What is brand voice?

The distinction matters because the two are managed differently. Voice is decided once and defended. Tone is chosen per piece: the same brand is warm in a launch announcement and careful in an incident notice, without becoming a different brand in either.

Confusing the two produces content that reads as inconsistent even when every individual piece is well written. A brand that sounds irreverent on X and corporate on LinkedIn has not adapted its tone. It has two voices, and readers notice.

A language model with no brand context returns the statistical middle of everything it has read, which is why unguided AI content has a recognisable sameness. Adding voice instructions to each prompt helps, but only for as long as someone remembers to add them, writes them the same way every time, and updates every prompt when the voice changes.

Drift is therefore a systems problem rather than a writing problem. The fix is to store the voice once, as configuration, and have every generation read it. A brand profile that lives on the account cannot be forgotten, retyped inconsistently, or lost when the person who wrote the prompt leaves.

Most brands have a voice without a document describing it, because the voice lives in published work. That work is the source: an existing site, a set of posts that performed, the pages a founder wrote personally. Reading the corpus is a faster route to an accurate profile than filling in a brand questionnaire, which tends to produce aspiration rather than description.