Most AI writing advice starts off simply wrong:
The above said to an AI that has never read you, that is usually slop.
You can tell it to write like a famous writer. Maybe that helps a little, but what you get is the average of how the training set imitates them. Not them, not you either.
Here is the interesting question: what changes when the model already knows how you write before it generates a single word?
I ran an experiment using the same brief and the same LLM model. The brief was simple: write an essay opening on “AI is computation.”
On the left, the model writes with no profile. On the right, it writes with a voice profile extracted from Gabriel Pickard’s writing, with Gabriel’s permission ofc.
The difference is visible in the first sentence.
The no-profile version starts like a competent explainer:
▎ There is a temptation to describe modern AI as if it were mostly a new kind of text interface...
That is clear, maybe even useful. It is the default register of a good frontier model: define the topic, frame the issue, proceed carefully.
The profile-guided version starts differently:
▎ There is a framing of large language models that treats them as oracles...
That opening already has a writer inside it. It moves differently, names the conventional frame, turns it into a mechanism, and sets up a contrarian landing. It does not just use different words like a Humanizer.
That is the part people miss when they talk about AI writing style.
Voice is not tone or vocabulary, it is not sprinkling in favorite phrases. It is structural: how you open, how you pivot, where you put pressure, what kind of claim you are willing to make first, how you move from abstraction to example and back. A voice profile gives the model those patterns before generation begins.
A prompt tells the model what to do, a profile tells it who is doing the writing.
Full comparison:
https://usenoren.ai/sample-voice-profile#generation
Disclosure: this post was written with my Noren voice profile. If it sounds like me, that’s the experiment working. Subscribe if you want more.
Until next time.