The complaint always sounds the same: "it writes well, but it's not me." That's normal. Without a reference, a model produces an average, and an average sounds like nobody. The full guide on the topic is writing with AI without sounding like AI ; this article is the how-to.
Step 1: pick five posts, not fifty
Take five posts you can reread without wincing. Not your best performers: the most faithful ones. A post that got lots of views in a tone that isn't yours is a counterexample.
Step 2: spot what repeats
Look at what comes back across your five posts. Your average sentence length. The fact that you often open with a negative. Your punctuation. The fact that you never use exclamation marks.
These details make your voice far more than your vocabulary does.
Step 3: name what you never do
This is the step everyone skips, and it's the one that works best. Write down what you refuse: slogan sentences, rhetorical questions as openers, emojis at the start of lines, the word "game changer".
A list of bans constrains a model more than a positive description of your style. To find out which of these tics already creep into your writing, run it through the generic content detector : it catches seventeen of them, em dash included.
Saying "write like me" doesn't work. Saying "never do these seven things" works right away.
Step 4: edit, don't regenerate
When a sentence sounds off, the instinct is to rerun the whole thing. That's a mistake: you lose what was working too. Fix that sentence, keep the rest, and note why it sounded off.
After two weeks, that list of fixes becomes your real voice reference. That's exactly the idea behind per-dimension edits in Kast, and why we don't offer a "rewrite everything" button.
Step 5: read it out loud
The final test takes thirty seconds. If you stumble on a sentence reading it aloud, your reader will too. No score and no tool replaces that test.
If you're stuck on the first line, the hook generator suggests eight on your topic, and this article explains why each of these structures works.