The failure mode is not laziness, it is structure. Ask a tool to "turn this into a LinkedIn post" and it will compress the article — same argument, fewer words, same shape. That is a summary, not a repurpose. A real repurpose changes what the piece does: an article explains, a post provokes, a checklist instructs, a script is spoken.
The method below starts by extracting the raw material, then rebuilds each format from it against a different job. Extraction can be delegated; deciding what each format is for cannot.
Step 1: break the original into material, not prose
Before asking for any new format, pull the source apart into four lists:
- Claims — the assertions the piece defends
- Evidence — figures, test results, named examples, dates
- Steps — anything the reader is told to do, in order
- Objections — the pushback the piece anticipates and answers
This matters because different formats draw on different lists. A checklist needs the steps. A post needs one claim and the objection it raises. A script needs the evidence delivered in spoken rhythm. Compressing the prose loses all four distinctions.
Step 2: assign each format a job
| Format | Job | Draws on | Length |
|---|---|---|---|
| Social post | Provoke one reaction | One claim + one objection | 80–150 words |
| Checklist | Enable an action | Steps | Steps only |
| Continue one relationship | One claim, conversational | 120–200 words | |
| Script | Be spoken aloud | Evidence + structure | 2–5 minutes |
| FAQ block | Answer before being asked | Objections | 4–8 pairs |
| Comparison table | Support a decision | Claims + evidence | Table |
Do not generate all six from one prompt. Run each separately with the job and the source material stated, or you will get six variations of the same paragraph.
Step 3: rebuild, do not compress
For each format, state the job, the audience, and the material to use. A checklist built from the steps section should contain only the steps — no claims, no evidence. That constraint is what makes it useful, and it is the one a generic prompt will ignore.
Then check each output against its job. Does the script read aloud? Tools produce sentences that look fine and trip the tongue — relative clauses, stacked prepositions, nominalisations. Reading it out loud catches these in seconds, and no amount of re-prompting is faster.
Step 4: verify before publishing
Repurposing propagates errors at speed: one wrong figure in the source becomes six wrong figures. Verify every number, date and name against the original source in each output, not just the first. In our testing, figures drifted most often in compressed formats where the surrounding context that qualified them had been cut.
Watch especially for qualifiers disappearing. "In our testing" becomes a universal claim. "Most" becomes "all". Check the modal verbs and scope words specifically — they are small and they carry the accuracy.
Step 5: keep a source of truth
When something is wrong in the original, fix it there and re-derive, rather than patching each version. Otherwise six versions drift apart, and you no longer know which one is correct.
What each format needs, briefly
Social post: one idea, a specific detail, no preamble. If it needs context to land, it is the wrong idea for a post.
Checklist: action first, one item per line, no explanations. If an item needs explaining, it is two items.
Email: one point, written as though to one person, one action at the end.
Script: short sentences, spoken transitions, and a reason to keep listening in the first ten seconds.
FAQ: the question a real person would type, then the answer without hedging.
Comparison table: same criteria on every row, and a blank where something is unknown rather than a guess.
Where AI genuinely helps
Extraction and format conversion are worth delegating. Pulling the steps out of a long article, reformatting a paragraph into a table, producing a first draft of an email from a section — these are mechanical, fast to review, and the errors are visible.
Judgement calls are not. Which claim is worth a post, whether a comparison is fair, whether the evidence supports the conclusion — these need someone who knows the subject. In our thirty days of testing, the drafts that caused problems were always the ones where judgement had been delegated along with the typing.
Convert fast, judge slowly
Let a tool do the reformatting. Keep the decisions about what each piece is for.
Try Rytr →Frequently Asked Questions
Can I generate all six formats in one go?
You can, but you will get six versions of the same paragraph. Run each format separately with its own job stated.
How do I stop the output sounding like the source?
Change the job, not the wording. A checklist built only from steps reads nothing like the prose it came from.
What needs checking before publishing?
Every figure, date and name in every output, plus the qualifiers — "in our testing" tends to become a universal claim.
Should I edit the original or the derived versions?
Fix errors at the source and re-derive. Patching each version lets them drift apart.
Which format is worth most for a small site?
In our testing, the FAQ block and comparison table — they answer the query directly and get reused in other pages.
Is repurposed content a duplicate-content risk?
When each format genuinely has a different job, no. The risk comes from near-identical text dressed differently.