Brand voice
The consistent way a brand sounds in writing and speech: its vocabulary, sentence rhythm, level of formality, humour, and the things it refuses to say. Voice stays broadly fixed across every channel and situation, which is what lets a reader recognise the source of a sentence before seeing any name attached to it.
Brand voice is not Tone of voice.
Voice is constant and tone is situational. A brand has one voice and many tones: the same underlying personality can be warm in a welcome email and plain and unadorned in an outage notice, and if the personality itself changes between the two, that is not tone, it is inconsistency.
In practice
Voice is the part of a brand that appears most often and gets defined least. You publish far more words than logos, and most people meet your sentences long before they notice your palette. Yet voice is usually recorded as three adjectives on a slide, and three adjectives cannot settle an argument about whether a specific sentence is right. The working test of a voice definition is simple: given the same brief, would two different writers produce recognisably the same paragraph?
The persistent confusion is with tone. Voice is who the brand is; tone is the mood it is in. Your voice does not change when you announce a price rise, apologise for an outage, or celebrate a launch. The tone does, and it should, because a brand that sounds identically cheerful through a refund request reads as a machine. Treating them as one thing produces either a voice so broad it constrains nothing, or a voice so rigid the brand cannot be serious when it needs to be.
Making a voice concrete means writing down things that can be checked. Whether you use contractions. Whether you write as "I" or "we". How long a typical sentence runs. Which words you use constantly, which you have banned, and why. How much you hedge. Whether jokes are allowed and where they are not. The single most useful artefact is a set of before-and-after rewrites: the same sentence written the way a generic tool would write it, then the way you would. Nobody argues with a pair of examples the way they argue with an adjective.
Voice is also the first thing to erode under volume, because writing in voice is harder than writing competently. The first post in a batch sounds like you and the fortieth sounds like whoever was fastest. This is where a written definition stops being a document and becomes infrastructure: it is the thing a collaborator or a generative tool reads before drafting, rather than something they are expected to have absorbed. Repic's AI twin works from your own writing and, for spoken voice, only from a recording you consented to and can delete.
One caution. A voice definition describes how you sound when you are being yourself, not a costume you put on. If the written definition and your actual best work disagree, the definition is wrong. Update it from the sentences you were proudest of rather than from the ones you thought you were supposed to write.