1 June 2026
Most people give up on AI sounding like them too early
Most people give up on AI sounding like them after one draft.
They prompt it. They read v1. It sounds generic. They conclude AI can't do their voice and either rewrite the whole thing themselves or stop using AI for writing.
That moment - the one where they reject v1 and walk away - is the moment the work actually starts.
Start small, not perfect
When you first start trying to brief AI on your voice, you don't really know what to say. So you start simple. Use Australian English. No exclamation marks. Don't say "unlock potential". Two or three sentences, paste-in-anywhere.
That's enough to get you off zero. Don't try to write a 20-page voice doc upfront - you'll never finish, and most of what's actually distinctive about you won't make it in anyway. The interesting bits of your voice aren't the rules you can articulate. They're the rewrites you make once you see something that isn't quite right.
Why a brief isn't enough
The thing nobody warns you about: the first draft is supposed to be generic.
How could it not be? It's read a million pieces of writing and the few sentences you gave it about you. The first cut lands somewhere in the middle of "competent writing" because that's all it has to work with. Of course it doesn't sound like you yet. You haven't shown it you yet.
People reject v1 like AI failed them. AI didn't fail. It made the only draft it could make from what you gave it.
The actual work is what happens after.
What I do
Recent example. Welcome email for new user signups. I asked AI to draft it. v1 came back - fine, polite, generic. Sounded like a thousand other welcome emails.
So I rewrote it. Not from scratch. Just the bits that make it feel like me - warmth, an exclamation where I'd actually use one, an emoji where it'd land, more context around things the AI couldn't have known mattered (it's the first cut, how would it?). I deleted a few words I'd never say. Smaller changes than you'd think.
Then I fed the edited version back: hey, I sent this version, please review and learn from it for next time.
I've done that exact loop three times now. Each round, the version it gives me needs less editing than the round before.
The advanced move
Sometimes I'll drop a long brain dump in chat - on a topic, on a problem, just thinking out loud. Then at the end I say: do you see how I wrote to you? That's how I speak. Analyse it. Take learnings from that, beyond what I'm literally telling you.
The dump itself is voice training data. The way I cut myself off, the way I correct mid-sentence, the words I reach for, the ones I never use - all of it is voice information that no brief can capture.
Why most people don't do this
Most people skip the loop because they're expecting AI to be a faster typist. Hand it a brief, get the answer, done. So when v1 doesn't sound like them, they reject the whole tool or they accept it as good enough and think AI “isn’t that great”.
The thing they miss is that the loop IS the work. Each edit you make, with the reason attached, is what AI needs to see to learn your voice. The brief just gets it started.
Next time AI drafts something and v1 lands flat, don't reject it. Sit with it. Rewrite the parts that aren't you. Tell it specifically what you changed and why. Then ask it to remember.
Do it three times. See what v4 looks like.