For people who write in someone else’s voice
Ghostwriters, agency writers, the comms lead who drafts the founder’s posts, the freelancer with six clients on retainer. The usual way to teach an AI a client’s voice is to feed it a pile of their writing, and the pile is rarely there. When it is, a fair amount of it was written by the last ghostwriter, or by a model.
Studio skips the pile. The client gets a link, answers the same questions anyone building a voiceprint answers, and a full voice file lands in your locker. Ten of them, under one key.
One payment. No account and no password: the locker link is the key. Fourteen days to change your mind.
How a voice gets into the locker
Open the locker
One checkout, and no voiceprint of your own required. The locker link arrives by email and opens on any device.
Send the intake link
It sits on the locker page, ready to paste into an email or a Slack thread. The same link serves every client.
They answer
About fifty-three choices between real passages, roughly twelve minutes. Nobody is asked for a writing sample, and nothing they have written is uploaded anywhere.
It files as a full voice
Everything the $59 file carries: a filter built from the phrases they named, their signatures held exempt from it, the exports cut for ChatGPT, Claude, Gemini and coding agents, the dials and the draft linter. They keep their own copy and their own key, and the locker holds a reference to it.
The brief you hand back
Each voice also renders as a short brief addressed to the person it describes, in mechanics rather than adjectives. Clients rarely get to see this part of the job: their own voice written down as numbers they can hold a draft against.
When a line is wrong they say so, the answer changes, and the file rebuilds from it. That conversation tends to be the useful one.
# Maren Osei — voice brief This is how you write, as far as your answers describe it. Nothing here came from reading your work: it was built from the choices you made in the questionnaire. **The Plainspoken Analyst** You hedge almost nothing. Claims land flat. ## The measurable part - **Sentence length.** Averages 10 to 16 words. - **Swing between sentences.** The longest roughly three to four times the shortest. - **Paragraphs.** Two to three sentences. - **Em-dashes.** About one per 100 words. This is a signature, and it is protected. - **Hedging.** Under one hedge per 100 words. Claims land flat. ## Words this voice keeps Held exempt from every filter above, on purpose: the failure mode, in practice, the load path. [ … continues: the words this voice does not use, and the rest in plain language ]
Rendered by the same function a locker uses, from the answers behind Maren Osei, one of the three voices on /voices. Trimmed for length; nothing reworded.
What the $199 covers
- Voices
- Ten, and each one is a full voiceprint with everything the $59 file has.
- Intake
- One link for every client. A full locker never costs anyone their work: their build still completes and they still get their file, it just isn’t filed with yours.
- Brief
- One per voice, written to the client.
- Licence
- Commercial use, stated in the terms. Bill for the work; there is no per-seat or per-project fee.
- Access
- No accounts. Whoever holds the locker link holds the locker, and a lost link can be mailed back to the address you paid with.
- Payment
- $199 once. Fourteen days, no questions, and the files stay yours.
Ten full files bought one at a time would come to $590.
A whole team writing in one shared voice is a different shape of problem. Get in touch and we will work it out with you.
Worth knowing before you send the first link
The voice is the one the client recognises
The questions ask people to choose between passages rather than describe themselves, because picking the paragraph that sounds like you is easier to get right than naming your own habits. Where a choice still lands wrong, the dials re-cut the file without sending anyone back through the questions.
What the file changes is measured, and so is where it does not
With no file, 0 of 15 drafts landed in a writer’s target sentence rhythm. With one, 13 of 15 did. How large the change is depends on which model you draft with, and on how far the client’s voice sits from that model’s habits. The research page has the numbers, including the arm that told us nothing.