OpenAI Shifts ChatGPT Success Metrics Away From Time Spent

OpenAI says it will stop treating time spent in ChatGPT and click counts as measures of success, and instead optimize for whether users finish what they came to do. "Our goal isn't to hold your attention, but to help you use it well," the company wrote in a blog post titled "What we're optimizing ChatGPT for."

The shift matters because it runs against the incentive structure that has defined consumer software for two decades, where engagement metrics drive product decisions. ChatGPT sits at the center of a competitive race with Google's Gemini, Anthropic's Claude, and others, and OpenAI's subscription business — the core of its revenue — depends on users finding the product worth paying for over the long term. "If ChatGPT genuinely helps you, you'll want it to do more for you and decide to subscribe for the long haul," the post reads.

Success redefined

OpenAI frames the new metric bluntly: "Instead of measuring success by time spent or clicks, we care more about whether you leave the product having done what you came for." The company also tracks whether users return daily, weekly, or monthly — retention as a signal of usefulness, not compulsion.

"Often, less time in the product is a sign it worked," the post states. With ChatGPT Agent, the company argues, value increasingly happens without the app open at all — booking a doctor's appointment, summarizing an inbox, or planning a birthday party.

The examples OpenAI gives point at personal, high-stakes territory: preparing for a difficult conversation with a boss, deciphering lab results, or untangling a stuck line of thinking.

Break reminders ship today

The most concrete product change is immediate. "Starting today, you'll see gentle reminders during long sessions to encourage breaks," OpenAI wrote. The reminders show a prompt — "You've been chatting a while — is this a good time for a break?" — with options to keep chatting or mark the session as helpful. OpenAI said it will keep tuning when and how the prompts appear "so they feel natural and helpful."

That feature lands in a policy environment where regulators and researchers increasingly scrutinize AI companionship and dependency, particularly for minors and people in emotional distress.

OpenAI admits the sycophancy failure

The post also acknowledges a known misstep. "We don't always get it right," OpenAI wrote. "Earlier this year, an update made the model too agreeable, sometimes saying what sounded nice instead of what was actually helpful."

The company said it rolled back that GPT-4o update, changed how it uses feedback, and is reworking how it measures real-world usefulness over the long term, "not just whether you liked the answer in the moment." The rollback followed public criticism that the model was validating users rather than helping them.

On emotional distress, OpenAI was direct about a specific gap: "There have been instances where our 4o model fell short in recognizing signs of delusion or emotional dependency." It described such instances as rare and said it is developing tools to better detect signs of mental or emotional distress so ChatGPT can respond appropriately and point people to evidence-based resources.

New behavior for personal decisions

OpenAI is also changing how ChatGPT handles intimate questions. "When you ask something like 'Should I break up with my boyfriend?' ChatGPT shouldn't give you an answer," the post states. "It should help you think it through — asking questions, weighing pros and cons." New behavior for high-stakes personal decisions is rolling out soon.

The company summarized its stance: ChatGPT should guide, not decide, when users face personal challenges — "helping you stay in control of your time, and guiding — not deciding — when you face personal challenges."

Expert review infrastructure

To back the changes, OpenAI is building external review capacity:

• Medical expertise. OpenAI worked with over 90 physicians across more than 30 countries — psychiatrists, pediatricians, and general practitioners — to build custom rubrics for evaluating complex, multi-turn conversations.
• Research collaboration. Human-computer-interaction researchers and clinicians are giving feedback on how OpenAI has identified concerning behaviors, refining evaluation methods, and stress-testing product safeguards.
• Advisory group. A standing advisory group of experts in mental health, youth development, and HCI will keep the company's approach aligned with current research and best practices.

OpenAI said this work is ongoing and it will share more as it progresses.

The test OpenAI sets for itself

The post closes with a personal benchmark the company says governs the effort: "if someone we love turned to ChatGPT for support, would we feel reassured? Getting to an unequivocal 'yes' is our work."

The stated commitment — usefulness over engagement, measured by long-term outcomes rather than in-the-moment approval — sets a standard OpenAI will now be judged against publicly, particularly as reports of user dependency on AI chatbots continue to accumulate.