You paste in a few paragraphs you actually wrote, tell the model to write the next section in your voice, and wait.
What comes back is fluent. Competent. Cleanly structured, grammatically spotless, and written, unmistakably, by no one in particular. It sounds like an AI doing an impression of a writer — any writer — and the one thing it doesn't sound like is you.
That gap is the whole problem. The model can mirror your topic and copy your structure, but it keeps missing the fingerprint: the rhythm of how you actually talk on the page, the words you reach for, the ones you'd sooner die than use. Closing that gap is what makes a model sound like you instead of like the average of everyone.
Why "write like me" never quite works
The model writes for the middle of everyone who ever typed a request like yours, and the middle is, by definition, generic.
Your voice is the opposite of that. It's the specific — the particular way you build a sentence, the joke you'd make here, the turn you'd never take. So when you tell a cold model to "write like me," you're asking it to reproduce something specific from almost nothing, and it does the only thing it can: it guesses, politely, at a version of you. Then the chat ends and even that guess is gone. Tomorrow you start over.
What "your voice" is actually made of
Here's the part most people skip. Voice isn't a vibe you gesture at — it's a set of concrete, repeatable patterns, and patterns can be written down.
It's how long your sentences run and how you vary them. The words you reach for and the ones you avoid. How you open a piece, how you land it, where your humor lives, the punctuation you lean on and the punctuation you never touch. Name those, and the model finally has something real to match instead of a mood to approximate. The thing is, you already know most of them — you've just never said them out loud.
How to make AI sound like you
You don't teach voice with a better adjective. You teach it with evidence.
- Gather real samples. A few things you actually wrote, in the register you want — not your most polished work, your most you work.
- Name your patterns. Sentence rhythm, your go-to transitions, the words you reach for, where the humor lives.
- List your never-words. The corporate filler and AI tells you reject, spelled out so the model knows what to avoid.
- Show, don't just tell. Hand it the samples next to the rules. Examples teach voice better than any description of it.
- Save it where you can reuse it. Somewhere portable, so you're not re-teaching your voice from scratch every single session.
Do that and the next draft comes back sounding like it came from your desk, not from the internet's.
The catch: voice is the easiest thing to lose
One good prompt works exactly once.
Tomorrow's blank chat doesn't remember the voice you taught it yesterday, and the careful custom setup you built inside one tool stays locked in that tool. Open a different model and you're a stranger with a generic accent again. If you want the longer version of why context keeps evaporating like this — and the handful of ways people try to fix it — it's the same trap, just pointed at your voice instead of your facts.
Where RUMO fits
This is exactly what the Writing Codex is for. It's your voice, written down once: the rhythms, the vocabulary, the never-words, the gap between your LinkedIn voice and your text-to-a-friend voice. You build it a single time and drop it into whatever model you're using that day, and the generic impression turns into something that actually reads like you.
It's one of six context anchors — durable pieces of who you are that any AI can read. The Personal Constitution is the free place to start, but the Codex is the one that fixes the voice.
Start with one
Go back to that hollow, competent paragraph the model handed you. Now imagine it opens already knowing how your sentences breathe, so the first draft sounds like a sharper version of you instead of a stranger doing your job.
You don't get there with a cleverer prompt. You get there by writing your voice down once and bringing it with you. Start free with your Constitution — about thirty minutes — then add the Writing Codex when you're ready to make every model sound like you.
Frequently Asked Questions
- How do I make ChatGPT sound like me?
- Give it your voice on purpose instead of hoping it guesses. Gather a few real samples of your writing, name the patterns that make it yours — sentence rhythm, the words you reach for, the words you'd never use — and hand the model both the samples and the rules. Then save that voice profile somewhere reusable so you're not re-teaching it every session.
- Why doesn't 'write like me' work on its own?
- Because the model writes for the average of everyone, and a single 'write like me' prompt gives it almost nothing to copy. It's guessing at your voice from one example, and it forgets the moment the chat ends. Your voice is specific — to match it, the model needs the specifics written down, not a one-time instruction.
- Can AI actually copy my writing voice?
- Yes, surprisingly well — but only when you give it the raw material. Voice isn't a vibe; it's concrete, repeatable patterns. Show the model real samples and spell out your rules, and it can match your rhythm, vocabulary, and tone closely. Hand it nothing, and you get the generic default.
- What is a Writing Codex?
- A Writing Codex is a structured profile of how you write — your sentence patterns, favorite words, the words you avoid, your openings, your humor, your punctuation habits. It's one of RUMO's six context anchors. You write it once and drop it into any AI tool so the model stops writing like the internet and starts writing like you.
- Do I have to teach my voice to every AI tool separately?
- Only if you set it up tool by tool. If you capture your voice as a portable document — a context anchor — you write it once and paste or wire it into ChatGPT, Claude, Gemini, or any agent. The voice profile travels with you instead of living inside one app.
