I’ve used Claude for a lot of things at this point—brand work, client projects, research, the French conversation prep system and Cycle Tracker app I wrote about on the blog. This one is different. Building a system that writes in my own voice, across multiple formats, with enough consistency to actually be useful professionally, required a different kind of thinking.
Getting Claude to replicate your own written voice is genuinely harder than briefing a brand voice, because with a brand you often have the foundational work already done—positioning, audience, approved examples, tone guidelines. With your own writing, the raw material exists but the documentation of it doesn’t. You have instincts. You know when something sounds wrong. But instincts aren’t a brief.
Going bottom-up
The only way to build a reliable brief from your own writing is to extract it from real examples rather than describe it from scratch. Introspection produces your ideal self—the writer you think you are. Real work produces actual patterns: the structural habits, the recurring moves, the things you reach for without thinking.
So that’s where I started. Real pieces, treated as source material.
Why I used older writing deliberately
I wanted this system to capture my natural written style including how it’s evolved over time, so I deliberately chose examples that span years—pre-AI supported writing that reflects how I actually developed as a copywriter, not just what I produce now.
The older pieces are part of the picture, not noise. They show the evolution. At some point I’d like to link to the specific examples I used as reference, partly for transparency and partly because I think it’s interesting to see which pieces felt most representative.
Knowing which examples to use and what each one is actually demonstrating is a craft decision. It requires looking at the work clearly enough to know what’s worth capturing.
One project, not four
Before this, I had separate Claude projects for different writing contexts—one for my Substack, one for website copy, social, and so on. It works, up to a point. The problem is maintenance. Every time something shifts in the way I write, every time I look at a piece and think that’s closer to it than anything I’ve done before, I’d have to update multiple places. The voice fragments across projects instead of evolving in one.
What I built instead is a single project with a master document that has two layers: core constants that hold across every format, and context layers that adjust the specific dials depending on what I’m writing. Formality shifts. How much technical detail is appropriate shifts. The register moves. But the underlying written identity lives in one place and updates once.
How it works in practice
The first prompt does the heavy lifting—you tell Claude what you’re writing, for which medium, what it needs to do—and it reads the instructions and pulls the relevant context layer automatically. A Substack piece and a website landing page pull from the same core document but surface different things from it.
This is what makes the single-project architecture worth the upfront thinking. The system stays coherent as the writing evolves, without having to maintain four separate versions of the same truth.
Check out what other projects I’m working on: