Skip to content

Part 1 · Understand

Why voice is the whole product

What makes generated writing sound generated, and what the product does about it.

01What it is

The product learns how you write from samples you paste in, then scores every draft against that profile and refuses anything below the floor. Voice is a gate, not a preference.

02Why it exists

Generic model output has a recognisable shape. It opens by restating the question, it balances every claim against its opposite, it reaches for "unlock" and "seamless", and it says nothing a competitor could not say.

Buyers recognise it now. A post that reads as machine-written costs more than no post, because it tells the reader that the person behind it did not care enough to write it themselves.

Nothing in a model fixes this on its own. What fixes it is a specific person's phrasing, specific claims that only you can make, and a gate that refuses the draft when it drifts. All three exist here, and the gate is the part that holds when the model has a bad day.

03How to use it

  1. 01Paste three or more real samples of your own writing during setup. Emails and posts you actually sent.
  2. 02Add your banned words. Anything you would not say out loud belongs on that list.
  3. 03Add your forbidden claims. Things that are not true of your product, and things you are not allowed to say.
  4. 04Read the first few drafts closely and reject the ones that miss. Rejections feed the revision loop.

04What good looks like

You give it
Four samples totalling 900 words, taken from real LinkedIn posts, plus six banned words the founder never uses.
You get
Drafts carry the founder’s sentence length, their habit of opening on a specific number, and none of the banned words. Voice scores land above the contract floor.
Why
The profile is built from enough real text to characterise sentence length, opener habits and vocabulary rather than guessing from two lines.

Derived from the rules in the code

05What weak looks like, and what it costs

You give it
One sample, 40 words, pasted from a company About page.
You get
The voice profile does not build. Drafts run in a neutral register and read like every other tool’s output.
Why
The profiler needs a floor of samples and words before it will characterise anything, and marketing copy from an About page is not how you write to a person.

Derived from the rules in the code

06Read next

Why voice is the whole product · InstinctGTM