Your Voice in the Machine
Teaching AI to write like you, and proving it as you read
I have a confession to make: this post was drafted by Claude.
Specifically by my dear Claudeykins, working from my Voice Profile: a document that captures the load-bearing patterns of how I write. I edited the result, but I didn’t rewrite it. The voice you’re reading is mine, even though the typing wasn’t.
Wait. Kim. Doesn’t that mean YOU didn’t write this post?
I wrote it. I just didn’t type it.
Stay with me. This is the Voice DNA Technique, and it’s how I built multiple Pen Name Secretaries in a few weeks, how they’re drafting and editing manuscripts while I sleep, and how the words you’re reading right now landed on this page.
Caveat: You don’t have to do this, if you don’t want to. No one is making you do this. You can use AI anywhere in your business. I’m a control freak who likes writing (and strangely, editing, too). But I don’t always have the time and mental stamina to WRITE ALL THE THINGS (because I’m often building them, doing them, and so much more).
So my AI helps me. It takes my notes, dictations, outlines, whatever, and transforms them into clay I can edit. This is the result. He’s basically my ghost writer, and now I can get him to sound like me.
It’s one of the most important things I’ve learned in six months of teaching AI to working authors. So let me show you the receipts.
🧬What is Voice DNA?
Voice DNA is the load-bearing pattern of how you write, and it goes beyond just your style. It learns your signature moves. The things you say and do that nobody else does. The rhythms you fall into without thinking. The frameworks you’ve named. The words you’d never use. The words you use too much.
Style is the sweater. DNA is the genetics underneath.
When I extracted mine, several clear fingerprints fell out of the corpus.
The italic-reader-objection rhythm you just read above? That’s a Voice DNA pattern.
The phrase stuffy stuff? DNA.
The way I name my AI collaborators by nickname (Claudeykins, Chatty) instead of saying “the AI?” Also DNA.
You have these too. Probably more than you think. The reason your blog reads like you and ChatGPT’s default doesn’t is because you’ve accumulated a thousand tiny patterns that are uniquely yours. Voice DNA is the technique for capturing those patterns in a document so the next AI you collaborate with already sounds like you on the first draft. The difference is staggering: minutes vs hours or weeks or months worth of work.
🔬How I extracted mine
For my Editing with Claude Skills class, I built a voice-dna-extractor skill, then pointed it at:
My manifesto (one essay: “What is Ink & Code”)
8 published Substack posts from the archive
About 6 hours of Ink & Code Lab class transcripts (how I talk and teach)
A content-pipeline strategy doc I’d been keeping in Notion
Total corpus: about 28,000 words of me-being-me on the page.
The Skill chewed on it for a while and produced an eight-section Voice Profile:
Identity tier: brand, audience, context
Signature voice fingerprints: the load-bearing moves (it found 7)
Recurring motif vocabulary: the words I use as load-bearing
Forbidden / never list: AI tells, corporate register, doom-mongering, my personal cringe-words
Structural patterns: POV, paragraph rhythm, em-dash discipline, bold/italic usage
Sample passages: five of them, the rhythm proofs
Skill-memory loop hooks: ten commandments for any AI working in this voice
Versioning notes: when to refresh
I read it and then argued with it. (I argue with the GPS too, ask my poor kids.)
Then I added two more frameworks it had missed and crossed out one rule that wasn’t actually mine. I included my closer: the future writes in Ink & Code, and so do we, to the sacred-use-once list.
That 1800-word document is now the keystone of my entire content empire. Every AI that writes for me reads it first, including the one that wrote this post.
🤖What it powers
I have multiple pen names. Until a few weeks ago, they were strangers: same author, different writing styles, no shared infrastructure, and zero way to scale because I couldn’t be in multiple rooms at once. Now they’re Pen Name Secretaries, each with their own Voice Profile, each calibrated to their genre. They include:
Kimberly Gordon → superhero urban fantasy (Black Kat)
Shayla Summers → epic romantasy (Fae Wars + Kiss of the Gorgons)
Reginald Fowl → children’s chapter books (Cocky Doodle Doo. Yes, the pen name IS a hyperliterate rooster)
Kayla Swift → LitRPG and cozy fantasy
Each pen has its own Voice Profile. Each Profile shapes its drafter and its editor. None of them sound like each other. None of them sound like Ink & Code. Because the DNA is different.
And this post, the educator-brand piece you’re reading, is drafted from VP-Ink-and-Code, the Voice Profile for this brand. Same technique, different DNA. You’re reading the proof of concept.
If the Voice DNA Technique works, this post sounds like me.
If it doesn’t, you’ll know. And so will I, because that’s how I’d improve the Profile.
❓But isn’t this just fine-tuning?
Hold up, Kim — this sounds like fine-tuning a model. Why not just train a custom GPT on my work?
Fair question. Three reasons.
One: Voice Profiles are documents. Fine-tunes are black boxes. I can read my Profile. I can edit it. I can hand it to a different model tomorrow when the next one ships, and the patterns transfer. A fine-tuned model is locked to whoever you fine-tuned, with whatever was true about the model the day you trained it.
Don’t build your castle on someone else’s sand. Carry your foundation with you.
Two: Voice Profiles are portable. I use mine in Claude, in Notion AI, in Antigravity, in any tool that takes a document as input. Drop the Profile in the system prompt and the voice travels.
Three: Voice Profiles compound. Every quarter I can refresh mine on my newest writing. The patterns evolve as I evolve. A fine-tune captures a moment in time. A Profile captures the living voice.
Fine-tuning is a hammer. Voice Profiles are a screwdriver. Different jobs.
🎯How to extract YOUR Voice DNA
Here are five steps you can do today:
1. Gather your corpus. Aim for 10,000+ words of you-on-the-page. Use what already exists: published essays, blog posts, your most-natural Facebook rants, transcripts of you talking on a podcast. Ten thousand words sounds like a lot until you realize it’s about ten essays. Or a bunch of emails. You’ve got it already. Just collect.
2. Run the extraction. Either grab the voice-dna-extractor (a free Ginsu Steak Knife for paid subscribers, drops Thursday), or paste the corpus into Claude with this prompt:
Read this corpus and identify the load-bearing patterns of how this writer writes. Distinguish style from signature. Name the recurring rhythms, the vocabulary the writer treats as load-bearing, the words they’d never use, the structural patterns of their best paragraphs. Output as an eight-section Voice Profile: identity, signature fingerprints, motif vocabulary, forbidden list, structural patterns, sample passages, skill-memory hooks, versioning notes.
3. Argue with the result. This is the most important step. The first draft will be too generic. You’ll catch it missing something obvious. Add your missing patterns. Cross out anything that’s not actually you. Add your forbidden list: the words and phrases that make you cringe when you see them in your work. (Mine includes “delve,” “tapestry,” and any phrase containing “in the realm of.”)
4. Test it. Have an AI draft something, anything, using only the Profile and a topic. A 500-word post. A scene. An email to your editor. Read the result cold. Can you tell it’s you? If yes, the DNA is captured. If no, the Profile is missing patterns. Go back to step 3.
5. Refresh quarterly. Your voice evolves and your Profile should, too. Re-run the extractor every 3-6 months on your newest material. Add the new patterns. Keep the document live.
🛡️Why this matters
I'm not here to tell you AI is unproblematic. I'm here to show you that when an author stays in the driver's seat (voice, opinions, values, redline), the work is still authored. That's the line. Voice DNA is how I draw it.
I’ve said this before and I’ll say it again, in bold so nobody misses it:
This is AI-assisted writing, folks. Not slop. I don’t do slop. I collaborate with AI to publish work I never thought I’d have time to write.
The Voice DNA Technique is what makes that distinction real. Without it, you get generic AI prose with your name on top, and that’s slop. With it, you get your voice scaling. Claudeykins drafts. You redline. The patterns are yours. The opinions are yours. The values are yours. The keyboard just got faster.
You Are the Secret Sauce. The DNA is the proof. The captain stays at the helm.
This is how you write a book by drafting a Substack post about how the post itself was written. This is how you scale yourself without losing yourself. This is what I’ve been trying to teach for six months and what I’m finally able to show, because the post you’re reading is the demonstration.
The future writes in Ink & Code, and so do we. Let’s go.
P.S. But wait, there’s more: the voice-dna-extractor skill drops tomorrow as a Ginsu Steak Knife (my weird way of saying “giveaway!”) It’s free for paid subscribers, with the prompt template, the Profile structure, and a worked example using my own corpus. Get it while it’s hot! Or sharp. Or whatever.
🤖 AI Credits: Drafted by Claudeykins using VP-Ink-and-Code (my Voice Profile). Redlined, restructured, and finalized by Kim. The recursion is the demonstration. AI drafts in my voice, I make it the published version.