Claude writes emails in my voice

Every AI email draft you've ever binned had the same problem. It sounded like a polite stranger, so you rewrote it from scratch and wondered why you bothered.

That happens because Claude has never seen how you actually write. It guesses, and a guess reads like a template. You end up spending longer fixing the draft than you would have spent typing the reply.

I fixed this by teaching Claude my voice before asking it for a single draft. I connected my email to the Claude desktop app and asked it to study everything I'd sent in the last month. The greetings I use, the sign-offs, how blunt I am, how long my replies run. Then I had it save what it learned as a skill, so it applies my voice every time it drafts. From then on, the drafts sound like me.

Next I put it on a schedule. Every morning before I'm at the desk, Claude reads the unread inbox, drafts replies to whatever it can handle, and lists the questions only I can answer. I check the drafts over the first coffee and fix a line here and there before they go.

The next level is adding your call transcripts to a searchable database, so Claude replies knowing the history with each client rather than just the thread in front of it.

You stay the editor. Your judgement and your relationships go into every reply before it leaves. But once this is set up, Claude does roughly three quarters of the work of replying, and email stops being the job that eats the first hour of your day.

Teach Claude your voice first. Everything else builds on that.