People are increasingly spending more time with generative AI assistants. In fact, OpenAI’s ChatGPT now has nearly 700 million weekly active users. In a blog post published today, OpenAI emphasized that ChatGPT was designed to help users make progress, learn something new, or solve problems—not to capture their attention indefinitely.
While most online services measure success through metrics like time spent or clicks, ChatGPT takes a different approach, focusing on whether users leave after accomplishing their goal. OpenAI also monitors how frequently users return, whether daily, weekly, or monthly, as a sign of how valuable they find the service. To better align with these goals, OpenAI announced the following changes to ChatGPT today:
Improved Distress Detection: While ChatGPT is trained to be honest, the GPT-4o model has sometimes fallen short in recognizing signs of delusion or emotional dependency. OpenAI is improving the model’s ability to detect signs of mental or emotional distress so that ChatGPT can respond more appropriately.
Break Reminders: Starting today, ChatGPT will display gentle reminders during long user sessions to encourage taking breaks. OpenAI will continue refining how these reminders appear to ensure they feel natural and helpful.
Handling Personal Questions: People often ask ChatGPT deeply personal questions, such as, “Should I break up with my boyfriend?” In such cases, ChatGPT will avoid giving definitive answers. Instead, it will help users think through their decisions by asking questions and weighing the pros and cons. This new behavior for handling high-stakes personal decisions is coming soon.
To design these behaviors, OpenAI collaborated with experts in relevant fields to improve how ChatGPT responds in sensitive situations. For example, the team worked with more than 90 physicians from over 30 countries to improve responses to medical-related queries. OpenAI also emphasized that this is an ongoing effort and that it will share more updates as the work progresses.