Voice Matching
How I learned to capture any writer's voice—from Joan Didion to fitness influencers—using examples instead of descriptions. A wizard for building reusable voice skills.
I found Joan Didion on a bench at Clark Kerr, the old deaf school turned Berkeley dorm. I was hungover, washed out from a night of partying, the sun too bright. Something about the title—Where I Was From—with its hint of dislocation made me pick it up.
I spent the next few days in prose that cut like glass. She wrote about California not as paradise but as contradiction. Her sentences were spare and unsparing. I had never read anything like it.
That encounter planted a seed.
Fifteen years later. I'm producing content for a client with a distinctive voice—short sentences, irreverent humor, occasional profanity. His audience knows what to expect. The team had tried every AI tool that promised voice matching. None delivered.
They called me as a longshot. An experiment.
What I learned: the best prompts are not elaborate. Five words carry a whole philosophy.
"In the style of [writer]."
When you invoke Didion, you give AI permission to be opinionated. To eschew the safe middle ground. To adopt her stance of unflinching clarity. The result crackles with specificity. It feels born of a particular consciousness.
But "in the style of" only works if the AI knows the writer.
For famous authors, five words is enough. For your client who runs a fitness empire? For your own voice? The AI has nothing to draw from. You need a different approach.
Most voice matching tools focus on describing voice. Categories. Adjectives. Spectrum sliders. You tell AI a voice is "conversational, witty, medium-length sentences"—and get generic slop.
Examples beat descriptions. Every time.
When you show AI actual sentences from the voice you're matching, it absorbs the rhythm. The cadence. The specific way this writer constructs meaning.
Consider Didion's signature sentences:
"The center was not holding."
"We tell ourselves stories in order to live."
"I have already lost touch with a couple of people I used to be."
No description captures what these examples do. They show the AI exactly what to aim for.
The Voice Matching Wizard is built around this insight. You collect signature examples across every dimension of writing—openings, transitions, closings, sentence architecture, sticky phrases, what they avoid. Then you synthesize them into a reusable skill file.
When I submitted my first draft for that fitness client, I had no idea what to expect.
The feedback: "This is Fantastic!!! I honestly wouldn't change a thing except the title."
The skill file I built has dozens of signature sentences. Each one teaches the AI something no description could convey.
The skill package includes:
SKILL.md— The complete analysis framework
WIZARD.md— Conversational setup with checkpoints
templates/— Blank template for your voice
references/— Example voice skill with signature sentences
Setup takes 15-20 minutes. You need 2-5 writing samples, 500+ words each.
Three ways to use it. Codify your own voice for consistency across newsletters, social, long-form. Build a voice skill before any ghostwriting project. Or analyze writers you admire—extract their signature sentences, absorb their patterns, make them your own.
This post is from the Skill Stack newsletter. Subscribe for weekly skill breakdowns.