Every brand running on AI tools ends up with the same average voice, because the tools learned to write from everyone at once. We measure how your brand actually sounds on 8 channels, formality through to boldness, and score each one from 0 to 10. Then we write the rules your AI has to follow to keep your voice intact, draft after draft. Install them once and everything your AI writes sounds like you, launch day and refund note alike.
When nobody engineers the voice settings, AI writes from the statistical average of the internet. Competent, but meh. Invisible in one email. Devastating across thousands, because every send teaches your customers you sound like everyone else. And once you sound like everyone else, price is the only reason left to pick you. With a solid brand voice, people recognise your email before they see the logo, and buy before they've compared a single price.
Hi [First Name],
We noticed you haven't shopped with us in a while! Come back and enjoy 15% off your next order.
Don't miss out. Shop now!
Hi [First Name],
We noticed you haven't shopped with us in a while! Come back and enjoy 15% off your next order.
Don't miss out. Shop now!
These are not the same screenshot. That's the problem.
"Friendly, bold, authentic" means nothing to a language model. (Your customers can't measure it either, btw.) So we stopped describing voice and started measuring it with eight channels, zero to ten, set like faders on a mixing console. Every position is enforceable in copy. Every position is checkable in whatever your AI writes.
How dressed up the language is. F2 texts you. F9 sends its regards.
The size of the words. L2 says use. L9 says utilize, and means it.
The rhythm of the sentences. Percussion at one end, strings at the other.
How close you stand to the reader. Firm handshake, or a hand on the shoulder.
Who's talking: a fellow traveler, or the expert in the room.
The voltage on the page. Spa-quiet to can't-skim-past.
How much play is allowed, and what kind. Sincere, wry, or full on wink.
How many fights you'll pick, and with whom.
Hi [First Name],
We noticed you haven't shopped with us in a while! Come back and enjoy 15% off your next order.
Don't miss out. Shop now!
Okay, we have to ask. How's your skin doing without us?
(We think about you. That's not weird.)
Your barrier called. It wants the ceramides back. 15% off says welcome home.
The spec is the difference. Your AI is exactly as distinctive as its instructions.
Spoiler alert: this ends with a voice guide your humans will actually use, and a machine spec that pastes into your AI stack whole. Klaviyo, ChatGPT, Claude, your agency's agents, whatever ships next.
We analyze what you actually send. Sentence length, hedging, warmth markers. The truth, measured.
A live identity and voice workshop. Word sorts, forced choices, gut faders. Recorded, always.
We compare three things: what you declare, what you demonstrate, what you want. Then the verdict.
The Human Guide and the Machine Spec, versioned and dated. Written in your voice, as proof.
Installed in your stack together, then calibrated for 30 days on the copy you actually write. You log what you changed, and a patch lands every Friday.
It lives in the founder's instincts and the best 20% of your copy, and gets diluted by every new writer and tool. We measure it, fix its contradictions, and encode it.
We build it from the founder's vision and anchor it in how you actually talk so it's sustainable. Or as a designed character your system performs. You approve the copy (not the numbers behind it).
Brand voice isn't an exact science. I turn it into numbers anyway, because numbers are what the machines can follow. The humans reading your copy (you included) still catch things the numbers miss: the word that's a shade too cheerful, the joke that lands a beat late, the sentence that's nearly you. Those nuances show up in your edits, and the calibration period exists to collect them. For 30 days after install, you write with the spec, I read what you changed, and every Friday the spec moves a little closer to you.
The spec has to be installed in the tools writing your copy. A spec in a drawer sounds like a drawer.
Sort a few words, answer six sharp questions, and watch your brand land as a coordinate on the console, with a distinctiveness score and your voice type. (Storyteller, Maverick, Confidant, and seven more.) It's free and your result opens the Starter Spec.
Pick the option that matches how deep the problem goes.
What you get: your voice type, your eight-channel print, and a full personal reading of what every number means for your copy. Think of it as the tasting: a real, measured pour of your voice, with the Starter Spec as the full bottle.
The quiz reads what you say about your own voice. It hands you your numbers and stops there. Writing the rules your AI follows is the Starter Spec's job.
The wrongness you can't name gets a name. That alone is worth three minutes.
I want the free voiceprint quizWhat you get
What you get
The included patches come first: 14 days of them on the Starter, 60 days of calibration and cover on the Session. After that, if a model update ever shifts the voice, reinstall the spec fresh, and it's back at full strength. Done-for-you Deployment of your existing email flows exists too, and we talk about that after your Session, not before. (Prices are in USD, and your bank handles any conversion at checkout. Prices rise as the founding 100 fills, and rises get announced first. There are no countdown timers here, because I don't do those.)
If you can't get enough of this, sign up to my email list. I talk about how brands lose their sound, and how they get it back.
NO SPAM, NO SELLING YOUR ADDRESS, UNSUBSCRIBE ANY TUESDAY. PRIVACY