Everyone says "write in your voice." Nobody tells you how to define it. Here's a 20-minute exercise that captures it on paper.
Your voice fingerprint is five decisions and three samples. If you can't articulate it, you can't ask a tool to imitate it, and you can't notice when it drifts.
Decision 1: the temperature
Pick one: cold (analyst), room temp (friendly neighbor), warm (encouraging), hot (firebrand). Don't pick two. Most content dies at "room temp / warm" because the creator can't decide.
Decision 2: the cadence
Short sentences? Long sentences? Mixed? Your voice has a rhythm. If you don't define it, the AI picks one at random, and it drifts.
Decision 3-5: vocabulary, references, banned words
Vocabulary: 5 words you use often, 5 you refuse. References: where your metaphors come from — sports, construction, cooking, music, theater? Pick one lane. Banned words: 3-5 you'd rather die than type.
The three samples
Find three pieces of content you wrote that sound most like you. Paste them into a doc. Mark the sentences that feel most you. These become reference posts for the Brain.
You can't protect a voice you haven't described.
Why this beats prompt engineering
You could spend hours tweaking a prompt to get the right output. Or spend 20 minutes defining the voice once, upload it, and let every draft inherit it forever.