or six pieces of your own best writing as examples instead. Rules about what you never do are also unusually effective: no emoji headers, no "in today's fast-paced world", never open with a question.
RepurposerepurposeDescribe it
Build a content repurposing agent. From each new blog post, draft: a LinkedIn post, an X thread of 5-7 posts, and a 150-word newsletter section. Match the voice in the examples I'm uploading. Never use emoji in headers, never open with a rhetorical question, no 'in today's fast-paced world'. Put every draft in Notion for review — never publish anything directly.Build itStep 2
Connect the source and the review destination
Note what is deliberately not connected: the publishing accounts. Drafts go to Notion or Docs for a human to approve. This is the single decision that keeps the output good, and it is worth being stubborn about even when the drafts start getting reliably decent.
RepurposerepurposeConnect
Notionread posts, write draftsConnectedGoogle Docsread voice examplesConnectedSlacknotify #contentConnectedStep 3
It builds the pipeline and the style constraints
It sets up the trigger, the per-format generation, and the style rules as actual constraints rather than suggestions. Each draft is produced against your examples, which is what separates this from pasting the post into a chat window and asking for a thread.
RepurposeTrigger: new post published in NotionVoice: 6 examples indexed from Google DocsFormats: LinkedIn, X thread (5-7), newsletter (150w)Banned: emoji headers, rhetorical openers, clichésOutput: drafts to Notion → notify #contentStep 4
Edit the first batch hard, and feed that back
Rewrite the first few drafts properly rather than nudging them, and tell it what you changed and why. Two or three rounds of that moves the output from recognisably generated to genuinely close. Keep the review step permanently — the goal is removing the blank page, not the editor.
content.trysomething.site3 drafts ready from "Why your internal tools backlog never moves" • LinkedIn — 180 words, leads with the backlog stat • X thread — 6 posts, opens on the spreadsheet line • Newsletter — 150 words, links back to the full postThe LinkedIn one is too explainy. Cut it to 90 words and start with the sentence about finance running payroll from a spreadsheet.Rewritten to 94 words, opening on the payroll line. Noted: prefer shorter LinkedIn posts that open on the concrete example.
What it does once it’s running
- One source turned into every format you actually publish
- Written against examples of your own work, not a generic house style
- Hard style rules — the clichés and tics you never want to see
- Drafts to Notion or Docs for review; publishing stays a human action
- Feedback on an edit carries forward into the next batch
Tools it connects to
What it won’t do
- It drafts, it does not publish — and keeping that boundary is what keeps the quality defensible
- Voice matching needs real examples; without them the output is the generic register everyone recognises
- Anything requiring a fresh opinion or new reporting is still yours to write
Start from a template instead
Each of these opens with the prompt already written. Edit it before you build.
Common questions
- Will it sound like AI wrote it?
- That depends almost entirely on whether you give it examples of your own writing. Described styles produce generic output; five or six real pieces plus a list of things you never do gets much closer. The review step exists because "much closer" is not "done".
- Can it post to social directly?
- It is deliberately set up not to. Drafts go to Notion or Docs for approval, and publishing stays a human action — this is the decision that keeps the output worth publishing, and worth keeping even once drafts get reliably good.
- How does it learn what I changed?
- You tell it, in the same conversation. "Too explainy, cut it to 90 words, open on the concrete example" is both an edit and a lasting instruction — it carries into the next batch.
- What source formats can it work from?
- Blog posts, long documents, transcripts and recordings. Teams most often point it at a Notion space or Drive folder so new pieces are picked up automatically.
Related use case
AI agent for market & competitive intelligenceMonitor competitor pricing, changelogs, and signals, and get alerted when something changes.
Build it in the next ten minutes
Start with a sentence. Leave with a working app on its own URL.