How to Use ChatGPT to Write Emails That Sound Like You

Last Tuesday I spent twenty minutes rewriting a ChatGPT email draft because it kept calling my client "esteemed" and signing off with "Warm regards" — something I've literally never typed in my life. The email was technically fine. It just wasn't me. And that's the problem nobody talks about when they say "just use AI for emails." The default output sounds like a robot wearing a suit to a backyard barbecue.

I've been using ChatGPT for email drafts for almost two years now, and I've finally cracked how to make it actually sound like me — not a slightly more polished, corporate version of me. Here's what actually works.

Feed It Your Voice First, Not Your Request

Most people open ChatGPT, describe the email they need, and then wonder why it sounds generic. That's backwards. Before you ask for anything, you need to show ChatGPT how you actually write.

Here's what I do: I keep a note with five or six emails I've written that felt authentically "me." A casual check-in with a client. A slightly annoyed but professional follow-up. A genuine thank you. A decline that didn't burn bridges. I paste two or three of these into ChatGPT and say something like: "Here are some emails I've written. Study my tone, sentence length, how I open and close, and any phrases that seem distinctive to me. Don't summarize — just absorb this for the next task."

Then — and this is important — I wait for it to acknowledge. I ask it to tell me three specific things it noticed about my style. This isn't just a checkup. It actually helps the model lock onto patterns it might otherwise ignore. When I tried this, ChatGPT told me I tend to start sentences with "So" and use em dashes constantly. It noticed I rarely use exclamation points and that my closes are abrupt. All true.

Only after that confirmation do I describe the email I need. The difference is night and day.

The Real Trick: Teach It Your Anti-Style

Here's the thing that took me months to figure out, and I've never seen anyone else mention this. Teaching ChatGPT what you don't sound like is more powerful than teaching it what you do sound like.

After feeding it my sample emails, I add a constraint list. Mine looks something like this:

"When writing emails for me, NEVER use: 'I hope this email finds you well,' 'Just circling back,' 'Per my last email,' 'Please don't hesitate to reach out,' 'Warm regards,' 'Best regards,' or any variation of 'looking forward to hearing from you.' Also avoid starting more than one sentence with 'I' in a row."

This negative constraint list does something weird but effective. It forces the model to find alternative phrasings that — by process of elimination — end up closer to your actual voice. It's like telling someone "draw me a house but you can't use any straight lines." The constraint creates creativity.

I've tested this extensively. Same prompt, same sample emails, with and without the anti-style list. The version with the "don't" list sounds more like me about 80% of the time. It's not magic, but it's the closest thing I've found.

Context Matters More Than You Think

A quick email to a coworker you've known for three years shouldn't sound like a message to a potential client you've never met. Obvious, right? But ChatGPT will default to the same register unless you explicitly tell it otherwise.

I started adding what I call "relationship context" to every email request. Not just "write a follow-up email" but "write a follow-up email to my contractor Mike who I've worked with for two years and we usually joke around a bit." That tiny addition changes everything. ChatGPT might open with "Hey Mike, checking in on this—" instead of "Dear Michael, I wanted to follow up regarding our previous conversation."

I also specify my actual emotional state when it matters. "I'm mildly annoyed but don't want to come across that way" is useful information. So is "I'm genuinely excited about this project and want that to come through without seeming unprofessional." The model can calibrate tone surprisingly well when you're honest about what you're feeling.

Real talk: sometimes I'm too close to a situation to write the email myself. I'm either too irritated or too worried about something, and my natural voice would be too much. In those cases, I ask ChatGPT to write as me but "dial down the frustration by 50%" or "make this more neutral than I would naturally be." That's a legitimate use case. The AI version might actually be better than what I'd write raw.

Save Your Prompts, Not Your Outputs

After all this testing, I've settled on a system. I have three saved prompts — one for casual professional emails, one for client communications, and one for cold outreach. Each one includes my sample emails, my anti-style list, and the general tone I want. When I need an email, I paste the relevant prompt, add my specific request, and I'm usually happy with the first or second draft.

The outputs themselves are disposable. I never save them. But those prompts? I've refined them over months and they're genuinely valuable. If ChatGPT disappeared tomorrow, I'd lose the tool but keep the self-knowledge. I now understand my own writing style better than I did before I started using AI to replicate it. There's something kind of funny about that.

One limitation I'll be honest about: really short emails are still hard. For anything under three sentences, I usually just write it myself. The prompt overhead isn't worth it, and quick messages are where your authentic voice matters most anyway.

My take: the goal isn't to automate your personality. It's to automate the boring structural work so your personality can come through without you having to think about format and phrasing at the same time. Once I stopped expecting ChatGPT to nail my voice automatically and started treating it as something I had to actively teach, the whole process got way more useful. Takes some upfront work, but now my email drafts actually sound like someone I'd want to get an email from. Which is me. Hopefully.

Heads up: Some links in this post may be affiliate links. I only recommend tools I've personally tested. Opinions are entirely my own.