Almost everyone can smell AI writing now. The empty enthusiasm, the “in today’s fast-paced world,” the three tidy bullet points that say nothing. It sounds like a press release from a company that sells nothing.
Here is why. You asked it to write “in a friendly, professional voice,” and it did exactly that: it wrote the average of every friendly professional thing ever posted online. Adjectives do not carry your voice. Your actual sentences do.
The fix is to stop describing your voice and start showing it. Below is the exact method I use so AI writes like me, whether it is a work email or a note to a friend.
The Voice Sample Method
The whole trick has a name I use for it: the Voice Sample Method. The key move is one line: paste two or three samples of your real writing instead of describing your voice in adjectives.
That is it. You do not tell the AI you are “warm but direct.” You show it three things you actually wrote and let it copy the pattern. How long your sentences run. Whether you use contractions. Where you break a rule on purpose. The AI is very good at matching a pattern it can see and very bad at guessing one you only named.
Three parts make it work:
- The samples. Two or three real pieces of your writing. Not your most polished corporate email. The ones that sound like you talking.
- The rules. The hard yes and no. My own list is strict. I never use em dashes, ever. I keep sentences short. I cut corporate filler. Yours will be different, and that is the point.
- The do-not list. The specific words and moves that make you cringe. Banned openers, buzzwords, fake enthusiasm. This is what stops the AI from over-correcting back into robot.
I train AI on my own voice this way constantly. I have a strict personal style with hard rules I do not break, and I feed those rules plus real samples into the tools before I ask for a single sentence. The same strict voice rules, never an em dash, my full ban list, run across all 35 AI agents I operate and everything I publish. The output starts in my voice instead of arriving as beige corporate soup I have to rewrite from scratch.
What it looks like, robot vs you
Here is the same message two ways. Left is what you get from “write a friendly follow-up email.” Right is what you get after the Voice Sample Method, with the instruction that produced it underneath.
| Robotic AI version | Your-voice version |
|---|---|
| ”I hope this email finds you well! I wanted to circle back and touch base regarding our previous conversation. I’m thrilled about the opportunity to leverage our synergies and move the needle together. Please don’t hesitate to reach out with any questions!" | "Hi Sam, quick follow-up on Tuesday’s call. I think the plan we landed on works. I can have a first draft to you by Friday. One open question: do you want me to loop in the design team now or after you review? Let me know.” |
The instruction that got the right column:
Here are three samples of how I actually write. Match this voice.
[paste sample 1: a real short email you sent]
[paste sample 2: a Slack message or text to a coworker]
[paste sample 3: a note to a friend]
Rules:
- Short sentences. Plain words.
- No em dashes, ever.
- Contractions are fine.
- No corporate filler.
Do NOT use: "I hope this finds you well," "circle back," "touch base,"
"leverage," "synergy," "move the needle," "thrilled," "don't hesitate."
Now write a follow-up email to Sam about Tuesday's call. Warm but direct.Notice the register line at the end. “Warm but direct” for a coworker. You would swap it for “casual and short” on a note to a friend. Same voice, different room. The samples set who you are; the instruction sets the moment.
The common mistakes
I hit every one of these before it clicked.
- Describing your voice in adjectives. “Friendly, confident, approachable” tells the AI nothing it can copy. It already thinks it is all three. Show samples, do not list traits.
- Skipping the do-not list. If you never tell it what you hate, it defaults to buzzwords and fake cheer. The banned list does half the work. Name the exact words that make you cringe.
- Accepting the first draft. The first pass is a starting point, not the answer. Tell it what is off: “too formal, cut the enthusiasm, shorter.” One round of that fixes most of it.
- Over-polishing until it sounds fake again. There is a point where “make it better” swings back into robot. A little rough is human. When it starts sounding smooth and empty, stop, or tell it to add back the plain edges.
Build more than one voice
Nothing says you only get one. You already show up differently for your boss, your team, your best friend, and your group chat. Build a small library instead of a single sample: a work voice, a casual voice, a just-for-friends voice, whatever rooms you actually walk into. Same method every time, just a different set of samples and a different do-not list for each one.
Before you use any of this on something that matters, play with it first. Feed it a few lines from your favorite TV character and have it rewrite your next email that way. Tell it to answer only in pig latin, just to see it work. Silly is fine, the point is not the output, it is getting comfortable pulling a voice out of a handful of samples so it feels automatic once you point it at your own.
Save it once, reuse it forever
You do not want to paste your samples every single time. Once your voice prompt works, save it. Put it in ChatGPT custom instructions or a Project, or a Claude Project, so every chat starts already knowing how you write. That is the same give-AI-a-memory habit pointed at your writing.
Two more guides pair with this one. If your instructions still come out muddy, start with how to write a good prompt. And if you are brand new to all of this, your first week with AI is the on-ramp.
Do this today
Find three things you actually wrote that sound like you. One work message, one casual one, one somewhere in between. Paste them into your AI with the rules and do-not list above, then ask it to draft your next email. Read it out loud. If it sounds like you, you are done. If it sounds like a robot, tell it exactly what is off and run it again.
Want my voice-training prompt and the full do-and-do-not list I feed AI so it stops sounding like a robot? Grab it below and I will send it straight over.