OpenAI is testing a Writing Style feature in ChatGPT that learns how you write from examples inside apps you already connect to. Bleeping Computer reported the early rollout on September 7. An onboarding screen tells users ChatGPT will write in your voice by referencing examples from connected apps. OpenAI confirmed the test. A wide release date is still unknown.
The setup splits samples into Messaging, Documents, and Email. Slack covers messaging. Google Drive and Notion cover documents. Gmail covers email. Once that is configured, ChatGPT can pull from how you actually write across those tools instead of making you paste samples into every chat or re-explain your tone.
Anthropic already shipped a Styles feature where you upload or paste writing samples. OpenAI's experiment goes further by reading live examples from connected services. That can give the model a bigger, more realistic set of patterns, including the difference between your Slack voice and your email voice.
I want this as a developer who writes docs, tickets, and customer notes all day. Matching my real tone saves time. I also want a clear off switch and a tight scope. Letting a chat product browse Gmail, Drive, Notion, and Slack for style examples is powerful. It is also a privacy decision. Treat the connectors like production access, not a cute demo.
For regular folks, the upside is drafts that sound like you on the first try. The smart move is to start with one low-risk app, review what ChatGPT produces, and only then connect inbox or docs. Personalization should feel like a helpful coworker, not a silent roommate reading everything.
Style mimicry is not mind reading. It is pattern matching on sentence length, greeting habits, and how formal you get when the stakes rise. That can make a draft feel like you wrote it after coffee. It can also copy bad habits, slang you use with friends, or confidential phrasing from a private doc if the connector is too broad.
If you try the beta, give ChatGPT three clean samples you would be fine seeing in public. Skip anything with passwords, customer secrets, or legal language. Then ask for a rewrite of a safe paragraph and check whether it sounds like you or like a cartoon of you. Small tests beat big connector dumps.
