
Academy track
Make it sound like you

AI that writes like you, checked against nine numbers measured from your own prose rather than adjectives like warm and punchy.
- 5 steps
- 2 hours 20 minutes in total
- A check you run at the end of every step
Step one is open to everyone, in full. No account, no card.
The other 4 steps are included in Membership, £79 a month.See Membership
By the end of this track you will have
- Step 1One plain text file of your own prose, one reader, one register, with a word count written on it.
- Step 2Nine numbers describing your own prose, computed from your corpus file, saved somewhere you can paste them into a prompt.
- Step 3Six to ten named moves you actually make, each one quoted from a real line in your corpus, and nothing on the list you could not find twice.
- Step 4A 400-word sample written to your profile, a printed gap table against your corpus, and a second version that closed the biggest gap.
- Step 5A closed list of what you are allowed to claim, and a blind audit pass that checks a draft against it and is allowed to return a failure.
The running example
Every worked example follows one invented case from the first step to the last: a solo wedding photographer.
Who it is for
You write to customers every week and every AI draft comes back sounding like somebody else.
Who it is not for
Not for you if you have never written anything a real reader acted on. There is nothing yet to measure.
The steps
Each one ends with a check you run yourself. The box is yours to tick when it passes.
0 of 5 steps ticked, kept in this browser
- Step 1.Collect the writing that was actually for someone25 minutesOpen to everyone
Done means
wc -w corpus.txtreturns a number between 1,500 and 3,000, and reading the first line aloud names a reader rather than a subject. Run it again after every deletion. If the count falls under 1,500 you have collected too narrowly, and if it climbs over 3,000 you have stopped choosing. - Step 2.Measure it, do not describe it20 minutes
Done means
Run the script on your corpus, then on 2,000 words published by somebody else in your field. Their median must differ from yours by two words or more, or their short share from yours by five points or more. If both land inside those margins, your corpus is too mixed to teach anything, and you go back to step 1 and cut a register out.
- Step 3.Name the moves, cite the line30 minutes
Done means
Cover the quotes and read the move names on their own. Search
corpus.txtfor a third occurrence of each move, one you have not already cited. Any move where you cannot find one inside two minutes comes off the list. What survives is the profile, and the same pass run next month will cut more. - Step 4.Write one sample and score it against yourself30 minutes
Done means
The sample sits within two words of your corpus median, within five points of your short rate, and its parenthesis, exclamation and question rates are no higher than yours. Where your corpus rate is zero the sample must be zero as well. Zero means zero. One exclamation mark against a corpus of none is a fail, and you delete it rather than argue with it.
- Step 5.The model will invent your life if you let it35 minutes
Done means
Plant one false claim in the draft on purpose, in your own rhythm, before you run the audit. Something like a client email you never received, and the audit must return that exact sentence marked UNSUPPORTED. If it comes back clean, your auditor is reading the voice. It is not reading the claims. An auditor is worth nothing until it fails that test.
I keep this tick in this browser and send it nowhere. It goes when you close the browser.
Included in Membership
The rest of this track is included in Membership.
Step one is open to everyone, in full. Every step after it opens on a member account.
Included in Membership — £79/month.
