ChatGPT Work Adds Feature to Learn Writing Style from Connected Apps
OpenAI's ChatGPT Work for enterprise users has recently added a feature that learns and mimics users' personal writing styles by analyzing historical writing content from connected third-party applications. This feature is currently in a limited testing phase.
The feature captures commonly used phrases, signature habits, and writing quirks by analyzing historical writing samples from connected accounts such as Gmail, Google Drive, Slack, and SharePoint. Users can configure this in the "Writing style" option under the "Personalization" settings on the web interface, allowing AI-generated emails, documents, and message drafts to better reflect the user's authentic tone.
Reports indicate that early testing users have positively evaluated the feature for maintaining consistent content style in professional contexts and aligning with personal expression habits. However, some users have expressed concerns regarding privacy issues and whether this feature will expand to other usage modes of ChatGPT in the future.
According to another English report, this feature is currently being tested with a limited number of users, and the system prompts that "ChatGPT will refer to examples in your connected applications to mimic your writing style." Users can provide writing samples for different categories such as messages, documents, and emails. OpenAI has confirmed that it is testing the "Writing Style" feature but has not disclosed a specific timeline for a broader rollout.
The report notes that OpenAI has not disclosed specific data processing details for this feature, including the storage duration of reference writing samples and whether these samples will be used for model training. The company currently has privacy protection options like "Lockdown Mode," but it has not clarified whether similar protective mechanisms will apply to this new feature.
The report compares this feature with Anthropic's "Styles" feature for Claude, pointing out that both approaches are similar, but ChatGPT's method involves directly connecting to third-party applications used by the user to extract writing samples, rather than requiring users to manually upload example texts.
Source: Public Information
ABAB AI Insight
OpenAI's iterative path for personalization in ChatGPT has evolved from early "user manual input of custom instructions describing preferences" to "memory functionality automatically accumulating conversation history," and now to directly integrating with third-party office applications like Gmail, Slack, Google Drive, and SharePoint to extract real writing samples. This path coincides with the launch of Claude's "Styles" feature earlier this year, but the methods of sourcing material differ—Claude relies on users manually uploading example texts, while ChatGPT opts for direct access to users' real-time office data sources.
From a product positioning perspective, this feature is first implemented in ChatGPT Work for enterprise customers, rather than the consumer version of ChatGPT. This prioritization is driven by clear capital logic—enterprise users have a higher willingness to pay and higher average transaction values than individual users. The ability to "make AI write like oneself" precisely addresses the pain points of enterprise employees in handling emails, documents, and internal messages in high-frequency office scenarios. OpenAI aims to further strengthen ChatGPT Work's competitiveness against Microsoft Copilot and Google Workspace's built-in AI functionalities in the enterprise productivity tools market, with capital investment focusing on transforming data integration capabilities in office scenarios into a differentiated selling point for enterprise subscriptions.
This aligns with Microsoft's Copilot deeply integrating data from its own office ecosystem, such as Outlook, Teams, and SharePoint, and Google Workspace's built-in Gemini analyzing Gmail and Docs' historical records. All major tech companies are competing for the ability to "most accurately replicate users' real writing habits." However, OpenAI does not own native office infrastructure like email and cloud storage, relying instead on third-party applications for data integration, which places it at a structural disadvantage compared to companies like Microsoft and Google that possess both office infrastructure and AI capabilities. The AI office assistant industry is currently transitioning from a "general large model capability competition" to a "personalized, scenario-based writing experience refinement competition."
Essentially, this represents a "shift in pricing power" coupled with a data sovereignty struggle—at the mechanism level, the core value of this feature does not stem from OpenAI's own model capabilities but from the historical data assets stored by users in third-party applications (Gmail, Slack, etc.). OpenAI "borrows" these user behavior data, originally scattered across different platforms and controlled by various vendors, through authorized connections to enhance its product experience. This process highlights that the key pricing power in current AI product competition is gradually shifting from "the intelligence level of the model itself" to "who can legally and compliantly connect to more third-party data sources." Users' privacy concerns regarding such cross-platform data scraping behaviors are becoming a compliance variable that vendors need to address directly in this competition.
ABAB News · Cognitive Laws
The intelligence of the model is public, while user data is scarce.
The more applications one can connect to, the more they resemble the AI that "understands you."
The end of personalization is the starting point for repeatedly testing privacy boundaries.