Your Voice Profile: The File That Makes AI Write Like You, Not Like Everyone
Telling an AI to "write in my voice" does almost nothing. The model doesn't know your voice, so it writes in its default one and hopes you won't notice.
What works is a voice profile: a short file that names the specifics of how you write — your sentence shapes, your rhythm, the words you'd never use, the constructions you reach for without thinking. Hand the model that, and it has a target. Without it, you're asking a stranger to imitate someone they've never met.
This is the difference between a request and an instruction. "Sound like me" is a wish. A voice profile is a spec.
Why "Sound Like Me" Fails
A language model writes toward the average of everything it has read, unless you pull it somewhere specific. The average is competent and forgettable. It hedges. It transitions smoothly. It uses the same handful of phrases that have come to read as "written by AI."
When you say "write in my style," you give the model no coordinates, so it stays at the average. It cannot infer that you write in short, clipped sentences, or that you never use exclamation points, or that you open with a flat declarative and let the rhythm build. It doesn't know you hate the word "delve." You have to tell it — concretely, in writing, every time, which is why a saved file beats a remembered intention.
The Five Layers of a Voice Profile
Sentence Shapes
Are your sentences short and stacked, or long and winding with clauses? Do you favor fragments? One-word lines for emphasis? Write down your default shape and your exceptions. "Mostly short. Occasionally one very long sentence when I'm building toward something. Rare fragments, used on purpose." That single description changes the output more than any adjective like "punchy."
Rhythm and Punctuation
Em dashes or parentheses? Semicolons, or never? Do you start sentences with "And" and "But"? Do you use a colon to set up a payoff? Punctuation is half of voice and almost nobody specifies it. Name your habits: "Em dashes for asides, never parentheses. No semicolons. I start sentences with conjunctions on purpose."
The Banned List
The fastest way to sound like yourself is to forbid the phrases you'd never write. Build a list: maybe it's "leverage," "robust," "in today's landscape," "it's worth noting," any sentence that opens with "Whether you're." A banned list does more work than people expect, because most generic writing leans on a small set of crutch phrases. Remove the crutches and the default voice collapses.
Your Tells
Everyone has signature moves. A way of opening — straight into the claim, no warm-up. A way of closing — circling back to the first image. A favorite construction, a recurring metaphor family, a habit of asking a question and answering it in the next line. Name three or four of yours. These are the fingerprints that make writing recognizably one person's.
Real Samples
Paste two or three short passages of your own writing — and label what makes each one yours. The model imitates what it can see far better than what you describe. The description gives it the rules; the samples give it the sound. Two paragraphs with notes beat ten pages of unlabeled text, because the model needs a clear target, not a big one.
Voice is not a vibe you request. It is a set of decisions you can write down — and once written down, a machine can follow them.
The Advice That Wastes Your Time
The common tip is to "feed the AI a huge corpus of your writing and let it learn your style." For most people this is the slow road to a worse result. A large pile of mixed writing — emails, work memos, half-edited drafts — dilutes your voice rather than sharpening it. The model averages your good writing with your rushed writing and lands somewhere bland.
The better move is curation over volume. Pick your best, most characteristic passages, the ones that sound unmistakably like you on a good day. Three of those, plus an explicit description, outperform a hundred unsorted documents. You are not training a system. You are pointing it at a target, and a clear small target beats a fuzzy large one.
The second piece of bad advice is to redo the instructions from scratch each session. Every time you re-describe your voice off the top of your head, you describe it a little differently, and the output drifts. A voice profile you save once removes that drift entirely.
Keep It as a File, Not a Habit
A voice profile only pays off if it's reusable. The instant you have it written down, it stops being something you re-explain and becomes something you paste. Same file, same starting point — in any tool, on any day. Switch from one AI to another and your voice comes with you, because it lives in the file, not in one product's memory.
This is also why a voice profile is worth treating as a real document rather than a throwaway prompt. It's an asset. It gets better as you notice more about how you actually write. And it belongs to you, not to whatever app you happened to build it inside.
Turning Your Writing Into the File
The hard part is the first draft of the profile — reading your own writing closely enough to name what makes it yours. Most people can feel their voice but struggle to describe it in instructions a model can use.
That extraction is what Soul Alchemy is built to do. You paste a body of your existing writing, and it produces structured markdown files that capture not just what you think but how you say it — the patterns, the register, the recurring shapes — in a form any AI can read from the first message. Instead of writing your own voice profile from a blank page, you get one distilled from the writing you already have. It's a one-time build, $99, no subscription, and you keep the files.
From there the loop is simple. Paste the profile, and the AI writes from your coordinates instead of its average. Not a perfect clone — but recognizably you, instead of recognizably everyone.
Frequently Asked Questions
Why does AI ignore me when I tell it to write in my voice?
Because "write in my voice" is not information. The model has no idea what your voice is, so it falls back to its default register. A voice profile fixes this by naming the specifics: your average sentence length, the words you never use, the rhythm you favor, the way you open and close. Concrete instructions get followed. Vague ones get ignored.
What goes into a voice profile?
Five things at minimum: your sentence shapes (short and clipped, or long and winding), your rhythm and punctuation habits, a banned-words list of phrases you'd never write, your recurring tells (favorite constructions, signature openings), and two or three real samples of your own writing. The samples anchor everything else — the model imitates what it can see.
How much of my own writing does an AI need to copy my style?
Less than people assume, if the samples are representative. Two or three paragraphs of your real writing, plus an explicit description of what makes it yours, beats ten pages of unlabeled text. The model needs a clear target, not a large one. A short, well-chosen sample with notes outperforms volume.
Should I keep a voice profile as a file or just retype it each time?
Keep it as a file. Retyping instructions every session guarantees drift — you'll describe your voice slightly differently each time, and the output will wander. A saved profile is consistent across sessions and across tools. Paste the same file into any AI and you get the same starting point, every time.
Can a voice profile make AI writing sound less generic?
Yes, mostly by subtraction. A lot of generic AI writing comes from a small set of tic phrases and a flat, hedged rhythm. A voice profile that bans those phrases and specifies your real cadence removes most of the "AI smell" on its own. The positive instructions add your character; the banned list strips the default character out.
Turn Your Own Writing Into a Voice Profile Any AI Can Read
Soul Alchemy reads your existing writing and produces structured markdown files that capture how you actually write, ready to paste into any AI. $99, no subscription.
Try Soul Alchemy