Blog / Four prompts that teach AI your writing voice

tutorial

Four prompts that teach AI your writing voice

Written by Gina Wang · July 29, 2026

A step-by-step method to make AI writing sound like you instead of everyone else's AI post, using your own old emails and posts, no style guide required.

I asked Claude to write a LinkedIn post about a client win. It came back smooth, confident, and it sounded like every other AI-written LinkedIn post I’d scrolled past that week.

Not bad writing. Just not my writing.

Here’s what I figured out: the problem wasn’t the tool. It was that I’d asked for a tone (“professional but approachable”) instead of showing it an actual pattern. Adjectives get you the same safe voice every time. Real samples don’t.

Why “write a post about X” always sounds like everyone else’s AI post

When you ask for a vibe, you get the vibe every other business owner asked for too. Professional, friendly, engaging. That’s not a voice. That’s the default setting.

Your actual voice lives in specifics: how long your sentences run, what you open with, the phrase you’d never say out loud. None of that shows up when you describe tone. It only shows up when you show your own writing.

Step 1: pick 2-3 pieces you’d stamp as “100% me”

Go find writing you already did. An email you’re proud of. A social post that got a real reaction. A text to a client that sounded exactly like you on a good day.

Two or three pieces is enough. Don’t overthink which ones. If your gut says “yeah, that’s me,” that’s your sample.

Step 2: have AI reverse-engineer the pattern

Paste the samples in and ask for the pattern, not a compliment.

Here are 2-3 examples of my writing. Don't tell me you like them.
Break down the actual pattern: average sentence length, how I open
a piece, words or phrases I repeat, anything I'd never say. Be
specific and concrete, not vague adjectives like "warm" or
"conversational."

You’ll get something like: short opening sentence, then a longer one, then a fragment. Starts with a specific moment, never a thesis. Never uses the word “leverage.” That’s a pattern you can reuse. “Warm and conversational” isn’t.

Step 3: build a short voice profile doc

Take what it gave you and turn it into a short doc, maybe 10-15 lines. Save it once.

Include:

  1. Sentence rhythm (short, then longer, then a fragment when it lands)
  2. Your real openers (a specific moment, never a warm-up line)
  3. Words you use and words you’d never touch
  4. One or two full example passages, kept word for word

This becomes the thing you paste into every writing prompt from here on. You built it once. You use it forever.

Step 4: the 3-round refinement loop

The first draft with your voice profile attached will be closer, but it won’t be perfect. That’s fine. Refine it out loud, the way you’d redline a draft from a new hire.

Round 1: More casual. Cut anything that sounds like a press release.
Round 2: Cut the corporate words. I'd never say "streamline" or
"leverage."
Round 3: I'd never say that last line. Just end it after the second
sentence.

Three rounds, plain language, no need to explain why. Each correction you make is worth adding to your voice profile, so you’re not making the same fix next month.

A before-and-after, side by side

Before, with just a tone prompt (“write a LinkedIn post about landing a new client, professional but approachable”):

“I’m thrilled to announce an exciting new partnership! This collaboration represents a fantastic opportunity to deliver innovative solutions and drive meaningful results for our valued client.”

After, with the voice profile attached:

“Landed a new client this week. The kind of project where they already tried three other people first. I like those. Means the easy stuff’s already been fixed, and what’s left is the actual hard part.”

Same news. One sounds like a press release. The other sounds like a person who was actually there.

Saving this as a system, not a one-time trick

The voice profile you just built isn’t a one-off file. It’s the first real system you own: same instructions, reused every time you need to write something, no rebuilding required.

Drop it into a Claude Project alongside your business brief, and every piece of writing you generate from here on starts already sounding like you. Not one good post. Every post, without doing this work twice.

Key takeaways

  • Give it 2-3 samples you'd stamp as '100% me,' and have it reverse-engineer the pattern (sentence length, your openers, your quirks) instead of asking it to match a vague tone word like 'friendly.' Save that pattern as a voice profile and paste it in every time.
  • Because most people prompt with adjectives ('professional, friendly, engaging') and every AI tool defaults to the same safe, smooth, slightly corporate voice when given adjectives.
  • Yes.
  • Run it through a refinement loop after the first draft.

Liked this one?

I send these out before they land on the site. Nothing here needs a developer, and nothing here needs you to be technical.

No spam, and you can leave whenever you want.

Common questions

How do I get AI to write like me instead of sounding generic?
Give it 2-3 samples you'd stamp as '100% me,' and have it reverse-engineer the pattern (sentence length, your openers, your quirks) instead of asking it to match a vague tone word like 'friendly.' Save that pattern as a voice profile and paste it in every time.
Why does AI writing always sound the same?
Because most people prompt with adjectives ('professional, friendly, engaging') and every AI tool defaults to the same safe, smooth, slightly corporate voice when given adjectives. Real writing samples work where adjectives don't.
Can I train Claude on my old emails or posts?
Yes. Paste in 2-3 pieces of your own writing and ask it to describe the pattern back to you: sentence length, how you open, words you'd never use. That description becomes your reusable voice profile.
How do I stop AI-written content from sounding fake?
Run it through a refinement loop after the first draft. Tell it what's off in plain terms: 'too corporate,' 'I'd never say that,' 'cut the last line.' Two or three rounds usually gets it there, and you keep the fixes in your voice profile so you don't repeat the correction next time.