Repurposing breaks brand voice at one point: the moment a machine rewrites your words without knowing how you sound. The fix is to automate the mechanical work (cuts, sizes, first drafts), put your voice in writing so the machine has something to follow, and check three things before anything publishes.
TLDR: Write a one-page voice card, let AI draft platform versions from the original post using it, and run every draft through three checks (opening line, claims, call to action). Automate the formats, never the decision about how you sound.
Repurposing steps that carry voice, and the ones that don't
Not every repurposing step carries voice. Split them like this:
| Step | Automate? | Why |
|---|---|---|
| Resizing and cropping for each platform | Yes | No voice involved |
| Cutting a long video into short clips | Yes, then review | Machine finds the cuts, you pick the moments |
| First draft of a caption per platform | Yes, from a voice card | Saves the blank page |
| Opening line (the hook) | Human | This is where voice shows most |
| Claims, numbers, names | Human | One wrong number does lasting damage |
| Call to action | Human | It is the point of the post |
The rule: machines produce options, a person chooses the first line.
Write a one-page voice card the AI can actually use
"Friendly and professional" is not a voice. Give the model material it can copy and avoid. One page, five parts:
- Four to six trait words with an example each. "Direct: we say 'it costs $X', not 'pricing may vary'."
- Words we use and words we never use (jargon, hype words, emojis you avoid).
- Sentence habits: short sentences, second person, no exclamation marks.
- Three posts that sound exactly right, pasted in full.
- One post that sounds wrong, with a note on why.
Our guide to building a reusable brand voice prompt shows how to turn the card into a saved prompt. If you run several brands, keep one card per brand and never mix them in a single prompt.
Repurpose in one direction, one change at a time
Voice drifts when you ask for too much in one pass. Start from the original post, the source of truth, and ask for a single change: shorter for X, more visual for Instagram, a hook for a Reel, a different tone for LinkedIn. Never repurpose a repurposed version, because each round loses a little more of the original.
The 15-pieces-from-one-post workflow is a good map for which formats to ask for. In Mydrop's AI content tools you can generate captions and post ideas, rerun them, and edit before they go anywhere.
One data point to keep in mind when an automated rewrite shortens things: in Mydrop's data, captions under 80 characters ran 9% below the account's typical post, while every length from 80 characters up sat within 3% of the median. A trimmed caption is not automatically a better one.
Three checks before anything publishes
Run every repurposed piece through the same quick review, in this order:
- Opening line: would your team say this out loud? If it sounds like a template, rewrite it.
- Claims and numbers: compare them with the source post, line by line. Machines round, merge and invent.
- Call to action: does it match the platform and the goal? "Link in bio" does not belong on a platform without a bio link.
Use a per-platform preview so the check includes how the post looks. Mydrop's post preview shows each version as the platform will render it, and approvals let a second person sign off before publishing when the stakes are higher.
Watch for drift every month, not every post
Reviewing each piece forever is not scalable. Instead, once a month pull 10 repurposed posts from different platforms and read them next to the voice card. If three or more sound off, the fault is in the card or the prompt, not the writer. The guide to spotting brand voice drift lists what to look for.
Write your voice card this week
Block 45 minutes, write the one-page card from your three best posts, and use it for your next repurposing batch. To run the drafts, previews and approvals in one place, start free with Mydrop.
Google review
Trustpilot review
Product Hunt comment
