The Repurposing Chain: Turning One Pillar Piece Into Nine Assets

Most repurposing programmes produce nine worse versions of one good thing. You publish a 3,000-word pillar, then a junior writer or a Zapier step slices it: the intro becomes a LinkedIn post, the H2s become a carousel, the conclusion becomes a newsletter. Every derivative is a lossy copy of the parent. The audience on LinkedIn gets a thinner version of the argument the blog readers got, and they can tell, because a compressed argument has a distinctive smell. It gestures at evidence it doesn’t have room to show.

The fix is structural, not stylistic. Stop treating the published pillar as the source. Treat it as the first derivative.

The source of record is the evidence file, not the article

When we ran the UK Content Retainer Benchmark this spring, the pillar was 3,400 words built on 41 anonymised retainer agreements and 12 practitioner interviews. That article is not the asset. Underneath it sits a folder we call the evidence file:

• 12 interview transcripts (recorded in Riverside, transcribed in Descript, cleaned to ~1,400 words each)
• a 41-row Google Sheet of retainer values, deliverable counts and effective per-piece rates
• 30 verbatim objections lifted from Fathom recordings of sales calls
• 19 quotes we marked as quotable, tagged by theme
• a one-page method note covering sample bias and what we refused to claim

That folder goes into a Claude Project or a custom GPT’s knowledge base once. Roughly 46,000 words of raw material. The pillar spends about 11% of it. Which means every derivative afterwards has 89% of the source still available, unused, and never seen by the reader on any other channel.

This is the whole mechanic of effective ai content repurposing: each asset is generated from the evidence, plus a channel brief, plus a stated argument. The pillar is never in the prompt. If you paste the pillar in, the model will paraphrase it, because paraphrasing is the cheapest path to a plausible output and models take cheap paths.

Nine assets, nine arguments

The same evidence supports different claims depending on who is reading and what they came for. A LinkedIn feed rewards a provocation with one number attached. A sales team needs the counter to “you’re more expensive than the agency we’re quoting.” A trade op-ed needs the industry-level version, not the how-to. Those are not the same argument at different lengths. They are different arguments.

Here is the chain we actually shipped, in order:

Total derivative build time: 6 hours 10 minutes across two weeks, with two people. Doing the same nine ad hoc, from the pillar, took us 11 hours the previous quarter and produced four assets we quietly never distributed.

Why the order matters more than the count

Front-load by distribution, not by effort. Asset 1 went out three days before the pillar published. That felt wrong the first time we did it and it is now non-negotiable, for two reasons.

The LinkedIn document post reached 28,400 impressions and pulled 61 comments. Nine of those comments were objections we had not anticipated, including a good one about how day-rate contracts distort the per-piece maths. We had 72 hours to fold that into the pillar before it went live. The highest-distribution asset became a free pre-publication review from the exact audience the pillar was for.

Second reason: sequencing protects the expensive assets. The calculator took 90 minutes and captured 210 email addresses off 640 downloads. We only built it because the LinkedIn comments proved people wanted to run their own numbers. Had we built it first, on instinct, it would have been a 90-minute bet on a hunch.

The slow, durable assets sit at the back deliberately. The answer hub and its schema took 35 minutes and drove 1,900 organic sessions in month three, long after the social assets had gone quiet. Ordering is the difference between a repurposing chain and a scattergun.

The prompt pattern

Every derivative uses the same three-part brief. Channel constraints, the argument, and a hard instruction against compression:

SOURCE: Use only the evidence file in this project. Do not reference,
paraphrase or summarise the published article at /content-retainer-benchmark/.

ARGUMENT: UK content teams price retainers on hours because that is what
procurement understands, and hourly pricing guarantees margin decay as the
team gets faster. Support this with at most three figures from the sheet
and one interview quote. Do not make any claim the sheet cannot support.

CHANNEL: LinkedIn document post, 8 slides, max 14 words per slide, plus a
110-word caption. The caption must open on the objection, not the finding.
No "we surveyed" framing. No list of what's inside.

BANNED: any sentence that could open a blog post.

That last line does surprising work. Without it, Claude and GPT-5 both drift toward article-shaped openings, because their training data is thick with article-shaped openings.

The difference in output is not subtle:

COMPRESSED (pillar in the prompt):
"We analysed 41 UK content retainers to understand how teams
are pricing their work in 2026. Here's what we found."

RE-ARGUED (evidence file only):
"Your retainer has been the same £4,800 for three years.
Your team is 40% faster than it was. Do the division."

Same data. The second one earned the 61 comments.

Where the chain breaks

Two failure modes show up reliably. The first is a thin evidence file. If your pillar was assembled from six competitor articles and a Perplexity session, there is no 89% remainder, and every derivative will be a compression no matter how you prompt it. Nine assets from thin research is nine ways to be generic. Run the chain only on pillars where you own original material: interviews, your own data, client work you have permission to describe, support tickets, sales call recordings.

The second is voice drift across nine outputs. We solved this with one shared instruction block, about 600 words, that lives in a custom GPT and gets pasted into every Claude conversation. It contains banned constructions, three before/after rewrite pairs from our own editing, and the two-em-dash rule. Standardising that layer, and the retrieval layer underneath it, is the part most teams skip; there’s a fuller treatment of the GPT and automation setup in our pillar on automation, repurposing and custom GPTs.

Worth saying plainly: not every pillar deserves nine. Ours earns it because the underlying data cost about 30 hours to gather. A well-written explainer with no original evidence earns three assets at best, and forcing it to nine is how teams end up with a content programme that publishes constantly and is quoted never.

Next time you brief a pillar, build the evidence file first and write the article second. The order of those two steps decides whether you have one asset or nine.