Repurposing a source into several drafts
Turning one article, page, document or recording into posts, captions, threads and hooks, with a record of where each draft came from.
Updated
Repurposing takes one source - pasted text, a public link or an uploaded file - and produces several ready-to-edit drafts in the formats you pick, each carrying a record of what it was made from. Every draft goes through the same governed generation path as any other AI writing, so it is priced, charged and provenance-tracked the same way.
Recordings and scanned documents cannot be read in this release. There is no speech-to-text provider and no optical character recognition engine, and the page says so rather than producing an empty draft.
BEFORE YOU START
- The "generate" permission to produce drafts, and "create" or "edit" to bring a source in.
- An AI text provider on this deployment. Without one, every target comes back honestly failed rather than as a fabricated draft.
- Credits, one governed generation per format you pick.
- The right to reuse the material. For a link or a file you must tick the rights acknowledgement; the product will not accept the source otherwise.
STEPS
- Open /app/repurpose.
- Choose a source tab: "Paste text" for something you already have, "From a link" for a public page or PDF, or "Upload a file".
- Fill in what that tab asks for. Pasted text has a title and a body; a link has an address; a file is picked from disk.
- For a link or a file, read the rights notice and tick "I have the right to reuse this material."
- Choose "Bring in this source". The page shows what was extracted, how it was extracted, and an excerpt.
- Under "Choose formats", tick what you want: Post, Caption, Thread or Hooks. Between one and six formats.
- Choose "Repurpose into N formats".
- Open each draft with "Refine in Create" to edit it properly.
WHAT YOU SHOULD SEE
A source summary with a status chip and a citation line, then one draft per format, each with a provenance line saying what it came from and how many credits it settled at.
WHAT THIS WILL NOT DO
- It cannot transcribe a recording. No speech-to-text provider is connected, nothing is sent anywhere, nothing is charged, and the recording itself is not retained - only its name, size and checksum. Paste a transcript if you have one.
- It cannot read text inside an image or a scanned PDF. There is no OCR engine connected.
- It will not carry a claim into a draft that does not appear in the approved source text. Selected claims are checked against the source and refused if they are not there.
- It does not claim a file is clean when nothing scanned it. The scan sentence only appears when a malware scanner is actually connected.
- The mobile screen is text only. It pastes source text and picks formats; links, files, rights acknowledgement, provenance and claim checking are web-only.
IF IT DOES NOT WORK
- "That URL cannot be ingested for security reasons." means the address is not one the product will fetch.
- "That URL could not be read. Check it and try again."
- "This document could not be opened. It may be corrupt, password-protected, or not the format its name suggests."
- "Text cannot be extracted from this type of file." means the format is outside the accepted list.
- "This file was read successfully but contains no text."
- "This upload was reported as malicious by the malware scanner and was discarded without being read."
- "Some selected claims do not appear in the approved source text, so they cannot be carried into a draft." lists which ones.
- "AI provider not configured yet" against a draft means no text provider exists. Nothing was charged for that draft.
- "That request could not be sent, so nothing was repurposed and nothing was charged. Your source is still here."
COMMON QUESTIONS
How big can a source be?
Pasted text is capped at 50,000 characters in the browser and 8,000 characters for the generation itself. Documents go up to 20 MB, recordings up to 64 MB - and a recording still cannot be transcribed.
Are failed drafts charged?
No. Usage is recorded for them so the attempt is visible, but a failed generation settles nothing.
Why does it say the AI provider is not configured when the error looked like a credit problem?
Because a missing provider takes precedence in the label. Two answers on one screen sent people to billing for a problem billing could not fix, so the provider answer wins.
RELATED GUIDES
- Writing posts with AI
Generating a first draft from a brief, checking what it will cost first, and what the output is and is not.
- Grounding content in sources
Adding a webpage or your own notes so generations can cite them instead of inventing a claim.
- Finding clip moments in a transcript
Reading and correcting a machine transcript, and asking it for places worth a look - which is a text search, not a judgement.