ChatGPT is learning to write like you, by reading your Gmail and Slack
OpenAI is testing a 'Writing Style' feature that studies your own messages and documents so the chatbot can mimic your voice. The privacy trade is real.
Key points
- OpenAI is testing a ChatGPT feature called Writing Style that copies a user's tone by reading examples from connected apps.
- The tool pulls samples from two broad source types: messaging via Slack, and documents and email via Google Drive, Notion and Gmail.
- OpenAI has confirmed the test to BleepingComputer but hasn't said when, or if, it will launch to everyone.
- Anthropic's Claude offers a similar feature called Styles, but users paste in samples themselves rather than granting access to live accounts.
- The design raises fresh questions about what a chatbot can see once you connect it to your personal inbox and workspace.
OpenAI is quietly testing a feature that lets ChatGPT write in your voice, by reading things you've already written elsewhere.
It's called Writing Style, and a small group of users can see it now. An onboarding screen tells them: "ChatGPT will write in your voice by referencing examples from your connected apps."
Instead of pasting old emails into a chat window every time you want ChatGPT to sound like you, the bot retrieves those emails itself.
What is ChatGPT actually looking at?
It looks at writing you've produced inside apps you choose to connect. OpenAI groups the sources into three categories: messaging, documents, and email.
The connected apps on offer are Slack for chat, Google Drive and Notion for longer documents, and Gmail.
| Category | Connected app |
|---|---|
| Messaging | Slack |
| Documents | Google Drive, Notion |
| Gmail |
Once switched on, ChatGPT can pull real examples of how you write in each setting. Most of us don't have one voice: you're chattier in Slack than in a board memo, shorter in email than in a Notion doc, and the model now gets to observe all of that directly.
How is this different from what Claude already does?
Anthropic's rival chatbot Claude has offered a similar tool called Styles. With that approach, you paste or upload a few writing samples and Claude builds a template from them.
OpenAI's version is bolder. Rather than asking you to hand over samples, it plugs directly into services where your writing already lives, giving the model far more raw material and a more honest picture of how your tone shifts between a quick message and a project brief.
It also means giving a chatbot standing access to your inbox and work notes. That's a much bigger ask than dropping three paragraphs into a text box.
Should ordinary users be worried?
Not panicked, but thoughtful. This is authentication and authorisation working exactly as designed, which is precisely the point worth pausing on.
When you connect Gmail or Slack to ChatGPT, you're using OAuth, the standard protocol by which one app requests permission to read your data. You log in, tick a consent box listing what ChatGPT can access, and a token gets stored. From then on, ChatGPT can reach into that account without asking again until you revoke it.
The security question isn't whether OAuth is safe. It's well understood. The real question is which scopes, meaning the specific permissions ChatGPT requests, it actually needs, and whether the average user reads the consent screen before clicking Allow. Most don't.
Multi-factor authentication, the extra verification code after your password, wouldn't change any of this. You granted the access on purpose. That's the awkward part of the modern AI stack: the biggest risk is often a permission you agreed to while half paying attention. Our OAuth coverage going back to 26 May 2026 has shown this pattern repeatedly across 47 stories.
What should you actually do?
A few sensible habits, if you try Writing Style.
Read the consent screen. If ChatGPT is asking to read all mail rather than just drafts, notice that. Check your Google account's connected-apps list every few months and remove anything you no longer use. Keep work and personal accounts separate, so a personal experiment doesn't quietly hand a chatbot your employer's data.
The feature worth watching next isn't the style mimicry itself. It's how the connector list grows: Slack and Gmail today, calendar and browser history tomorrow. Each addition widens the scope of what the model has seen, and consent fatigue means most users won't notice.
OpenAI hasn't said when Writing Style will roll out widely.
