How to Train AI to Stop Sounding Like Everyone Else (and Start Sounding Like You)

Every AI tool has a default voice. You know it when you read it. Clean, competent, a little hollow. It's the writing equivalent of hotel art. Nothing wrong with it, nothing memorable about it either.

Here's what I've noticed. People blame the tool for that. They shouldn't. The tool didn't decide to sound generic. It sounded generic because nobody told it what specific, sounds like.

Generic is the default, not a limitation

Think about it from the model's side for a second. It has no idea you say "y'all" and never say "folks." It doesn't know you'd rather cut a sentence in half than use a semicolon. It doesn't know the phrase you always reach for when you're closing out a post, or the three words you'd delete on sight if you saw them in your own writing.

Without that information, it makes reasonable, safe, forgettable guesses. That's not a flaw. That's what happens when you ask for specificity and provide none.

Step 9 in the AI DNA Workflow is Personalize the Voice, and it exists because this is fixable in about twenty minutes, once, instead of re-explaining your tone in every single conversation for the rest of your life.

Build the file

Here's the actual thing to build: a markdown voice file. Not a prompt. A document. Something you write once and hand to Claude at the start of a project, or drop into a custom instruction, or paste in whenever you're starting something that matters.

What goes in it:

Phrases you actually use. Not aspirational ones, ones you catch yourself saying in real emails, real texts, real posts. If you always start a paragraph with "here's the thing," write that down.

Words you'd never say. This one does more work than people expect. "Synergy." "Leverage" as a verb. "Circle back." "Unlock your potential." Whatever makes you wince when you see it in someone else's copy, list it as banned. Negative examples train a model just as well as positive ones, sometimes better.

A real writing sample. Something you wrote yourself, no AI involved, that actually sounds like you. An email, a caption, a rant in your notes app you never sent. Paste the whole thing in. Let the model study the rhythm, not just the vocabulary.

How you structure things. Do you write short punchy lines? Do you use fragments on purpose? Do you hate exclamation points? Say so.

That's it. That's the file. It doesn't need to be long. It needs to be true.

Feed it once, not every time

Here's the part that saves you actual time. Once that file exists, you're not rewriting your tone instructions in every new conversation. You paste the file in, or you drop it into custom instructions if your setup allows it, and you're working with something that already knows how you sound before you've typed a single sentence.

Compare that to what most people do instead: typing some version of "write this in a casual, friendly tone" every single time, and getting a slightly different flavor of generic every single time, because "casual, friendly tone" describes about half the internet.

This isn't about finding the right prompt

I want to be direct about something. This is not a magic prompt. There isn't a clever one-liner that unlocks your voice in a single shot. That idea sells because it sounds easy, but it skips the actual work, which is sitting down and getting specific about how you sound, once, on paper.

A prompt is a request. A voice file is a reference. One evaporates the second the conversation ends. The other gets more useful every time you reuse it.

Start small tonight

You don't need a polished document. Open a blank file. Write down five phrases you actually use. Write down three words you never want to see in your writing again. Paste in one real thing you wrote that sounds like you.

Save it. Use it next time.

That's the whole system. It's not fancier than that, and it doesn't need to be. The goal was never a perfect prompt. It was a model that finally sounds like it's known you for a while.