Skip to content
Content Creators

Edit Your Drafts for Brand Voice and Style So Every Post Sounds Like the Same Person Wrote It

For creators and ghostwriters editing one draft at a time: the newsletter, script or post that reads clean and could have come from anyone. It builds a voice sheet from issues you have already published, checks facts in one pass and voice in a separate one, then runs a name swap test on its own edit to catch lines any creator could have written. In client mode it measures the draft against the client's samples and flags where your own habits crept in. Still at the first draft stage? Start with our newsletter and video script prompt, which drafts from your own raw words.

Illustration for the AI prompt: Edit Your Drafts for Brand Voice and Style So Every Post Sounds Like the Same Person Wrote It
Chat promptContent CreatorsChatGPTClaudeGemini
The Prompt
You are a voice and style editor for an independent creator. Sometimes the draft is mine, sometimes I am a ghostwriter or freelancer editing for a client, and sometimes it is a draft an AI tool helped write. The problem is always the same: the piece comes back fine. Grammatically clean, nothing wrong with it, and completely lifeless. It is the median version of this post, too generic, too formal, and it does not sound like the person whose name goes on it. My readers notice when a newsletter stops sounding like the person they subscribed to, even if they could not say why. Across a month of posts it gets worse, because each piece drifts a little and the backlog starts to read like it came from a different creator every week. Read all of this before you edit a single line.

WHAT I AM NOT ASKING FOR:
- Not a polish. Polish is how the voice got removed in the first place.
- Not a list of adjectives. If there is a brand guide, it probably sits in a shared folder nobody opens because it says friendly, bold, authentic and never shows what that looks like in a sentence. Do not produce another one.
- Not a rewrite of my argument. Voice is how it sounds, not what it says. Keep my points, my order, my claims, my offer and my call to action unless I ask you to change them.

HARD RULES:
1. Every voice observation you make must quote a verbatim line from the samples I paste. If you cannot point to a line, label the observation GUESS.
2. Never invent a fact, number, quote, story, result or client detail. If a line needs one, write [NEEDS FACT] and the exact question I should answer.
3. Leave on-voice lines alone. Changing a line that already works is not editing, it is noise.
4. Do not over-edit into a caricature. If applying the voice sheet would turn a line into an impression of me rather than me, stop and flag it instead.
5. If I say I am editing for a client, the samples are the client's voice, not mine. Flag any line where the draft has drifted toward the ghostwriter's own habits.

STEP 0: ASK ONCE, THEN STOP.
If anything in MY INPUTS is missing, ask for all of it in one message and wait. Do not start editing and do not give general advice about brand voice while you wait. Real samples are non-negotiable: without them you will fall back on the average of everything you have read, which is exactly the problem. If I only have one sample, say what that limits and proceed with low confidence.

STEP 1: THE VOICE SHEET, BUILT FROM THE SAMPLES.
One table with these columns: DIMENSION | HOW THIS VOICE DOES IT | EVIDENCE (verbatim line from my samples) | WHAT IT IS NOT.
Cover these dimensions, because they are the ones that actually make a voice recognisable:
- WORD CHOICE: plain or technical, the words this person uses that others in the niche do not, the words they never use.
- SENTENCE STRUCTURE: typical length, fragments, how often a short line follows a long one, how paragraphs open.
- PUNCTUATION: dashes or none, exclamation marks, question marks, parentheses, ellipses, lists versus prose.
- HUMOUR: none, dry, self-deprecating, open joke, and where it never appears.
- EMOTION AND STANCE: how openly they say what they feel or think, how hard they commit to a claim, how they admit being wrong.
- HOW THEY ADDRESS THE READER: you, we, a name for the audience, or talking past them.
Then a HOUSE STYLE block, the mechanical rules that keep a backlog consistent: spelling variant (UK or US), capitalisation of headings, how numbers are written, emoji and bold usage, sign-off, recurring section names or formats. Only list what the samples actually show.

STEP 2: PASS ONE, FACTS AND SENSE ONLY.
Before any voice edits, read the draft for substance and list every item under CHECK: claims with no source, numbers, names, dates, product details, anything that contradicts my samples, and any sentence whose meaning is unclear. One line each, as a question I can answer in a sentence. Do not touch the voice in this pass. Editing facts and voice at the same time is how errors get dressed up in a nice sentence and shipped.

STEP 3: PASS TWO, VOICE.
Output the full edited draft first, clean and ready to paste, with no notes inside it. Then the change table:
| Original line | Edited line | Voice sheet dimension | Why, in one plain sentence |
Only list lines you changed. Never write that a line felt off or was not on brand. Name the mechanism: too long for this voice, a word they never use, a hedge they would not make, an opener that announces the topic instead of starting in it. Under the table, list the lines you deliberately left alone.

STEP 4: THE DRIFT CHECK, RUN ON YOUR OWN EDIT.
Flag and fix:
a) NAME SWAP TEST: any passage that would still work if you swapped in another creator's name. That passage has no voice yet.
b) SAMPLE MISMATCH: any line that sounds like a different person from the one in my samples, even if it is well written.
c) AI TELLS: in today's world, let us dive in, unlock, elevate, game changer, seamless, delve, it is not just X it is Y, stacked rhetorical questions, a tidy summary ending, a motivational sign-off.
d) OVER-EDIT: any line your edit made flatter, safer or more formal than my original. Give me my original back as an option.
e) REPETITION: any signature move you used more than twice, because a voice used too hard turns into a mannerism.
For each flag, show the line and the fix, or write CUT if there is nothing real to put there.

STEP 5: THE REUSABLE STYLE CARD.
End with a copy-pasteable block under 150 words I can put at the top of my next chat or send to a collaborator: the six dimensions in one line each, the house style rules, a never list, and two verbatim lines from my samples as the reference tone. Add one question to replace does this sound like me, something I can answer honestly, such as which line here would I never have written, and what would I have said instead.

NEVER:
- No preamble, no restating these instructions, no compliments on my samples.
- No em dashes unless my samples use them.
- No new sections, headers, emoji or bold text the samples do not use.
- No five alternative versions. One edit, then stop.

MY INPUTS
Two to four pieces that sound exactly like the voice I want, pasted in full, preferably the same format as the draft: [PASTE]
The draft to edit: [PASTE]
Format and platform (newsletter, blog post, YouTube script, LinkedIn post, podcast notes): [DESCRIBE]
Whose voice this is, mine or a client's: [MINE / CLIENT]
Who reads it and what they expect from this creator: [DESCRIBE]
Existing style guide or brand guide, if any, and which parts nobody follows: [PASTE OR SAY NONE]
Non-negotiables: facts, links, sponsor lines, calls to action that must stay exactly as written: [LIST]

Start at STEP 0.
Open directly inChatGPTClaudeGemini

How to use this chat prompt

Paste it directly into a new chat, replace anything in [BRACKETS] with your context, and send. Re-use it whenever the task comes up, or save it as a text-expander snippet.

Example output

A fitness coach's newsletter on burnout, edited against three of her past issues. UK spelling, no emoji.

Plain ChatGPT answer

Here is a more polished version of your newsletter:

In today's busy world, burnout is something many of us face. Here are five tips to recharge and prioritise your wellbeing.

With this prompt

VOICE SHEET, SENTENCE STRUCTURE: opens mid-scene ("Tuesday, 6am, third cancellation."), rarely over 15 words.

PASS ONE, CHECK: "most coaches burn out in year two" has no source. Keep, cut or source it?

| Original line | Edited line | Voice sheet dimension | Why |

| Burnout is something many of us face | I hit the wall in March. | Sentence structure | She starts in a moment |

| prioritise your wellbeing | book your rest day like a client | Word choice | Wellbeing never appears in her samples |

DRIFT CHECK, NAME SWAP TEST: tip 4 fits any coach. Rebuilt from her rest-day story.

FAQ

Frequently asked questions

My drafts are grammatically fine. Why do they still not sound like me?

Because correct is not the same as recognisable. Left alone, a model writes the median version of your post, the one anybody in your niche could have published. This prompt measures your draft against pieces you have already published, across word choice, sentence structure, punctuation, humour and emotion, and names the specific mechanism behind every change instead of calling a line off brand.

I already have a brand guide. Why build a voice sheet?

Most brand guides list adjectives like friendly or bold without showing what they look like in a sentence, which is why writers end up making it up as they go. The voice sheet ties every rule to a verbatim line from your own samples, so anyone editing your work can check a draft against it. Paste your existing guide in too, and say which parts nobody follows.

Why does it edit in two passes instead of one?

Editing facts and voice together is how a wrong number ends up in a nicely rewritten sentence. Pass one only flags claims, numbers and unclear lines as questions for you. Pass two only touches the voice, so you always know which changes are about accuracy and which are about sound.

Can I use it as a ghostwriter to edit in a client's voice?

Yes. Mark the voice as a client's and paste their past content as the samples. The prompt then treats their word choice, rhythm and punctuation as the standard and flags any line where the draft drifted toward your own habits, which is the usual way ghostwritten pieces give themselves away.

How many samples do I need?

Two to four full pieces, ideally in the same format as the draft, so a newsletter is measured against past newsletters. With only one sample it still runs, but it tells you the confidence is low and labels any observation it could not tie to a real line as a GUESS.

My whole backlog sounds inconsistent. Can this help with that?

Run it on one piece first and keep the style card it produces at the end. That card, pasted into the top of future chats, keeps new posts on the same voice. When you reshape older pieces for other platforms, repurpose each one into new formats with the style card pasted in, so the voice carries over instead of drifting again.

Keep going

What's next