Particle.news

OpenAI Tests ChatGPT ‘Writing Style’ That Reads Users’ Emails and Messages

The experiment signals a push to personalize workplace AI while privacy rules for connected apps remain unsettled.

Overview

  • OpenAI confirmed late Monday that it is testing a “Writing Style” or “Reference my writing style” option that lets ChatGPT learn from a user’s existing writing.
  • The tool can access examples from linked services such as Gmail, Google Drive, Outlook, Slack, Microsoft Teams, SharePoint and Notion to model an individual’s voice.
  • The feature is visible only to a small set of users and is reported to be available to paid ChatGPT subscribers and ChatGPT Work customers on the web.
  • A PCWorld hands-on test found ChatGPT could reproduce personal quirks like preferred phrases and punctuation but the output still read as synthetic and was flagged by the AI detector GPTZero.
  • Key questions remain about how much content will be scanned, what privacy controls users will have, and whether outputs will be traceable, which will shape adoption, misuse risks and regulatory scrutiny.