A ChatGPT Prompt That Makes Copy Sound Like You Wrote It

Most people don't want AI to write for them. They want AI to write with them, in a voice that sounds like theirs. That's a meaningful distinction, and getting there requires more than just saying "write in a casual tone."

The good news: a well-built prompt can get ChatGPT surprisingly close to your natural voice. Not identical, but close enough that readers won't flag it as a bot. Here's how to build one.

Why default ChatGPT copy sounds fake

Before we get to the prompt, it helps to understand what's going wrong with the generic output.

ChatGPT defaults to a style that is technically fluent but statistically averaged. It uses the phrasing patterns that appear most commonly across its training data, which means it gravitates toward:

• Sentences that run in sets of three ("fast, clear, and actionable")
• Transitions that fake depth ("it's worth noting that...", "it's important to understand...")
• Verbs that describe the act of doing things rather than just doing them ("serves as a reminder", "stands as an example")
• Em dashes and semicolons used too often, in ways real people rarely write

None of those are wrong in isolation. But together, they produce that uncanny valley effect where the prose is grammatically fine and completely hollow.

A human writer makes different choices. They repeat a word when it's the right word. They write a three-word sentence when that's all the situation needs. They have opinions about things. Your prompt needs to push ChatGPT toward those habits.

The core prompt structure

Here's a template that actually works. Fill in the bracketed sections with specifics from your own writing.

Rewrite this in my voice. My writing style:
- [1-2 sentences about how you naturally write, e.g., "I write short sentences. I don't hedge."]
- I never use: [list 3-5 words or phrases you'd never say, e.g., "utilize, it's worth noting, let's dive in"]
- I do use: [words or phrases you actually reach for, e.g., "honestly, the thing is, which means"]
- Tone: [one adjective, e.g., "direct" or "conversational" or "skeptical"]
- Audience: [who you're writing for]

Here's what I want to rewrite:
[paste your AI draft here]

Do not add new ideas. Do not change the structure. Only adjust word choice and sentence rhythm so it reads like I wrote it.

The last instruction matters more than people think. Without it, ChatGPT will often expand the piece, smooth out the transitions, and generally undo your original structure. Pinning it to "word choice and rhythm only" keeps the changes tight.

How to fill in your voice details

Most people stall here because they don't know how to describe their own writing. A simple exercise: pull up three emails or messages you've sent this week. Read them out loud. What do you notice?

You probably have habitual phrases. You probably have a default sentence length. You might hedge a lot, or you might not hedge at all. You might always start with a short declarative and then explain.

Write those down, in plain language. You don't need to be precise. "I write like I'm talking to a colleague, not a client" is more useful than "I employ a conversational register with low formality markers."

For the "never use" list, think about what makes you cringe when you read it in your own drafts. That's your list. Common ones: "utilize," "leverage," "it's important to note," "in today's world," "this is because." Whatever words you'd delete on sight.

Before and after: what this actually changes

Here's a real example of what the prompt does.

Before (ChatGPT default):

"It's important to note that email marketing serves as a crucial component of any comprehensive digital marketing strategy. By leveraging personalized content, businesses can foster deeper connections with their audience and drive meaningful engagement."

That sentence has five of the patterns I listed above: "it's important to note," "serves as," "crucial," "leverage," "foster." It reads like a whitepaper abstract from 2019.

After (with the voice prompt applied):

"Email still works. People check it every day, and a well-timed message from a brand they trust gets read. The mistake most companies make is writing emails that sound like ads, not like people."

Same idea. Different shape. The second version has an opinion in it and makes choices about what to emphasize.

Your results will vary depending on the specifics you give the model, but even a basic voice brief produces noticeably less generic output.

Feeding the model your own writing as reference

The template above works from a description of your voice. You can go further by giving the model actual samples.

Add this to the top of your prompt:

Here are 3 examples of how I actually write:

[Example 1 — a paragraph from something you've written]
[Example 2]
[Example 3]

Now, using that as your reference for my voice and rhythm, rewrite the following...

Three examples is usually enough. One is too thin to pattern-match against. More than five and the model starts averaging across them rather than picking up the consistent thread.

Pick examples that are representative, not aspirational. The sample you're proudest of might not be how you write under normal conditions. A few paragraphs from an email or a blog post you dashed off in twenty minutes is often more useful.

If you write regularly, this kind of personal voice prompt is worth building once and saving. You can paste it in at the top of any new conversation.

The pieces that prompt-engineering can't fix

A good voice prompt reduces the most obvious tells. It won't eliminate all of them.

ChatGPT has structural habits that survive prompt instructions. It will still tend toward symmetrical paragraph lengths, even if you tell it to vary them. It will still sometimes write transitions that announce themselves ("Now let's look at..."). And it will occasionally backslide into banned vocabulary mid-draft.

This is why the prompt should be the start of your editing pass, not the end of it. After you run the rewrite, read it out loud. Anything that makes you stumble, rewrite by hand. Any phrase you wouldn't have chosen, swap it out.

The goal isn't to get AI output that passes as human to a machine detector. It's to get a draft that you can finalize quickly because it already sounds like your thinking. That's a more useful framing, and it tends to produce better results.

For a deeper look at how these patterns work across different models, the comparison of humanizer prompts for ChatGPT, Claude, and Gemini is worth reading before you settle on a single approach.

If you want to go further on the system-prompt side (particularly for people who use the same model repeatedly), how to write a system prompt that strips out AI tells covers that in more detail.

---

FAQ

Does this work better in ChatGPT than other models?

The same prompt structure works in Claude and Gemini too. The results differ a little because each model has different default tendencies — Claude, for instance, tends toward longer explanations and more hedged language, while GPT-4o writes shorter paragraphs by default. You may need to tune the "never use" and tone sections differently for each. If you switch between models often, it's worth keeping separate voice briefs.

Will using a voice prompt get me flagged as AI by detectors?

Humanizing your output does reduce the probability score on most AI detectors. But detector results vary significantly across tools and are not a reliable signal — the same passage can score 20% on one tool and 80% on another. If you're in a context where that matters (academic, journalistic, etc.) you should understand the policies that apply to your situation. Our advice here is about making writing sound natural, not about gaming any particular system.

How specific do I need to be in the voice brief?

More specific is generally better, but the law of diminishing returns kicks in fast. A brief that runs more than about 150 words tends to create conflicting signals that confuse the model. Aim for tight and clear over comprehensive. Five concrete examples of phrases you'd never say beat two paragraphs of general stylistic description.

What if I don't know how to describe my own writing style?

Start with what you'd cut. Paste a recent piece of your writing into a doc, then go through and delete every sentence that doesn't sound like you. What's left is your voice. Notice what patterns hold: how long are the sentences? What words do you repeat? That's enough to build a brief from.

Should I use this for everything I write?

Not necessarily. For short outputs (subject lines, captions, quick replies), the friction of building a full voice prompt isn't worth it. This technique pays off most on longer-form content: blog posts, essays, LinkedIn articles, email newsletters. Anything where the cumulative effect of word choice shapes how readers perceive you. For short pieces, the best prompt to make AI writing sound human has a lighter-weight version that works well at smaller scale.