The 6-step system to keep your voice while editing with ChatGPT

Most writers use ChatGPT wrong. This is how I fixed it.

Using ChatGPT to rewrite your content can feel like hiring a copy editor with unlimited energy and zero ego.

It’s fast. It’s obedient. It never rolls its eyes when you ask for the tenth revision.

Especially for people like me, working in marketing, ChatGPT isn’t just a writing tool. It’s a time-saver across editing, translating, and proofreading.

Not to mention the superpower of content repurposing: turning one great piece into many, without starting from scratch.

And yet, something about it didn’t sit right.

I bet you did it too. You open a fresh ChatGPT thread, paste in your draft, and type the magic words: “Can you make this better?”

In seconds, you get something smoother. Clean. Grammatically flawless.
And yet, somehow, emptier.

You scroll through the result and realize: the words are yours. Everything reads well, but the tone is off.

Even when you ask it to make something sound “more casual” or “more engaging,” it might comply, but it still won’t sound like you.

You know what I’m talking about.

Many writers have noticed it too. They ask the model to improve their draft, and the tone of voice disappears. The phrasing softens. The choices that made the sentence yours dissolve into something that reads like a dull Wikipedia page.

AI gives you speed, clarity, structure. But if you’re not careful, it also erases the mark that made the piece yours.

And I used to spend hours rewriting, editing, trying to reinfuse the tone I had in the original message.

Only to find it smoothed out again at the next proofreading round.
Arrgh.
Sometimes I’d just close everything and start over.
Other times, I’d skip the last edit just to avoid undoing all the work again.

To fix it, I first had to understand how the system sees writing, and where it gets things wrong.
Eventually, I stopped fighting the model and started training it.

And when I did, everything changed: my voice stayed in the room, the edits got easier, and the process finally felt like mine again.

Let’s see how…

Why ChatGPT Flattens Your Voice

ChatGPT doesn’t see your voice. It analyzes it.

To the model, writing style is just a pattern: sentence length, tone, word choice, rhythm, structure.

It can mimic patterns and echo certain choices. But unless you explain what defines your voice (what’s intentional, what can shift, what must stay) it will flatten everything into something neutral.

This is where most rewrites fall apart: you give the model something raw, and it gives you back something clean, but in the process, it kills your style.

And now that everyone has access to the same tools for editing, research, and drafting, what actually makes a piece stand out, if not your ideas and your tone of voice?

That’s why you should use ChatGPT as your editor, not your ghostwriter.

Otherwise, you risk becoming just a prompter, not a writer.

The good news is: you don’t need to choose between clarity and identity.

You just need to learn how to guide the system before it takes over your voice.

How to protect your voice when using AI

If the model can’t preserve your style by default, then you have to make that part of the brief. You can’t just ask it to “make this better” and hope for the best.

Once that voice is gone, it’s hard to get it back.

So the solution is not to stop using ChatGPT. The solution is to prompt like a writer who knows what they’re protecting (examples below).

Before you ask it to rewrite anything, you need to show it who you are: how you think, how your sentences move, what makes your voice yours. You’re not just feeding the model words. You’re teaching it to recognise the shape of your voice, so it knows what not to erase.

The goal is not to get a perfect draft in one go.

The goal is to build a process: example, edit, reflection, correction.

So what does that look like in practice?

Here’s how I actually use it, from first prompt to final draft.

✅ Step 1: Open a new project dedicated to your writing

Start with a clean space. Create a dedicated ChatGPT project and write within its chat. This ensures the model isn’t carrying over tone cues or instructions from previous sessions.

Inside the chat, use the Canvas to collect your tone guides and preferences. Unlike uploaded documents or static instructions, the Canvas stays visible while you work and you can update it in real time without resetting the chat. That makes it easier to keep your edits consistent and your voice intact.

Avoid relying solely on the project instruction field. It has character limits, doesn’t support live changes, and requires you to restart the chat every time you adjust something.

Note: when you open a new chat inside the project, you will need to ask ChatGPT to retrieve the previous canvas, otherwise it will be blank

Pro tip: If memory is on, check what’s active. Any saved style guides or past examples will affect how ChatGPT responds, even in a new thread. They can override or blend with what you’re giving now, unless you reset or update them.

✅ Step 2: Paste 3–4 of your best articles

Once your space is clean and context is reset, the next step is giving the model something to learn from: your voice.

Paste a few published pieces that sound the most like you. These give ChatGPT enough context to pick up on your rhythm, sentence flow, tone, and quirks, the things that define your voice.

Once you've pasted your samples, guide the model with a clear prompt like this:

I’m building a style guide for my writing. I’ll share a few article samples one at a time.
Open a new canvas and for each sample, do the following:

1. Analyze the tone, sentence structure, and stylistic devices.
2. Summarize the key traits in bullet point format.
3. Label the analysis clearly under a heading like: “✅ Confirmed from sample 1” (and increment with each sample).
4. Do not overwrite previous entries — only add new sections.
5. Wait for my confirmation before continuing to the next sample.

The final result should be a layered style guide where each sample builds upon the last.

✅ Step 3: Add ideas and write an outline

Once your voice is mapped, it’s time to shape what you’ll say, before figuring out how you’ll say it.

But don’t start writing yet: start with structure first.

Use ChatGPT as a writing assistant to help shape your outline. Provide your ideas in bullet points or short paragraphs, then prompt the model like this:

Act as my writing assistant, we are writing a new article for substack with my tone of voice that is outlined in the canvas "[Name of the Canvas with style instructions]".
I will now add some ideas to insert in the article. Write an outline based on these.

✅ Step 4: Write paragraph versions

Once the outline is solid, you can start writing, one section at a time. This keeps the voice consistent and gives you more control over structure, tone, and rhythm.

For each paragraph, I like to experiment by asking for multiple versions: one in the style of Sample 1, another in Sample 2, and so on.

Then I manually edit, rewrite, mix lines, remove sentences, or add new ideas.

It’s an iterative process. After I land on a rough version I like, I submit it again for polishing and proofreading. Usually four or five rounds before I’m satisfied.

When you're ready to start writing, try prompting this way:

Now, let's write the article in accordance with [insert writing style] of the Canvas called [name of the Canvas] and follow the outline. 
Let's write it section by section, skipping the introduction. We'll write the introduction separately.

This approach avoids generic, templated articles and lets you review voice alignment as you go. If something feels off, you can pause, revise, and prompt again, without losing the flow of the overall piece.

✅ Step 5: Clean up and streamline a merged draft

When working with multiple drafts or alternate versions of a section, it’s easy to end up with overlaps, contradictions, or broken flow. This step focuses on merging without flattening.

Use a prompt like this when you’re ready to consolidate two versions into one:

The text I’m about to share is a merged version of different drafts.
Do not rewrite everything.
Your task is to:
– Remove repeated ideas or duplicate phrasing
– Smooth out any awkward transitions or structural overlaps
– Keep original sentences intact wherever possible

Do not simplify the language.
Do not adjust tone, rhythm, or structure unless it's absolutely necessary for clarity.
This is not about improving the writing — it’s about fusing several versions into one clean, coherent draft.

When you're done, return the streamlined version and briefly explain what you changed and why.

✅ Step 6: Proofreading and AI fingerprint removal

Now it’s time for the final pass, still one paragraph at a time.

This is the most important step: the one where you clear AI red flags and bring the text fully back into alignment with your voice.

This is where the final layer of polish happens.

Use a precise prompt like this to protect your voice:

You’re a human editor reviewing the previous draft. Your job is to remove AI fingerprints while preserving my natural writing style, rhythm, and tone.
Follow these exact instructions:

❌ Remove all dashes — both em dashes (—) and en dashes (–). Rewrite sentence flow using commas, colons, or full stops instead. Do not just replace dashes with commas. Restructure as needed to preserve rhythm and clarity.

✍️ Proofread carefully: Fix any grammatical errors, awkward phrasing, or typos without simplifying the language.

✅ Don’t make tone changes or rewrite sentences unless it’s necessary for flow or clarity.

🔍 Specifically remove or revise these signs of AI-generated writing:
– Generic sentence starters like “In today’s world,” “Let’s dive in,” or “In conclusion”
– Empty conclusions (e.g. “You’ll be well on your way to success”)
– Overly balanced or generic tone that avoids opinion or specificity
– Default emojis like 🔹 ✨ used without voice-specific purpose
- Artificial triads or stacked patterns

Use bold and italics to emphasize ideas when appropriate, but never all-caps.

✅ Bonus Step: Identify weak points and offer fixes

Use this prompt when you're almost done, and want a final check for friction points, unclear ideas, or spots that might miss the mark with your audience:

Identify 3 specific weak points, contradictions, or moments where the argument could lose clarity, tone, or reader trust.

For each one, do the following:
– Describe the issue clearly (e.g. confusing phrasing, repetition, logic gap, tonal mismatch, etc.)
– Explain why it might not land well or feel off
– Suggest a possible fix (rewrite, cut, reframe, or clarify)

Focus on precision and impact — this is not a general edit, it’s a final quality check.

Return only the 3 flagged issues with suggested fixes.

Or if I don’t want to change the content but just proofreading the flow:

This is an excerpt from an upcoming article. Can you identify any weak points in the flow, especially transitions, pacing, or logical structure? Keep your focus on clarity and narrative momentum.

Final thoughts

Writing with AI doesn’t mean giving up your voice. It means learning to guide it.

The real magic isn’t in asking ChatGPT to “make it better.” It’s in showing the model who you are, what matters in your tone, and where the edges of your thinking live.

When you use AI as an editor, not a ghostwriter, you stay in control.

And when you pair your ideas with prompts that reflect your process, you don’t just write faster.

You write better.

Let the machine do the mechanical work. But don’t hand over the pen.

---

🔗 If this resonated, you might also like:

Lately, I’ve been reading more from people who are thinking beyond prompts and features. Voices exploring what this shift really means, not just what it enables. Here are a few pieces that stayed with me:

🧠 Outsourcing Ourselves, Part 2: Thinking — by Human, Rewritten
A quietly brilliant reflection on what happens when we delegate not just tasks, but decisions and how AI subtly shifts our sense of agency in the name of efficiency.
It’s not about rejecting the tools. It’s about reclaiming the thinking they can’t do for us.

📈 The AI Productivity Shift: Unlocking Focus or Fueling Technostress? — by AI & Marketing Ethics
A sharp, honest look at what happens when we mistake more output for progress — and how to build systems that don’t just scale, but make sense.
Productivity isn’t the problem. Misalignment is.

---

Read also: