Content Repurposing Workflow: Automate It with ChatGPT

The most expensive thing in content marketing is the first version. Orbit Media’s annual blogger survey, which asked 1,042 content marketers this year, puts the average blog post at 3 hours and 20 minutes of writing time - and that is just the writing, before research, review and publishing. Every derivative of that post - the LinkedIn version, the newsletter section, the video script - costs a fraction of it, because the thinking is already done. Repurposing is extraction, not creation, and extraction is exactly the work a language model is good at.

Most teams still treat it as a nice-to-have they never get to. In Content Marketing Institute’s B2B research, 37% of the 980 marketers surveyed named repurposing a content creation challenge. My read is that it stays a challenge because people approach it as eight separate writing tasks instead of one extraction followed by seven reformats. This post is the workflow I use to make it the second thing: one source post in, a week of derivatives out, with a review gate that keeps the derivatives worth shipping. It sits on the same chain-of-prompts logic as my bulk content pipeline, pointed at a different job.

Why repurposing is the highest-leverage prompt workflow

Two facts make the case. The first is the arithmetic above: the source post carries almost all the cost, so every honest derivative improves the return on hours you already spent. If a post took four hours all-in and 40 minutes of workflow turns it into six more assets, the cost per asset just fell by roughly a factor of five. I measure it that way deliberately, because “AI content ROI” claims usually compare against nothing.

The second fact is what consistency is worth on the distribution side. Buffer’s engagement study, built on more than 52 million posts across ten platforms, found that accounts that didn’t post in a given week consistently underperformed their own baseline growth, and that “any posting was better than not posting at all” held across platforms. Repurposing is how a small team gets that consistency without a content calendar full of net-new work. The weeks I skip social are never the weeks I had nothing to say. They are the weeks nobody converted what I had already said.

The principle: derivatives, not translations

Before the steps, the rule that decides whether the output is any good: a derivative is a piece that stands alone in its format, built from the source post’s ideas. A translation is the blog post squeezed into a different box - the intro paragraph pasted as a LinkedIn post, the headings read aloud as a video script. Translations are what naive repurposing prompts produce, and they read exactly as lazy as they are.

The practical difference is where the prompt starts. A translation prompt starts from the text (“turn this post into a tweet thread”). A derivative prompt starts from an extraction: the claims, numbers and opinions in the post, pulled out as raw material, then rebuilt for the format’s own logic - hook first for social, one idea per email, spoken rhythm for video. That is why my workflow has an extraction step in the middle instead of jumping straight to formats.

The workflow, step by step

Step 1: pick a post that earned it

I repurpose winners and evergreens, not everything. A post qualifies if it ranks or converts, if it answers a question my audience keeps asking, or if it states an opinion I want associated with me. Publishing derivatives of a post nobody wanted multiplies the silence. One source post per week is plenty, and the discipline of choosing forces the portfolio question - which of my posts deserves more distribution - that most teams never ask.

Step 2: run the extraction pass

This is the step that separates derivatives from translations. I paste the full post and run:

Here is a published blog post. Do not summarize it.
Extract, as separate lists:
1. Every distinct claim or opinion, one line each,
   quoted or tightly paraphrased.
2. Every statistic, with its source as cited in the post.
3. Every concrete example, step or rule a reader could
   act on.
4. The three most contrarian or surprising lines in
   the post, quoted exactly.
Use only what is in the post. If a list is short,
leave it short - do not pad it.

[PASTE POST]

The output is an inventory, and the inventory is reusable: every format prompt that follows works from it rather than from the full text. The “do not pad” line matters. Left alone, the model will invent an eighth claim to make the list look complete, and an invented claim in the inventory ends up in a LinkedIn post with my name on it. The same anti-invention rule runs through all my newsletter prompts for the same reason.

Step 3: map formats to the inventory, not the other way

My default map for one substantial post: three to five short social posts (each built on ONE claim from the inventory), one longer LinkedIn or thread-format piece (the argument in miniature), one newsletter section, one short-video script, and a set of quote graphics if the extraction surfaced genuinely quotable lines. That is a realistic week of distribution from one source.

The map flexes with what the extraction found. A post that is mostly a step-by-step process makes a strong carousel and a weak quote graphic. A post that is mostly opinion makes strong short posts and a bad checklist. When I force a format the inventory does not support, the model pads - it has nothing to extract, so it invents connective fluff - and padding is the tell the whole piece is a translation.

Step 4: generate per format, one prompt each

Each format gets its own prompt with its own constraints, fed the inventory plus at most the relevant section of the source. The per-platform mechanics - hooks, fold limits, hashtag rules - are the subject of my social caption prompts, and those prompts slot into this step directly. The shape is always the same:

You write [PLATFORM] posts for [AUDIENCE] in my voice
(voice rules below). From the inventory, build one post
on claim #[N] only. Requirements:
- Open with the claim or its consequence, not context.
- [PLATFORM LIMIT, e.g. under 280 characters / hook
  above the fold]
- No hashtags unless I ask. No emoji.
- Use only facts from the inventory. If the claim needs
  a number the inventory lacks, say MISSING instead of
  inventing one.
Voice rules: [PASTE]
Inventory: [PASTE]

One claim per post is the constraint that does the most work. Ask for “five posts from this article” in one go and you get five overlapping summaries. Ask five times for one claim each and you get five different posts. The voice rules block is the standing setup from my brand voice prompt guide - without it, every derivative sounds like the same articulate nobody.

Step 5: review like it is net-new, then log it

Every derivative passes the same bar as an original draft: would I accept this from a competent freelancer? HubSpot’s State of Marketing has 80% of marketers using AI for content creation now, which means the floor of sameness is high and rising - the edit is where a derivative earns its existence. I check three things per piece: the claim survived the reformat intact, the numbers match the source post, and the piece would make sense to someone who never reads the blog.

Then I log source post, formats produced and dates into a plain spreadsheet. The log is what makes this a workflow instead of a burst of enthusiasm - it tells me which posts have been mined, which formats actually got engagement, and when a winner is due for a second round.

Automating the repetition without automating the judgment

Once the prompts are stable, the assembly is worth automating. I keep the extraction prompt, the format map and each format prompt as a standing setup - a Claude Project with the voice files and prompts as project knowledge does this well, and a ChatGPT setup with the same material works the same way. Repurposing day then collapses to: paste post, run extraction, run five format prompts, edit, schedule. Under an hour in my experience, against an afternoon doing it freehand.

What I refuse to automate is the two ends: choosing the source post and reviewing the output. Orbit’s survey found 92.4% of bloggers now use AI - and, in their words, it correlates with nothing: AI users and the 8% of holdouts report “strong results” at the same rate, while the share of bloggers reporting strong results at all sits at 13.9%, an all-time low. Tools do not separate you when everyone has them. The selection and the edit are the only steps in this workflow a competitor cannot copy, which is exactly why they stay manual.

The order I would build it in

Start where I started: one source post and two formats - a LinkedIn post and a newsletter section - with the full five steps run honestly, including the log. Add short video scripts once the text formats are routine, because scripts need the most editing for spoken rhythm - the retention-first version of that discipline is in the YouTube script prompts. Add scheduling automation last, not first. Teams that start by wiring a blog-to-everything Zapier chain automate the translation mistake at scale, and then the fix costs more than the setup saved.

FAQ

How many pieces should one blog post become? As many as the extraction inventory supports and no more, which for a substantial post is usually five to eight. The inventory is the honest ceiling. When the model starts padding, you have passed it.

Should the derivatives link back to the blog post? The newsletter section yes, always. Social posts sometimes - a derivative that fully delivers its one claim earns more trust than a teaser, and platforms reward posts that keep people on-platform. I link when there is genuinely more to get, and I say what it is.

How soon after publishing should I repurpose? I run the first round within a week, while the post is fresh, and a second round three to six months later for anything that turned into a steady performer. Old winners are the most under-repurposed asset most teams own.

Does republishing the same ideas hurt SEO? No. Derivatives live on other platforms and in email, not as duplicate pages on your site. If anything the distribution earns the source post links and branded searches. The thing to avoid is republishing near-identical pages on your own domain, which is a different practice with a different name.

This workflow is part of my prompt and workflow library. The volume pipeline it borrows its review gates from is prompt chaining, the platform mechanics live in the social caption prompts, and the voice setup it assumes is the brand voice guide. The wider practice starts at the complete guide to AI marketing.