Want to access all premium content?
Become a paying subscriber to get access to all courses, guides like this, prompts and skills library, and $500+ worth of AI tools.

OpenAI has announced an invisible text watermark called textGrain, with a rollout to eligible ChatGPT and Codex users in the EU. It works through patterns in word choices, rather than a visible label attached to the text.
That raises a separate question worth answering: how do you make an AI-assisted draft sound like something you would actually write?
We’ll build a ChatGPT project around your writing samples, use your own notes to shape each draft, and finish with an edit you control. This is a writing workflow, not a promise to remove a watermark.
You’ll need a ChatGPT account, three pieces of your own writing, and one draft or idea to work on.
Step 1: Check what changed. Understand the watermark and its limits.
Step 2: Choose your samples. Find three pieces that sound like you.
Step 3: Build your voice project. Save the writing preferences you want ChatGPT to follow.
Step 4: Start with your notes. Give the draft your ideas and examples.
Step 5: Make the final edit. Read it aloud, check the claims, and fix anything that feels off.

Step 1: Check what changed
OpenAI announced textGrain on October 5, 2026. Its launch plan covers eligible ChatGPT and Codex output in the EU, with optional access for API customers globally on selected models.
The watermark is a statistical signal in the wording. Copying and pasting doesn’t strip it away like file metadata.

OpenAI says detection has limits. A result doesn’t identify the user, establish ownership, or measure how much a person contributed. Detector access is initially restricted to approved researchers and expert organizations.

The announcement doesn’t make watermarking a global ChatGPT default. The writing steps below work regardless of where you live: they help you develop a draft around your own voice and judgment.
Step 2: Choose your samples
Pick three pieces you wrote yourself. A social post, an email, and a longer article give ChatGPT a useful range. Aim for around 200–500 words each where possible.
Choose work that sounds like you on a normal day. A polished piece can be useful, but skip anything someone else substantially rewrote.
Remove private names, contact details, and confidential client information before uploading.

Save the samples in one file called my-samples.txt, with a label above each one.
If you don’t have three written pieces, use a transcript of yourself explaining an idea. It can reveal phrases and sentence patterns that formal writing leaves out.
Read each sample before moving on. Would someone who knows your writing recognize it? If a sample feels unusually stiff or unlike you, replace it.
Step 3: Build your voice project
Create a project in ChatGPT called My Voice.

Start a chat inside it and paste this prompt, followed by your samples:
These are three examples of my own writing.
Create a voice profile I can review and paste into this project's instructions. Keep it under 250 words and use only these six numbered points:
1. Sentence length and rhythm, including how I use short sentences
2. Words and phrases I use repeatedly
3. How I open and close a piece
4. Punctuation and formatting habits
5. Tone, described in one sentence
6. Five practical writing rules and five habits that would feel out of place
Support points 1–4 with a short, exact quote from my samples.
Separate clear observations from tentative ones. Don't assume I avoid a word just because it doesn't appear in these samples.
[SAMPLE 1]
[SAMPLE 2]
[SAMPLE 3]
Read the profile line by line. You’re reviewing its observations, not accepting a diagnosis of how you write.
For example:
“I use short sentences for emphasis, but I don’t want every sentence shortened. Adjust that rule.”Or:
“The exclamation marks are specific to that email. Don’t make them a general preference.”
Once it looks right, open the project’s settings and paste the profile into its instructions. Add this above it:
Use the voice profile below when writing or rewriting in this project.
Preserve my facts, examples, and opinions. Don't invent personal experiences or commitments.
If information is missing, ask or mark it for review. Follow the format and length I request for each piece.

Start a new chat inside the project and ask:
“Describe the writing preferences you’ll follow in three lines.”Check that the answer reflects your saved profile. You can also ask for a short test paragraph and correct anything that feels exaggerated.
Step 4: Start with your notes
Your writing style is only part of what makes a piece yours. The opinion, example, and point you’re making matter just as much.
Before asking for a draft, write three to six rough bullets. Include what you think, something you’ve noticed or experienced, and what you want the reader to understand or do.
The notes don’t need to be polished. A specific detail is more useful than a perfectly phrased instruction.
Inside My Voice, paste:
Turn these notes into a [LINKEDIN POST / EMAIL / NEWSLETTER SECTION] of about [WORD COUNT] words.
Follow my project voice profile.
Use only the points and examples below. Don't add statistics, stories, personal experiences, or claims I haven't supplied.
Keep my opinion as direct as I've written it. Preserve any phrases that already work.
Notes:
[PASTE NOTES]

Already have a draft? Use:
Rewrite this draft using my project voice profile.
Keep the facts, main points, and requested structure. Replace wording that doesn't fit my style without weakening the meaning.
Don't invent examples or add new claims. Mark unclear or unsupported statements [CHECK].
Draft:
[PASTE DRAFT]
Compare the result with the source. Check that the rewrite kept your point, rather than replacing it with a broader, safer-sounding statement.
If a sentence could appear in anyone’s post, ask what specific example would make it useful. Supply that example yourself.
Step 5: Make the final edit

Read the draft aloud. Notice where you hesitate, where the language feels too formal, and where a sentence takes too long to reach its point.
Change those lines yourself or give a precise instruction:
“Keep the example, but shorten the setup. I’d say ‘we tried it’ rather than ‘we implemented this approach.’”Add a relevant detail from your own experience if the piece needs one. Don’t force a personal story into an email or explanation that already works.
Then ask for a review:
Check this edited text against my notes and voice profile.
Return only two short lists:
1. Claims to review
List factual claims, numbers, names, and commitments. Mark whether each is supported by my notes or needs checking elsewhere.
2. Voice mismatches
Quote sentences that don't fit my saved preferences and explain briefly why.
Don't rewrite the piece or search the web. Don't treat a claim as verified just because it sounds familiar.
Notes:
[PASTE NOTES]
Edited text:
[PASTE TEXT]

Work through the flagged claims. Confirm details you know firsthand and find sources for claims that need outside evidence.
The review is another set of eyes, not a fact-checking guarantee. Read the finished piece once more before publishing.
If your client, employer, or school requires disclosure of AI assistance, editing the wording doesn’t change that requirement.
Use cases
Try these three jobs inside My Voice once you’ve tested the profile.
Use case #1: Turn a voice note into a post
Record yourself talking through an idea, then paste the transcript.
Turn this voice-note transcript into a post under 200 words.
Keep my strongest phrases and main point. Remove filler and repetition, but preserve the meaning.
Don't add facts or personal experiences.
Transcript:
[PASTE TRANSCRIPT]
Check names, numbers, and any phrases the transcription may have misheard.

Use case #2: Reply to an email
Give ChatGPT your answer before asking it to phrase the reply.
Write a reply in my voice.
My answer: [YES / NO / NOT YET]
Reason: [REASON]
Next step: [NEXT STEP]
Keep it under 120 words. Don't add promises, availability, or explanations I haven't supplied.
Email:
[PASTE EMAIL]
Read it for tone, then check the dates and commitments before sending.

Use case #3: Refresh an old post
Keep the original argument while updating the details you provide.
Refresh this older post in my voice.
Keep my opinion and main argument. Use only the updated examples I provide below.
List any time-sensitive facts that need checking. Don't silently replace them or invent current examples.
Original post:
[PASTE POST]
Updated details:
[PASTE DETAILS]
Compare the versions. Make sure the rewrite hasn’t softened your opinion or presented an old fact as current.

The project can also help with bios, captions, and product descriptions. Keep giving it the facts and purpose of each piece rather than expecting the voice profile to supply them.
Confirmed limitations
OpenAI’s announced rollout is regional, and detector access is restricted. The illustration below reflects that launch scope, rather than a claim that the feature can never expand.

You can’t use this workflow to certify that a draft contains no watermark. Making the wording more personal and checking provenance are different tasks.

Three samples give ChatGPT a starting point. They won’t capture every tone you use, and it may exaggerate patterns or miss context.
A voice profile also doesn’t verify facts. Keep checking names, numbers, quotes, and commitments against the material you supplied.
The original guide was checked on November 3, 2026. Rollout details, detector access, and ChatGPT features may change.
Closing
Start with one short piece. Choose your samples, review the profile, and draft from notes that contain your actual point. Read the result aloud and change the lines you wouldn’t naturally use.
As you write more, add examples that represent your current style and remove ones that no longer fit. Keep the profile useful and specific, with enough flexibility for an email to sound different from a social post.
See you in the next one :)
