Every other Tuesday I sing at a bar in Southeast Portland with a floor that grabs at your shoes. The host calls everyone baby. Two weeks ago I did Fleetwood Mac, badly, with total commitment, and a guy at the bar knew it was me before he turned around.
He didn't recognize my face. He recognized how I sing. The same wrong notes in the same places. The same half second pause before the chorus because I never remember the timing.
That is the entire voice-first problem, worked out in a sticky room by a stranger holding a beer.
A client asked me last month what their brand should sound like on a smart speaker. They handed me sixty pages of identity guidelines. Fifty eight of those pages were about seeing: logo clearspace, secondary palette, approved photo crops. Two pages were about words, and both of them were adjectives.
Voice strips a brand down to two things most companies have never actually written down, which are what you say and the order you say it in.
Start with the part everyone skips: refusal. Half of what a speaker does is tell someone no. It can't find the order, the item is out of stock, the question was outside its skill. And here is where I keep thinking about HAL, whose voice never changed. Calm, courteous, warm, and identical whether it was reading a chess move or declining to open the pod bay doors. The courtesy was the interface, which is why the refusal landed the way it did. When a voice is your only channel, tone carries the bad news too, and if your warmth doesn't shift at all between good news and no, people learn quickly that the warmth was never about them.
So write your error states first. They are your brand, more than the greeting is.
Then the vocabulary. Not adjectives, an actual list. Words you use, words you never use, and the specific nouns for your own things. Oatly does this in print better than almost anyone, and the reason it works out loud is that the vocabulary is fixed. A voice with no visual anchor has nothing but repetition to build recognition on.
Then length. Every brand I have watched build a skill made the answers too long, because the deck rewarded thoroughness and the kitchen counter punishes it. Twelve seconds is a monologue when the thing talking has no face.
Twelve seconds is a monologue when the thing talking has no face.
I have sat in these rooms from inside brand strategy, and the voice question almost always gets handed to whoever owns the app. It shouldn't. It is a writing problem, and writing is the only part of identity that survives the screen going away.
The guy at the bar knew me by my mistakes, consistently repeated. Your brand should be so lucky.