Lately I keep getting the same worry, almost always in the same words: “I started using AI and my content began to sound like everyone else’s.” People with a good, worked-in voice who suddenly look at what they publish and don’t recognize it. It sounds correct. It sounds clean. And it sounds like anyone.
It’s a fair fear, and the easy answer would be to not use it. But giving up a tool that solves the form at this speed isn’t judgment, it’s fright. The good question isn’t whether to use it, it’s how, so it adds without eating the thing that makes you you.
I’ll give it to you in one line and then take it apart: AI solves the form at a speed that was impossible three years ago, but the voice, what you say and where you say it from, is still yours, and it all comes down to the order you bring them in.
In short (TL;DR)
- AI is great for the form (producing, varying, accelerating) and bad for the substance (what you say, from where, and why). The problem isn’t using it, it’s handing it the part that was yours.
- It sounds generic because AI is trained on the average of what exists: if you ask it for the what-to-say, it gives you the mean, and the mean sounds like everyone.
- Order changes everything: if you start from the blank prompt, you come out at the mean; if you start from your substance and use AI only for the form, you keep your voice.
- Feed it your own material before you ask it for anything. It hands you back a version of what you give it.
- Use it for drafts, not for decisions. Your voice returns in the edit, cutting more than you add.
- Keep the last mile human: the line you’re risking, the final call, the part that costs you. That doesn’t get delegated.
What “losing your voice” to AI means
Losing your voice to AI is letting it decide, for you, the part that set you apart. It doesn’t happen all at once. It happens out of convenience, asking it for a little more each time: first help with the form, then a suggested angle, then the whole text from a one-line idea. And the more you ask, the closer you get to one very specific place: the center.
Ted Chiang put it in an image I can’t shake: AI is like a blurry copy, a compressed JPEG of everything on the internet. It doesn’t invent from you, it averages what already exists. So when you ask it to generate the substance, it gives you back the statistical mean of how that’s been said a thousand times before. Correct, competent, and with nobody inside.
Joe Burns named the landscape that comes out of that: the Sloppening. Everything starts to sound the same, same cadence, same phrases, same transitions. Not because people use AI, but because they ask AI for exactly what AI can’t give: a voice. The voice is what’s left over from the average, and the average is exactly what the machine hands you.
Where AI adds and where it empties you
The form belongs to the machine; the substance is yours. That’s the whole rule, and it reads better as a table:
| AI, for the form | You, for the substance |
|---|---|
| Drafts, variations and first versions | The thesis: what you say and why |
| Transcribing, summarizing, reformatting, cleaning | The judgment to pick what works and what doesn’t |
| Exploring paths you wouldn’t have thought of | The last mile: what you publish and what you keep |
| Getting off the blank page | The voice: where you speak from and what you’re risking |
The left column is where AI hands you hours back and takes nothing. The right column is where, if you delegate it, you slowly lose the one thing that was hard to copy. Most people who “lose their voice” aren’t using the left column, they’ve been quietly moving things from the right column into the left without noticing.
Order matters: substance first, machine second
The order in which you bring AI in decides whether you keep your voice or lose it. And almost everyone does it backwards.
The usual move is to open the chat with a blank page and ask the machine to start: “write me a post about X.” In that order, the substance is set by AI (that is, by the average) and you’re left tweaking the form. You come out at the mean no matter what, because the substance was born generic.
Turn it around. Before you touch the machine, you set the substance: what your thesis is, where you’re saying it from, what you’ve seen that nobody is saying. Even if it’s four ugly lines written by hand. And only then do you use AI for what it’s great at: giving that a form, ordering it, offering you versions. In that order, the substance is yours and the machine only dresses it. It shows, and it shows a lot.
How to use AI without it showing
There’s no trick, there’s a way of working. Four concrete things.
Feed it your own material, and build a voice reference. Don’t start from a cold prompt. The highest-leverage move is to build a reference and reuse it every time, and there’s one technical detail that changes a lot: build it as a .md file, plain text, no PDF or oddly formatted doc. Any model reads it clean and you can carry it from one tool to another without fighting the format. Inside you put three things: eight or ten of your own pieces in full as examples, your hard rules (how you open, how you close, the words you never use, your punctuation tics) and one line on what you are and what you’re not. And the key: the AI reads it before anything else. As a system instruction, as project knowledge, or pasted at the top of the conversation, so it works from your voice from the first token instead of adjusting to it at the end. It’s basically building your own voice model without training anything or touching code. AI imitates what you give it; with that corpus in front of it, it comes out like you and not like the internet.
Use it for drafts, not decisions. Let it give you ten versions, get you unstuck, do the heavy lifting. But the decision of which one works, what stays and what goes, you make. That’s where judgment comes in, which is exactly what the machine doesn’t have.
Edit by cutting more than you add. AI text comes inflated, with extra transitions and phrases that sound like everyone. Your voice shows up when you cut: when you remove the automatic connector, the marketing word, the too-smooth cadence. The edit isn’t a touch-up, it’s where you sound like you again.
Keep the last mile human. The line you’re risking, the turn only you would take, the final call to publish or keep. That part, even when it’s hard, do it by hand. It’s small, and it’s the part people remember.
Where to start this week
Take a piece you were going to let AI write end to end. And before you open the machine, write the substance yourself in five lines: your thesis, your angle, what you’ve seen that nobody says. By hand, unpolished.
Then yes, use AI to give that a form, and afterward edit out everything that sounds like anyone. Compare the result with what you’d have gotten by asking for the post from scratch. That difference, the one you can feel between the two versions, is your voice. And now you know where it lives and how not to lose it.
Frequently asked questions
Does using AI make my content sound generic?
Only if you let it decide the substance. AI is trained on the average of what already exists, so when you ask it to generate what to say, it drifts to the mean, and the mean sounds like everyone. It doesn’t sound generic because you used it, it sounds generic because you handed it the part that was yours: the angle, the thesis, the voice. If you bring that and AI only helps with the form, it doesn’t show.
How do I get AI to sound like me and not like AI?
Feed it your own material before you ask it for anything. Give it your past writing, your notes, your way of putting things, instead of a cold prompt. AI hands you back a version of what you give it; if you give it the average of the internet, it gives you the average of the internet. And then edit, cutting more than you add: your voice is usually in what you delete.
Which parts of the process should I NOT use AI for?
The decisions, not the production. Keep out of the machine what you say and where from, the judgment to pick what works and what doesn’t, and the last mile: what you publish and what you keep. That’s the substance, and it’s exactly what sets you apart. AI is great for the form and for getting unstuck; for deciding, it’s still you.
Will AI replace creatives?
It will replace anyone who only brought form, because form is what the machine does for free. Anyone who brings substance (judgment, voice, a way of seeing) AI makes faster, not disposable. The question isn’t whether it replaces you, it’s which part of your work was only production and which part nobody else could do.
Before you go
The whole idea fits in one line: AI hands you the form by the bucketload, so the only thing that sets you apart now is the substance, and the substance doesn’t get delegated.
Use it without fear for everything that’s production, and guard jealously the part that was hard before AI and is still hard now: having something to say and a way of your own to say it. This week, write the substance yourself before you turn the machine on. And tell me whether the difference doesn’t show.
References
- Ted Chiang, ChatGPT Is a Blurry JPEG of the Web (The New Yorker, 2023) — AI as a blurry copy of the web’s average, and why it drifts toward the generic.
- Joe Burns, the term The Sloppening — content optimized until it loses its soul, what happens when you ask the machine for the voice too.
Reading recommendations
If you want to pull the thread of working with AI without losing yourself, three books:
- Ethan Mollick, Co-Intelligence. The most level-headed guide to working WITH AI: where to delegate and where never to cede your judgment.
- Rick Rubin, The Creative Act: A Way of Being. On taste and voice, the source of what only you can say.
- Austin Kleon, Steal Like an Artist. Where a voice of your own comes from, and how it’s built out of influences without sounding like a copy.
* The book links use my Amazon affiliate: if you buy through them, I earn a small commission, at no extra cost to you.