Content repurposing fails for a boring reason: teams automate the generation step and leave every other step manual. You end up with forty AI-written drafts nobody has time to check. The workflow below fixes the sequencing, not the writing.
Start from a source asset, not a content calendar
Repurposing only compounds when the source carries information that cannot be generated — a real job walkthrough, a customer call, a teardown, an original number you measured. Generic source material produces generic derivatives, and the AI cannot add substance that was never there.
Most repurposing advice inverts this. It starts with a calendar — five posts, one newsletter, two shorts — then hunts for something to fill the slots. That is feeding a schedule, not publishing because you had something to say. The source is the input; the calendar is downstream.
A source worth repurposing holds at least one of four things: a number you measured, a decision you can explain, a mistake with a cost attached, or a walkthrough of the work. If yours has none of those, no workflow saves it.
The five-stage workflow
Five stages, in order. The order matters more than the tools — you can run this with a phone recorder, a transcript and a text editor, and beat a paid stack running the same stages out of sequence.
- 01Capture — one substantial asset per cycle.
- 02Transcribe and mark — tag the three or four passages that carry the actual insight.
- 03Derive — generate platform-native variants from the marked passages only, never the whole transcript.
- 04Edit for fact — check every claim, number, name and price.
- 05Schedule and space — distribute derivatives over weeks, not days.
Stage one: capture one asset, not five
One substantial source per cycle — a week or a month, whichever you can hold. Operators who try to capture three a week quit inside a month, because capture competes directly with billable work. One is sustainable. One compounds.
Stage two: transcribe, then mark
Transcribe the whole thing, then mark three or four passages. Not summaries — the sentences where something true and specific got said. Ten minutes with a highlighter. It is the only step where your judgment is irreplaceable, and the one every tool wants to sell you a way to skip.
Stage three: derive from the marks
Hand the model the marked passages and a format brief, nothing else. Constrained input is the whole trick. A model given a full transcript averages it. A model given one sharp paragraph and a target format keeps the sharpness, because it has nothing else to fall back on.
Stage four: edit for fact, not for style
Style edits are optional. Fact edits are not. Check every claim, number, name, date, price and availability statement. Assume the model invented anything specific it was not handed. This step does not shrink as your prompting improves — it is a fixed cost of publishing.
Stage five: schedule with space between
Space the derivatives across weeks. Six variants of one idea in three days teaches your audience that you repeat yourself. The same six across a month reads as a point of view. Nothing changed except the spacing.
Your content repurposing matrix
The matrix is two lists and a rule. Down the side: the kinds of source you can actually produce. Across the top: the derivative formats you can maintain. The rule is that you fill cells, not calendars.
Start with the columns, because they are the constraint. Pick no more than five derivative formats — the ones you will still be producing in six months. A working set: one long written post, one short video, one email, two or three social posts, and one sales-facing asset such as an objection answer.
Then the rows. These five cover almost everything an operator can produce without hiring anyone.
Row 1 — the recorded job walkthrough
Strongest into short video — the footage already exists — plus a written how-it-works post and an objection answer for sales. Weak into email unless something went wrong; process descriptions do not carry an inbox.
Row 2 — the customer conversation
Recorded with written permission. Strongest into objection answers, FAQ entries and email — the only row in the customer's own words. Weak into video. Never publish a name, address or number without permission in writing; that is the fastest way this workflow costs you a client.
Row 3 — the teardown
You take apart something public: a competitor's quoting flow, your own old sales page, a pricing page you found confusing. Strongest into social and long-form. Highest reach, easiest to do badly — the version that works names what you would do instead.
Row 4 — the measured number
One number you measured yourself: how many calls went unanswered last month, how long an estimate sits before it goes cold, how much of this quarter came from repeat customers. This row feeds every column, and a competitor cannot copy it by rewriting your post.
Row 5 — the question you answered this week
The cheapest row to fill. If somebody asked you something that took two minutes to answer well, that answer is already a derivative. Strongest into email, social and FAQ. Weak into long-form unless several questions stack into one theme.
Five rows by five columns is twenty-five cells, and you should never fill them all. Three to six cells per source is the working range. The empty cells are the point — that is where you decided a format was wrong for the material.
Audio and video sources
Recorded material is the highest-yield source because it is dense and cheap to capture — you were going to have the conversation anyway. Transcription is close to solved, so the constraint moved to selection: a sixty-minute call might hold four useful minutes, and the workflow exists to find them.
One practical limit: consent. Recording a customer call is a legal question before it is a content question, and the rules differ by state and country. Get permission in writing, keep it, and treat identifying details as off-limits until the customer has seen the draft.
Where the tools fit — and why their pricing pages do not line up
You can run this workflow on almost any stack. What differs is which stage a product attaches to and what it meters — worth knowing before you open two pricing pages side by side, because side by side is where this goes wrong.
On the generation side, the meters have split apart. Jasper: per seat, plus a separate credit meter for its newer agent features. It documents a repurposing feature — “Remix Content” — on both published tiers; the top tier is quote-only with a twelve-month commitment. HubSpot's Content Hub: per seat at entry, then a flat platform fee with bundled seats, marketing-contact tiers, and consumable credits that reset monthly and do not roll over, plus a required one-time onboarding fee. It documents “Content Remix” on upper tiers. Writesonic: metered on tracked prompts, projects, articles per month and site-audit volume — an SEO and AI-visibility platform, not a per-word one. Copy.ai: a low-cost per-seat chat plan, then workspace plans on bundled workflow credits and seat pools, nothing self-serve between them, a gap of roughly two orders of magnitude. Rytr: flat per account, free tier metered in characters, paid tiers advertising unlimited generation.
Two things worth noticing. Only some document a repurposing feature at all — Jasper and HubSpot name one, Copy.ai names none. And most no longer sell what their category implies: Jasper's homepage now reads “Put AI agents to work for marketing,” Copy.ai calls itself an AI-native GTM platform, Writesonic leads with winning customers from AI search. Buy one expecting a writing tool and you are buying a product that already moved.
The scheduling stage is worse, because those products bill in units that are not comparable. Sprout Social bills per seat, so cost scales with team size, not connected profiles. Hootsuite bills per user, the account allowance going unlimited above the entry tier. Buffer bills per channel — you pay only for accounts you actually manage. Later sells bundles of social sets, one set being eight profiles, plus seat caps and add-ons. Metricool bills per brand, a brand being one connected set of profiles. Publer is modular: a base covering one social account, then published fees per extra account and per extra member.
Six products, and barely any two bill in the same unit: seats, users, channels, brands, bundles of eight profiles, and pure à la carte. Side-by-side comparison of those pricing pages is close to meaningless. The question is not which is cheapest but which unit matches the shape of your operation.
What to automate and what to leave alone
Automate the mechanical, deterministic steps. Leave anything with a judgment or a fact in it to a person. The rough rule: automate transitions between stages, not the decisions inside them.
- ›Safe to automate — file handling, transcription, formatting, scheduling, cross-posting, and moving an approved asset to the next stage.
- ›Keep human — passage selection, fact and claim checks, anything naming a price or a customer, and final approval to publish.
- ›Never automate — publishing without an approval gate. The cost of one wrong published claim about your own business exceeds the time the gate saves.
Approval gates are unpopular because they look like the bottleneck. They are not. The bottleneck is stage two. A gate you clear in ninety seconds slows nothing down — the forty unmarked drafts queued behind it do.
Where this breaks
- ›You have nothing to say. No workflow manufactures substance. Fix the source, not the pipeline.
- ›Nobody owns stage two. If marking belongs to whoever has time, it belongs to nobody, and you are back to unattended generation.
- ›The formats outrun the maintenance. Five columns you sustain beats nine you abandon in March.
- ›You measure output. Volume climbs, replies fall, and it takes a quarter before anyone notices.
Making it scale without adding people
A repurposing system scales when per-cycle human time stays flat as output grows. If each new derivative adds edit minutes, you built a treadmill. Measure minutes per cycle, not posts per week — volume is the vanity number here.
The three numbers worth tracking
- ›Human minutes per cycle — capture, marking and fact-editing added together. This is the number that decides whether the system scales.
- ›Distinct derivatives per source — count formats, not posts. Six near-identical posts is one derivative with five copies.
- ›Cells left deliberately empty — if that number is zero, you have gone back to filling a calendar.
Notice what is not on that list: posts published, words generated, credits consumed. Every tool in this market reports those, because they are the numbers that make the tool look load-bearing. None tell you whether the system works.
What this looks like in a service business
Concrete version, for an operator with trucks. Monday, a tech records ninety seconds of a repair he has done a hundred times and explains why the cheap fix fails. Tuesday it gets transcribed and three sentences get marked — the one about the cheap fix is the whole asset. Wednesday, four derivatives come out of it: a short video using the footage, a written post, an email, and a paragraph the office pastes when a customer asks why the quote is higher. Thursday someone checks the part names and numbers. Then it ships across three weeks, not one afternoon.
Total human time that week: well under an hour, and the recording was already happening. That is the whole argument. Not that AI writes for you — that one hour of your own expertise, marked properly, can carry a month of distribution.
Next: how to compare the tools that do the generating