Brand Voice Styler

Every brand with more than one writer drifts. This skill stops the drift by making voice explicit: it reverse-engineers a voice sheet from copy that already sounds right, scores the voice on named axes with quoted evidence, then rewrites target copy against the sheet — with every single change justified by a rule, so "on brand" stops meaning whatever the loudest reviewer feels today.

When to use this skill

• Copy needs rewriting to match a house voice, documented or not
• No voice guide exists and one must be reverse-engineered from good samples
• Multiple writers or agencies produce copy that does not sound like one company
• A rebrand or tone shift needs existing copy migrated to the new voice
• Two stakeholders disagree about whether a draft is "on brand" and need a referee

Instructions

1. Collect three to five samples the user confirms are on-voice — ideally across formats: a headline, a product page, an email, a support reply. The sheet is only as good as its evidence, and one sample is an anecdote.
2. Score the samples on five axes, one to five, each score backed by a quoted phrase: formal to casual, warm to neutral, playful to serious, plainspoken to technical, bold to careful. Disagreements between samples are findings — surface them; do not average them away.
3. Harvest the lexicon: words the brand reaches for, words it never uses, how it addresses the reader, sentence-length habits, punctuation tics, how it says no, how it apologizes.
4. Write the mini voice sheet (template below) and get it confirmed before touching the target copy. Editing without an agreed sheet just swaps one writer's taste for another's.
5. Rewrite the target against the sheet, sentence by sentence. Headlines, body, button labels, and error text all get the treatment — voice breaks loudest in the smallest strings.
6. Self-diff: for every changed sentence, name the axis score or lexicon rule that justified the change. A change with no rule behind it gets reverted, however nice it sounds.
7. Run the same-author test: place the rewrite beside the original samples and read straight through. If a stranger could spot where the samples end and the rewrite begins, iterate.

Voice sheet template

Voice: <brand>                     Evidence: <the 3-5 confirmed samples>
Axes (1-5, each with a proving quote):
  formal 2 · warm 4 · playful 3 · technical 2 · bold 4
Always: <habits — verbs first, "you" before "we", numbers over adjectives>
Never: <banned words and moves — no exclamation marks, no jargon, no scolding>
Sentences: <typical length and rhythm — "short. then shorter.">
Edge registers: errors <how it apologizes> · legal <how formal it is allowed to get>

Failure modes to catch

• Voice cosplay: adjectives from the brand deck ("bold, human, witty") pasted over copy that still reads like a terms-of-service update. The quoted evidence on each axis exists to stop this — no quote, no score.
• Over-rotation: every sentence maximally on-voice until the page is exhausting to read. Voice is seasoning; the plainest sentence in the samples is part of the voice too.
• One-format sheets: a voice extracted only from headlines will fail the first time it has to apologize in a support reply. Sample across registers or say the sheet is partial.

Quality bar

• Facts, offers, prices, and legal phrasing survive the rewrite byte for byte.
• Every deviation from the original traces to a named axis score or lexicon rule.
• The voice holds across sizes: the button label sounds like the manifesto.
• The sheet is reusable — the next writer could produce matching copy without this session.
• When the sheet and clarity collide, clarity wins and the collision is reported to the guide's owner, because a voice that requires confusion is a voice-guide bug.