toolkit

Brand voice profile & audit · Playbook

Brand Voice Profile Standard

The toolkit.brand-voice/v1 JSON profile is the source of truth shared by primarina and zoroark. Store the approved copy in the team wiki as <Brand> Brand Voice, or <Brand> Brand Voice (<locale>) for a language-specific profile, marked AI-Safe: Yes. Keep a local export beside the migration evidence for reproducibility.

Approval workflow

  1. Initialize a Draft profile.
  2. Have content/brand ownership confirm audiences, traits, terminology, capitalization and examples.
  3. Have clinical/legal owners confirm any safety notes that constrain claims.
  4. Change wiki status only through the normal human review process.
  5. Pull the approved profile and run primarina before accepting a rewrite or migration candidate.

Profile design

  • Write traits as observable guidance, not adjectives alone.
  • Use preferredTerms and forbiddenPhrases only for rules that can be applied literally.
  • Set paragraphMaxWordsMobile to the largest paragraph that remains comfortably scannable on a phone.
  • Use pageTypes for real exceptions; do not fork a whole profile for every page.
  • Keep proper nouns exactly cased in properNouns.
  • Derive multilingual sites one locale at a time. English and Spanish pronoun/hedging measurements use explicit dictionaries; other locales keep those three measurements unset until a reviewed dictionary exists. Do not treat English-only counts as evidence about translated copy.

The audit deliberately leaves subjective voice assessment to a reviewer. Zoroark can draft copy in the approved voice, but it is output-only and needs a factual/editorial review before implementation.