The Voice Fingerprint
Writing Quality Measured as Data
BY ANDY HOUSTONContext
A family-portrait studio with campaign copy written by AI and rejected the day it landed. The client could not name what was wrong with it, and never needed to. Their customers would have felt it too.
Problem
The copy was statistically unlike the client's real voice, and there was no way to argue about it productively. Voice disputes die as taste debates: one person feels the slop, another defends the draft, and nobody can point at anything. A rejection that is correct and unexplainable is also unfixable.
Approach
Measured the client's writing across hundreds of real assets: lexical range, punctuation habits, sentence rhythm, sentiment shape, nine metric families in all. The measurements resolve into a fingerprint, a numeric profile of how this one business sounds. The campaigns got rebuilt against that fingerprint, and the before-and-after shipped as a drag-to-compare artifact, so the client could slide between the rejected copy and the rebuild and watch the distance close.
Drag-to-compare echo: a slider revealing rejected copy against the rebuild.
The before-to-after, silent-legible: where it started, what the data showed, the decision that turned it, where it landed. Past performance guarantees nothing. The diagnosis method is the product.
Stack
- Python metric engine, nine metric families spanning lexical, rhythm, punctuation, and sentiment axes
- Corpus extraction across hundreds of client-authored assets
- Drag-to-compare interactive deliverable
Result
Campaigns rebuilt and accepted. The artifact doubled as its own proof: nobody had to trust a writer's ear, because the client could read the numbers and slide the comparison themselves.
Distance-to-reference gauge closing as the rebuild lands.
Impact
Writing quality at this studio became an inspectable number rather than a taste argument. When a future draft drifts from the voice, the drift shows up as a measurable distance before a customer ever sees it.
Lessons
AI-written content is converging on one voice, a new dialect of the internet. Conversion copy that works in the coming decade gets written in a way no other agent writes and most humans don't think, and the only way to prove you're doing that is to measure it.
Why this matters to you
For anyone whose brand voice is the asset: this is what defending it with instrumentation looks like.
Sounds like you? Get a quote