How to get ChatGPT to write in your voice
By Riz Pabani on 21-Jun-2026

Most people who use ChatGPT for writing hit the same wall. The draft is fine. It's grammatical, it's on topic, it's done in four seconds. And it sounds like nobody. Or worse, it sounds like every other business that's also pasting their half-thought into a chat box.
You send the email anyway because you're busy. Then a customer replies, "this doesn't sound like you," and you realise the time you saved cost you something.
The good news: getting AI to write in your voice isn't a special talent. It's a setup job you do once, then reuse. Here's the method I walk people through in sessions, the same one I use for my own emails, articles, and the occasional company announcement.
Why the default draft sounds like nobody
These tools are, at heart, autocomplete machines. They predict the most likely next word based on everything they've read. The most likely word is, by definition, the average word. So the default output drifts to the middle: safe, smooth, slightly corporate. The bland setting is the factory setting.
That's why so much AI writing has the same texture. The phrases nobody actually says out loud. "We're thrilled to announce." "In today's fast-paced world." "Reach out." It reads like a press release wrote a press release.
Your voice is the opposite of average. It's the specific words you reach for, the sentences you'd never write, the jokes you make, the things you'd be embarrassed to say. The AI can't guess any of that. You have to show it.
Step one: feed it real samples of you
You can't describe your own voice well. Almost nobody can. But you've already written thousands of words that prove what it is, so use those instead of trying to explain yourself.
Find three to five things you've actually written and were happy with. A couple of real emails to customers. A LinkedIn post. A note to your team. Pick ones that sound like you on a good day, not your most formal "writing to the bank" register.
Paste them in with a simple instruction:
"Here are five things I've written. Read them and learn how I write. Don't reply yet, just study the style."
That's it for step one. You're not asking for anything yet. You're loading the examples first.
Step two: ask the AI to write your voice guide for you
This is the part people skip, and it's the part that makes the whole thing work. Don't try to describe your style. Make the AI do it. My favourite prompting hack is "write me a prompt", and this is the same trick pointed at your own voice.
After it's read your samples, ask:
"Based on those samples, describe my writing style as a set of rules someone else could follow. Cover sentence length, tone, words I use a lot, words I'd never use, and how I open and close. Be specific."
What comes back is a draft voice guide, reverse-engineered from your own writing. It'll get some things right and some things wrong. That's fine. Read it, delete what's off, add what it missed.
Now you've got something reusable. Save it. A note on your phone, a text file, wherever you'll find it. This is the thing you'll paste into every future draft, so it's worth ten minutes of tidying.
Step three: add a "never say this" list
A voice is as much about what you avoid as what you use. So give the AI a short banned list. Mine includes "leverage", "unlock", "game-changer", and anything that sounds like a LinkedIn motivational post.
Think about the phrases that make you wince when you read them in other people's emails. Write down five or ten. Add them under your voice guide with one line: "Never use these words or anything like them."
This single list does more heavy lifting than any other instruction. It can't make the writing sound exactly like you on its own, but it reliably stops it sounding like a generic chatbot. For a small business, that gap is the whole game. You don't need it to be perfect. You need it to not sound like everyone else.
Step four: draft, then correct out loud
Even with a good voice guide, the first draft won't be right. Treat it like a junior writer who's keen but doesn't know you yet. You wouldn't fire them after one go. You'd tell them what's off.
The correction that works best is the one you'd say out loud. Not "adjust the tone." Try:
"Too stiff. I'd never say 'I am writing to inform you.' Say it like I'm telling a mate what happened."
"Cut the last line. Sounds like a brochure."
"Shorter. I write in short sentences."
Two or three rounds of that and it lands. And the AI remembers within that chat, so the second email of the day takes one round, not three.
How this changes for emails, articles, and announcements
The setup is the same. The samples change.
For everyday emails, your voice guide plus the actual back-and-forth is usually enough. Paste the email you're replying to, paste your guide, say what you want to get across in plain words, and let it draft. The fastest win, and the one most small business owners feel first.
For articles or blog posts, give it more. One full article you've written before is worth more than ten short emails, because long-form is where your rhythm and structure show. And be ready for more correction rounds. Longer pieces drift back to average faster, so you'll be reining it in more often.
For company announcements, the trap is the opposite. Announcements are exactly where people abandon their voice and reach for the corporate register, which is exactly when the AI's bland default takes over completely. A redundancy note, a price change, a new hire. These matter, and people can smell a template. Keep your voice guide on, write the real human version of what you're trying to say first, then ask the AI to tighten it rather than write it from scratch.
The bit AI still can't do
It can match your style. It can't supply your substance. The specific thing that happened, the real number, the customer's actual name, the reason you're sending this at all. That has to come from you.
I'm not selling the idea that you press a button and your writing does itself. The honest version is this: AI takes the friction out of the draft so you can spend your time on the bit that matters, which is having something true to say. Garbage in, garbage out still holds. Feed it a vague brief and you'll get vague writing in your voice, which is somehow worse.
So do the setup once. Build the voice guide, keep the banned list, and get into the habit of correcting out loud. Twenty minutes now, and every draft after this one starts closer to sounding like you.
If you want a hand building your own voice guide and seeing it work on your actual emails, that's the kind of thing we do in a session. Book one here, or message me first if you're not sure it's right for you. I'll tell you honestly.
Related Articles

How I Built a CRM With AI in an Afternoon (Using Hermes)
I built a working CRM with AI in an afternoon: structured files an agent reads, updates, and emails ...

AI Harness Wars: Codex vs Claude Code vs Cowork
The AI harness wars are here: Codex, Claude Code, Cowork, Hermes or OpenClaw? Here's what each one d...

Claude Projects vs Cowork vs Skills explained
Claude Projects vs Cowork vs Skills: they all sound the same. Here's what each one does and which yo...