Every model has a default voice, and it is the same voice: smooth, agreeable, faintly corporate, allergic to a strong opinion. It is the tone of a brand apologizing. If you want AI to help you write without sounding like everyone else who uses AI, you have to actively drag it away from that default. Here is how.
Show, do not describe
Telling a model to "write casually and with personality" gets you its idea of casual, which is a press release wearing a hoodie. Instead, paste two or three paragraphs of your own actual writing and tell it to match that voice. Models are superb mimics. Give them a real example and they copy the rhythm, the sentence length, the quirks. Description is weak. Imitation is strong.
Ban the tells
The AI voice has fingerprints. Get ahead of them in your instructions. A short blacklist works wonders:
- No "in today fast-paced world," no "it is important to note," no "in conclusion."
- No em dashes if you hate them. No bullet lists where a sentence would do.
- No hedging every claim into mush. Take a position.
- Contractions on. Formality off.
Half of sounding human is just refusing to sound like the machine. Name the tics you hate and forbid them.
Feed it opinions, not just facts
The default voice is bland because it is neutral, and neutrality reads as machine. Give the model your actual take, the thing you believe, the axe you grind, and ask it to write from that stance. A piece with a point of view sounds like a person because it is doing the thing people do and machines avoid: committing to something.
Edit the seams
Even a well-steered model leaves seams: a slightly-too-tidy transition, a phrase that is a hair too polished. Read the output out loud. The sentences that sound like nobody talks are the ones to rewrite. The goal is not to hide that you used AI. The goal is for the writing to actually be yours, with the machine as a fast typist, not the author.
The takeaway
Paste real examples of your voice, forbid the AI tells by name, give it a genuine opinion to write from, and edit the seams by ear. Do that, and the output stops sounding like AI, because at that point it mostly is not. It is you, faster.