OpenAI Optimizing ChatGPT for User Utility and Wellbeing

OpenAI has announced a strategic shift in how it optimizes ChatGPT, prioritizing user utility and wellbeing over traditional engagement metrics. The company is moving away from measuring success by time spent or clicks, focusing instead on whether users achieve their goals efficiently and whether they they return to the product because it is useful.

Optimization for Utility Over Engagement

ChatGPT is being optimized to help users solve problems and learn new things quickly so they can return to their daily lives. OpenAI states that success is measured by whether a user leaves the product having accomplished their intended task, rather than how long they remain in the app.

This approach extends to new capabilities such as the ChatGPT Agent, which allows the model to achieve goals—such as booking doctor's appointments, summarizing inboxes, or planning parties—without the user needing to be active in the app.

Enhancements to User Wellbeing and Healthy Use

To promote healthy use and prevent emotional dependency, OpenAI is implementing several changes to the ChatGPT experience:

  • Time Management: Users will now see gentle reminders during long sessions to encourage them to take breaks.
  • High-Stakes Decision Support: For high-stakes personal decisions (e.g., "Should I break up with my boyfriend?"), ChatGPT is being updated to avoid giving direct answers. Instead, it will act as a sounding board, asking questions and weighing pros and cons to help the user think through the problem.
  • Emotional Distress Detection: OpenAI is improving the 4o model's ability to detect signs of delusion or emotional dependency. The goal is to provide grounded honesty and point users toward evidence-based resources when mental or emotional distress is detected.

Expert-Led Safety and Evaluation

OpenAI is collaborating with a wide range of experts to refine how ChatGPT responds during critical moments, particularly regarding mental health and emotional distress:

  • Medical Expertise: The company worked with over 90 physicians, including psychiatrists, pediatricians, and general practitioners from over 30 countries, to create custom rubrics for evaluating complex, multi-turn conversations.
  • Research Collaboration: OpenAI is engaging with clinicians and human-computer-interaction (HCI) researchers to refine evaluation methods and stress-test product safeguards.
  • Advisory Group: An advisory group comprising experts in mental health, youth development, and HCI has been convened to ensure the approach aligns with current research and best practices.

Sources