OpenAI added Writing Style to ChatGPT Work on paid plans. Connect Gmail, Slack, Google Drive, or SharePoint and it pulls real emails, messages, and files to copy your phrases, sign-offs, sentence rhythm, and even capitalization quirks. After setup it applies that voice to new drafts on web and mobile, so you stop restating tone in every prompt.
Until now you either pasted samples or wrote a style block and hoped the model obeyed. This version skips the paste step. It reads work you already produced. Setup is Settings > Personalization > Writing Style, then connect the apps. OpenAI says it only sees content you can already access and that you authorized. Business, Enterprise, and Edu work data is off-limits for model training by default.
A reply that used to take a prompt plus two rewrite passes can now leave ChatGPT looking like yesterday’s sent mail. That is the useful part: client follow-ups, proposals, and status updates stop sounding like a template. Pair it with webhook tasks and the loop is concrete. New client mail arrives, Work summarizes it, drafts a reply in your voice, and flags the one decision that still needs you.
Do not connect a shared inbox on day one. Start with your personal sent mail and a private Slack DM history. Read two or three generated drafts against real ones before you let it send anything. If the sample set is mostly Slack one-liners, your proposals will get sloppy. Feed it longer docs if that is the work that matters.
Test it this week on one mailbox and one content type. If the drafts need only a light edit, keep it. If they leak a tone you use with friends but not clients, disconnect Slack and keep Gmail only.