Open app
Moonborn — Developers

Multi-locale brand voice — keeping the through-line across six markets

Brand teams expanding into new markets inevitably watch their voice dilute. Lineage-tracked locale forks + a custom distinctiveness baseline keep variants close to the canonical brand.

The brand-at-scale failure mode is consistent: a sharp EN voice in year one, a press-release DE voice by year three, a uncanny-valley TR voice by year five. The dilution is gradual and rarely catches anyone's attention until a competitor's voice starts feeling sharper.

This post is the brand-team-side reasoning behind the locale-fork pattern Moonborn supports.

The dilution mechanic

Each new market involves translators, content strategists, marketers who weren't in the room when the original voice was decided. Each one rounds the corners. By the time a campaign hits the third locale, the corners are gone.

The traditional fix is a "brand voice document" — a 12-page PDF listing tone principles. It's the right idea, wrong artifact. PDFs don't constrain content production; they get read once at onboarding and ignored thereafter.

What lineage forks fix

A canonical brand persona — the voice as authoritatively as you can articulate it — sits at the root of the lineage tree. Every locale forks from it, locking Soul + Self + Mask, refining Surface for language and cultural details only.

The fork keeps the Soul + Self + Mask as Moonborn objects, not PDF descriptions. When a Spanish content marketer asks an AI tool to "write in our brand voice," they're hitting the Spanish persona — a fork of the canonical — and the voice is enforced at the model level, not at the translator's discretion.

Distinctiveness against your own brand

By default, distinctiveness scores against chatgpt-default — "does this persona sound like a generic assistant?" For brand-governance work, set the canonical brand persona itself as the baseline:

await client.config.setItem({
  key: 'consistency.distinctiveness.custom_baseline_persona_id',
  value: canonicalBrandId,
  scope: 'workspace',
});

Now every variant is scored against your brand. A score of 0.85 means "very close to the brand"; 0.40 means "drifting toward generic." Set min_score = 0.65 to require variants stay close.

The audit trail

Every refine on every locale variant creates a version in the lineage tree. The full history is diffable in the product UI. With Git sync, it's also diffable in your repo as YAML + markdown.

Brand governance becomes a pull-request review pattern. The Spanish marketing manager refines their locale variant; the change shows up as a diff; brand leadership approves or rejects.

What this doesn't solve

The actual content production — the campaigns, the support replies, the social posts — is still a human + LLM workflow. The locale variants don't auto-generate copy. They keep the copy from going off-brand once a human or downstream tool produces it.

Think of the variant set as a brand voice enforcement layer that sits between content production and content delivery. It catches the drift; it doesn't replace the writers.

Patterns from production

  • One canonical, four locales is the working point for most customers. EN canonical + DE / FR / ES / PT locales.
  • Two tones per locale for B2B customers (formal for sales, casual for community).
  • Pairing with Moderation — brand-safety classifiers per locale, customized for local cultural sensitivities.

Next