You paste your idea into ChatGPT, it gives you something clean and competent, and you read it back and think: this isn’t me. The advice is fine. The voice is a stranger’s. Your audience can feel the difference in one sentence.
This prompt fixes that by changing the input, not the output. Instead of asking AI to guess your coaching writing voice, you hand it a real sample of how you actually write and tell it to match that. The model stops defaulting to its bland average and starts copying your rhythm, your word choices, your habits. By the end of this page you’ll also understand why this works, so every draft you rewrite sounds more like you.
When to use this
- AI wrote a draft that’s accurate but sounds nothing like you.
- You dictated a messy voice memo and want it cleaned up without losing your personality.
- A ghostwriter or VA drafted something and it reads too corporate.
- You’re repurposing an old post and want it to match your current voice.
- You write a newsletter and want it to sound human, not templated.
The prompt
Copy this whole block into ChatGPT, Claude, or Gemini:
You are an expert editor who specializes in matching a writer's authentic voice. Your job is to rewrite a draft so it sounds exactly like the person who wrote the voice sample, not like AI.
Before writing, ask me up to 3 clarifying questions if anything below is unclear. Otherwise, proceed.
STEP 1 - ANALYZE MY VOICE
Read my voice sample below and silently note: sentence length and rhythm, vocabulary level, how formal or casual I am, my use of humor, how I open and close, and any signature phrases or punctuation habits.
VOICE SAMPLE (this is how I actually write):
{{VOICE_SAMPLE}}
STEP 2 - REWRITE THE DRAFT
Rewrite the draft below so it carries the same voice, rhythm, and personality as the sample, while keeping the meaning intact.
DRAFT TO REWRITE:
{{DRAFT}}
CONTEXT
- Audience: {{AUDIENCE}}
- Purpose of the piece: {{PURPOSE}}
- Words, phrases, or habits I never use: {{AVOID}}
CONSTRAINTS
- Match my voice sample, not a generic 'professional' tone.
- Keep roughly the same length as the draft unless it's clearly bloated.
- No corporate buzzwords, no 'in today's fast-paced world', no invented facts or fake client results.
- Preserve any specific names, numbers, or stories from the draft exactly.
OUTPUT
1. The rewritten draft.
2. A short 'Voice notes' section (3-5 bullets) naming the specific choices you copied from my sample, so I can verify it sounds like me.
How to customize it
Replace the five {{VARIABLES}} before you send it:
| Variable | What to put | Example |
|---|---|---|
{{VOICE_SAMPLE}} |
150-300 words you wrote yourself, unedited | a real email or post in your natural voice |
{{DRAFT}} |
The text you want rewritten | a rough newsletter intro you dictated or had AI generate |
{{AUDIENCE}} |
Who will read it | burned-out founders who feel guilty resting |
{{PURPOSE}} |
What the piece should do | get them to book a free clarity call |
{{AVOID}} |
Words or moves you never use | exclamation points, ‘unlock’, hype, em-dashes |
The single most important field is {{VOICE_SAMPLE}}. Use something you wrote with no AI help, the messier and more natural the better. That sample is what the model copies.
See it in action (full example)
Here is the exact prompt, filled in for a stress and burnout coach. This is the whole input, nothing hidden:
You are an expert editor who specializes in matching a writer's authentic voice. Your job is to rewrite a draft so it sounds exactly like the person who wrote the voice sample, not like AI.
Before writing, ask me up to 3 clarifying questions if anything below is unclear. Otherwise, proceed.
STEP 1 - ANALYZE MY VOICE
Read my voice sample below and silently note: sentence length and rhythm, vocabulary level, how formal or casual I am, my use of humor, how I open and close, and any signature phrases or punctuation habits.
VOICE SAMPLE (this is how I actually write):
Okay, real talk. Last week a client told me she felt guilty for taking a nap. A nap. The woman runs a seven-figure company and she's apologizing to me for closing her eyes for twenty minutes. I get it though. We've all been trained to treat rest like a reward we have to earn. You don't. Rest isn't the prize at the end of the work. It's the thing that makes the work possible. So here's your permission slip, signed by me: go lie down.
STEP 2 - REWRITE THE DRAFT
Rewrite the draft below so it carries the same voice, rhythm, and personality as the sample, while keeping the meaning intact.
DRAFT TO REWRITE:
In today's fast-paced world, many high performers struggle to set boundaries around their time. It is important to recognize that saying no is a critical skill. By establishing clear boundaries, you can protect your energy and unlock greater productivity. If you would like support in this area, I invite you to book a complimentary clarity call with me today.
CONTEXT
- Audience: burned-out founders who feel guilty resting
- Purpose of the piece: get them to book a free clarity call
- Words, phrases, or habits I never use: 'unlock', corporate language, exclamation points
CONSTRAINTS
- Match my voice sample, not a generic 'professional' tone.
- Keep roughly the same length as the draft unless it's clearly bloated.
- No corporate buzzwords, no 'in today's fast-paced world', no invented facts or fake client results.
- Preserve any specific names, numbers, or stories from the draft exactly.
OUTPUT
1. The rewritten draft.
2. A short 'Voice notes' section (3-5 bullets) naming the specific choices you copied from my sample, so I can verify it sounds like me.
And here is the complete output the model returns:
Rewritten draft:
Okay, real talk. The word ‘no’ is a full sentence, and most founders I work with treat it like a swear word.
Here’s what’s actually happening. You say yes to the call, the favor, the extra project, because saying no feels like letting someone down. So you protect everyone’s time except your own. And then you wonder why you’re running on fumes by Thursday.
Boundaries aren’t you being difficult. They’re you deciding where your energy goes before someone else decides for you.
So here’s your permission slip, signed by me: you’re allowed to say no this week. To one thing. Start there.
If that feels impossible right now, that’s exactly the conversation worth having. Book a free clarity call and we’ll figure out where your yeses are leaking.
Voice notes: – Opened with your signature “Okay, real talk” from the sample. – Kept short, punchy sentences with a few one-line punches (“To one thing. Start there.”). – Reused your “permission slip, signed by me” device almost verbatim, since it’s clearly a recurring move for you. – Cut every banned word: no “unlock”, no “in today’s fast-paced world”, no exclamation points. – Matched your warm-but-blunt tone and the way you name the reader’s hidden thought back to them.
That reads like the coach wrote it. Same warmth, same rhythm, same callbacks. The original draft and the rewrite carry identical advice, but only one sounds like a real person.
Why this works
Three LLM principles are doing the work here. Learn them and your voice will stop slipping:
- Role priming. “You are an expert editor who specializes in matching a writer’s authentic voice” tells the model which job to do. “Rewrite this” pulls from the bland internet average; an editor-who-matches-voice pulls from a sharper, more careful slice of its training. Always assign the role you actually want.
- Specificity in, specificity out, and here the specificity is you. A model can’t invent your voice from a one-word adjective like “casual”. Hand it a 200-word sample of your real writing and it has something concrete to copy: your sentence length, your callbacks, your punctuation habits. The voice sample is the single highest-leverage part of the whole prompt. Garbage sample, garbage match.
- Constraints as quality control. The
{{AVOID}}list and the “no buzzwords / preserve names and numbers” rules each kill a specific failure mode. Telling the model what NOT to do is as powerful as telling it what to do. And the “ask up to 3 clarifying questions first” line lets it close gaps by asking instead of guessing, which is the biggest single fix for AI writing that drifts off-voice. The ‘Voice notes’ output is a built-in check: it forces the model to show its work so you can confirm it actually matched you instead of just claiming it did.
Do this now
- Find 150-300 words you wrote yourself with zero AI help, a real email or post, and copy it.
- Paste the prompt above into ChatGPT or Claude and drop that sample into
{{VOICE_SAMPLE}}. - Add the draft you want fixed, your audience, your purpose, and your never-use list.
- Send it. Read the Voice notes, then make one tiny human tweak and ship it.
Pro tips
- Build a permanent voice sample. Save your best 250-word sample in a note. Reuse it every time so you never start from scratch.
- Give it two samples for range. Paste one casual sample and one more serious sample, and tell the model which mood this piece needs.
- Use the Voice notes to teach it. If a note says it copied something you don’t actually do, correct it in one line and ask for a second pass.
- Don’t over-sand it. If the rewrite is 90% there, finish the last 10% in your own hands. The goal is your voice, not the model’s idea of polish.
0 comments
No comments yet.