How to make ChatGPT write in your voice (and stop sounding like AI)

To make ChatGPT write like you, stop describing your voice and start showing it. Feed it three to five pieces of your own writing, have it extract a written voice profile, add a list of words and patterns you never use, then save that profile where every new chat can see it. This guide gives you the prompts for each step, plus an editing pass that strips the AI feel out of any draft.

Why "write in my tone" does not work

Most people try something like "write this in a casual, friendly tone" and get the same polished, slightly breathless prose everyone recognizes. The problem is that "casual" and "friendly" describe millions of writers. The model fills the gap with its own defaults: balanced three-part lists, tidy summaries, safe vocabulary, a neat closing line.

Your voice lives in details you probably have never named. How long your sentences run. Whether you start with the point or the story. Which words you overuse. Whether you write "I think" or just say it. The method below captures those details from real samples, so you do not have to guess at them. It builds on the same basics covered in how to write better ChatGPT prompts: specific beats vague, every time.

Step 1: collect the right writing samples

Pick samples that already sound like you on a good day. A useful set:

• 3 to 5 pieces, roughly 300 to 1,500 words in total.
• The same kind of writing you want help with. If you want newsletters, use newsletters. Your voice in a work email differs from your voice in a personal blog post.
• Lightly edited or unedited. Heavily polished pieces, or anything someone else rewrote, carry their voice instead of yours.
• Nothing written with AI help. Otherwise you are teaching the model to imitate itself.

Good sources: emails you sent, old posts, Slack messages you were proud of, voice memo transcripts. Remove anything confidential before you paste it in.

Step 2: extract a voice profile with one prompt

Instead of asking ChatGPT to imitate your samples directly, ask it to analyze them first. A written profile is reusable, editable and much shorter than the samples themselves.

I'm going to share samples of my own writing. Analyze them and write a voice profile that another writer could follow to sound exactly like me.

Cover these points, with a short quoted example from my samples for each:
1. Sentence length and rhythm (average length, how often I use very short sentences, fragments)
2. How I open a piece and how I end one
3. Vocabulary: words and phrases I use often, level of formality, any slang or jargon
4. Punctuation habits (parentheses, colons, questions, contractions)
5. How I make a point: story first, claim first, examples, numbers
6. Humor: whether I use it, and what kind
7. What I never do (things a generic writer would do that I avoid)

Then write the profile as a list of 10 to 15 direct instructions, starting each with a verb.

My samples:
"""
[paste sample 1]
"""
"""
[paste sample 2]
"""
"""
[paste sample 3]
"""

Read the result critically. If a line feels wrong ("uses vivid metaphors" when you never do), delete it. If something obvious is missing, add it. The profile is yours to edit, and editing it is where it gets accurate.

Step 3: write a banned-words and banned-patterns list

Even with a voice profile, models drift back to their favorite habits. A short ban list stops the most recognizable ones. Start with this and adjust it to taste:

Never use these words or phrases: crucial, robust, leverage, foster, realm, embark, testament, vibrant, streamline, harness, comprehensive, furthermore, moreover, "in conclusion", "at the end of the day", "let's dive in", "it's worth noting".

Never use these patterns:
- Em dashes. Use commas, colons or full stops.
- "It's not X, it's Y" or "This isn't just X, it's Y."
- Lists of exactly three adjectives or three parallel phrases, unless the content truly has three items.
- Opening with a rhetorical question.
- Ending with a summary of what was just said, or a motivational closing line.
- Exclamation marks.
- Starting sentences with "So," or "Now,"
- Hedging phrases like "it can be said that" or "in many ways".

Then add your own. Read three AI drafts on your topic and note every phrase that made you wince. Those go on the list. Keep it under 30 items; very long ban lists are harder for the model to apply consistently, and some lines will start contradicting your profile.

Step 4: fix rhythm and specificity

Two things make text sound human more than anything else: varied sentence length and concrete detail. Add these instructions to your profile:

Rhythm:
- Mix sentence lengths. Follow a long sentence with a short one. Allow one-line paragraphs.
- Don't make every paragraph the same size.

Specificity:
- Prefer a concrete example, number, name or object over an abstract claim.
- If you make a general point, follow it with a specific case from [my work / my industry].
- If you don't have a real detail, leave a placeholder like [EXAMPLE] instead of inventing one.

The last line matters. Asked to "be specific" without real material, a model will happily invent a client, a percentage or an anecdote. Placeholders keep the draft honest and show you where your own experience needs to go.

Here is the difference in practice.

Before:

Write a LinkedIn post about why I stopped taking meetings on Fridays. Make it casual and authentic.

After:

Write a LinkedIn post about why I stopped taking meetings on Fridays.

Follow my voice profile below exactly. Facts to use: I run a 6-person design studio called Northlight. I made the change in March. Fridays are now for client work only. Our turnaround on revisions got faster and I stopped working Saturdays.

Under 180 words. Open with the decision itself, not a question. No hashtags. Leave [DETAIL] where a personal example would help and I'll fill it in.

Voice profile:
[paste your profile and ban list]

Why the after works: "casual and authentic" gives the model nothing to hold on to, so it produces the familiar hook, three-point list and inspirational ending. The second version supplies real facts, a length, a structure rule and your actual profile, and it tells the model what to do when it runs out of facts.

Step 5: save your voice so you never paste it again

Once your profile works, store it so it applies automatically.

• ChatGPT custom instructions apply to every new chat. You will find them under Settings, then Personalization. The field has a character limit that depends on your plan, so keep the profile tight. See OpenAI's custom instructions help page for the current details.
• ChatGPT Projects let you set instructions for one project only, and those override your global custom instructions inside that project. This is the better choice if you write in more than one voice, for example a "Newsletter" project and a "Client emails" project, each with its own profile.
• Other assistants have similar features. Claude and Gemini both let you save standing instructions for a project or a reusable assistant, so the same profile works there too.

You can also upload your original samples as files in a project, alongside the written profile. The profile tells the model what to do; the samples show it. That combination is a form of few-shot prompting, and it tends to beat either one alone. For more on how standing instructions are structured, see these system prompt examples.

Step 6: run an editing pass on every draft

Even good drafts need a second look. This prompt turns the model into an editor that checks its own output against your rules:

Act as a strict editor. Review the draft below against my voice profile and ban list.

1. List every sentence that breaks a rule, and say which rule.
2. List any phrase that sounds generic or could appear in anyone's writing.
3. List any claim, number or example that isn't in the facts I gave you, so I can verify or delete it.
4. Then rewrite the draft with all fixes applied. Change as little as possible beyond the fixes.

Voice profile and ban list:
[paste]

Draft:
"""
[paste draft]
"""

Step 3 of that prompt is your safety net. It flags anything the model made up, which is the fastest way for AI-assisted writing to damage your credibility.

Here is a before and after rewrite produced with this method, for a short newsletter intro.

Before:

In today's busy world, finding time to cook can feel overwhelming. But what if we told you that meal prep doesn't have to be stressful? This week, we're diving into three simple, delicious, and budget-friendly recipes that will transform your weeknights.

After:

I cooked four dinners on Sunday afternoon and didn't touch the stove again until Thursday. That's the whole pitch for this week. Below are the three recipes I used, what each one costs per portion, and the one I'd skip if you only have an hour.

The after follows a profile that says: open with a concrete thing I did, short sentences, no rhetorical questions, no stacked adjectives, promise specifics. The facts (four dinners, Sunday, Thursday) came from the writer, not the model.

A shortcut for the prompt structure

The voice profile is the part only you can write. The prompts around it, with a clear role, task, context, format and constraints, are repetitive to build. If you write with AI every day, a prompt generator such as Promptay can turn a one-line idea into that full structure, and you paste your voice profile into the context. Pair it with ready-made templates like our ChatGPT prompts for emails or ChatGPT prompts for marketing, and swap their tone lines for your profile.

Frequently asked questions

Can ChatGPT really learn my writing style?

ChatGPT does not permanently learn your style from one chat, but it imitates a style well when you give it samples and a clear written description in the prompt or in saved instructions. The most reliable approach is a voice profile extracted from your own writing, stored in custom instructions or a project, plus a short list of phrases to avoid.

How many writing samples should I give ChatGPT?

Three to five samples, totaling roughly 300 to 1,500 words, is enough for a useful voice profile. Choose pieces of the same type as the writing you want help with, and avoid anything heavily edited by someone else or written with AI help, since those carry a different voice.

How do I stop ChatGPT from sounding like AI?

Ban the specific words and patterns that give it away, such as stacked three-part lists, rhetorical question openers, em dashes and summary endings. Ask for varied sentence length and concrete details, and tell the model to leave placeholders instead of inventing examples. Then run an editing pass that checks the draft against your rules.

Where should I save my voice profile in ChatGPT?

Use custom instructions if you write in one main voice, since they apply to every new chat. Use a project with its own instructions if you write in several voices, because project instructions override the global ones inside that project. Keep the profile short, since the custom instructions field has a character limit.

Is it okay to publish writing that ChatGPT drafted in my voice?

That depends on where you publish and what you have agreed to, so check the rules of your employer, client, school or platform. Whatever the context, review every draft yourself, verify any facts or claims, and add your own experience. The voice profile makes the draft sound like you, but you are still responsible for what it says.