Codifying a brand voice into a reusable profile from approved copy
The complaint clients have about AI copy is that it sounds like everyone. You can fix that once and reuse it: feed the model several pieces of copy the client already approved, have it extract the actual patterns into a voice profile plus a reusable prompt, and every future draft starts closer to on-brand. It compounds — build it once, paste it before every job.
You are a brand-voice analyst. Below are {{count}} pieces of copy that {{brand}} has approved and published. Analyze only these samples and build a reusable voice profile. Approved samples: {{samples}} Produce: 1. Voice attributes — 4 to 6 traits (e.g. "dry, not jokey"), each with a do example and a don't example drawn from or contrasted with the samples. 2. Mechanics — typical sentence length and rhythm, person (first/second), contractions, punctuation habits (e.g. em dashes, no exclamation marks). 3. Vocabulary — words and phrases the brand uses, and words it clearly avoids. 4. A reusable system prompt — a paste-ready paragraph I can put before future drafting so any model matches this voice. Rules: - Base every guideline on a pattern actually present in the samples. Do not import generic "brand voice" advice or invent rules the samples don't show. - Quote a short example from the samples for each attribute so I can verify it. - If the samples are inconsistent on something, say so rather than picking one.
Fill in your details and the prompt updates live — then copy.
Voice attributes: - Understated, never hype. Do: "It's warm. That's the whole pitch." Don't: "Revolutionary thermal technology!" - Second person, direct. Do: "You'll notice it at the trailhead." Don't: passive brand-speak. Mechanics: Short sentences (avg ~11 words). Contractions throughout. Em dashes common; no exclamation marks in any sample. Vocabulary: uses "gear," "trail," "built to last"; avoids "luxury," "premium," "game-changing." Reusable prompt: "Write as Fernweh: understated, second person, short sentences, contractions, dry humor, no exclamation marks or hype words. Never claim performance you can't back."
The full workflow
- Gather 4-6 pieces the client actually signed off on — approved copy, not drafts
- Run the prompt and sanity-check each guideline against the samples it quotes
- Save the reusable system prompt and paste it before every future draft for that client
- Revisit the profile when the brand refreshes its voice, since an old profile drifts out of date
Watch out for
A voice profile makes copy consistent, not correct. It still can't authorize claims — every product statement generated in that voice must be substantiated separately under FTC rules before it runs.
Copy generated purely by an AI system isn't copyrightable, per the U.S. Copyright Office's 2025 guidance. If your client needs to own and protect the final copy, your human editing and arrangement — not the raw generation — is what carries the authorship.
Where this comes from
Every use case on this site is grounded in real reports from working copywriters — not invented by us.