You've noticed it. You ask the AI to write an email and what comes back is⦠fine. Clear. Correct. And nothing like how you actually talk.
That's because AI was trained on the whole internet β so by default it sounds like the average of the whole internet. Smooth, slightly hollow, a little formal. The "delve into" voice.
There's a fix. And Step 1 takes five minutes.
**Step 1 β Gather 3β5 real samples of your own writing**
Open a scratch note (Apple Notes, Google Keep, a blank doc β anywhere). Paste in 3β5 short things you actually wrote:
β’ One work email you sent this week (any topic β doesn't have to be important).
β’ One Slack, Teams, or WhatsApp message that's more than two lines.
β’ One paragraph from a document or report you wrote.
Rules: real, not aspirational. How you actually write, not how you wish you wrote. Strip names and anything confidential β replace with `[Name]` or `[Project]`.
That's your raw material. With those samples, you can run a **Voice Analyzer prompt** that tells the AI what makes *your* writing recognisable β your sentence rhythm, the words you use, the phrases you'd never say.
**Why this matters**
The AI doesn't sound like you because it doesn't know you yet. Once it does, the difference is immediate β and you can reuse that knowledge on every piece of writing you touch from here on.
Step 1 is free. The Voice Analyzer prompt, the 5-section Voice Card you build from the findings, and the save-it-as-Custom-GPT / Claude Project path (so you never have to paste it again) are inside Pro today.