Brand voice

Two settings decide how your product sounds in every language: a style guide and, on higher plans, a fuller AI profile.

Owners and admins.

The project's Settings → Voice holds two sections, in a deliberate order.

Style & voice comes first: "Tone, formality, and brand voice the AI follows when translating this project. Set a project-wide base and per-locale overrides." You fill in Formality, Tone, Brand voice, Regional notes and Additional guidance, then Save style guide. Per-locale overrides let German be formal while Spanish is not.

If your plan allows it, Infer from reviewed translations reads the translations your team has already approved and proposes a guide from them. What comes back is a draft: "Review and edit it below, then save. Nothing is applied until you save."

AI profile is the second section, on Team and above. It goes further: an Audience, its own formality and tone, a longer Brand voice description ("Describe how your product sounds: personality, dos and don'ts, words to prefer or avoid."), and Style examples: real source/target pairs, each with an optional note, that show the model what good looks like rather than describing it.

The profile's Grounding switches decide what the model is given alongside your text: Use glossary, Use style guide and Use semantic translation memory. The two sections share a page precisely because Use style guide governs the guide above it: turning it off silences the guide.