ChatGPT product improvement: choosing a segment, top pain point and success metrics

ChatGPT is OpenAI's AI assistant. How would you make it a better product? Decide where to focus, explain why, and walk through what you would build and how you would know it worked. ```hint One job, one segment People use ChatGPT for writing, coding, studying, research and advice, and some of them stop coming back after a few tries. Pick one segment and one job, then look for the moment where ChatGPT lets them down. ``` ### Clarifying Questions - Is the scope the ChatGPT app on web, desktop and mobile, or also the developer API? - Should "better" mean more usage and retention, higher answer quality, or more conversion to paid plans? - Which users should I prioritize: free users, paying individual subscribers, or companies rolling ChatGPT out to employees? - Are changes to the model's capabilities in scope, or only the product built around the model? ### What a Strong Answer Covers - A stated goal and a segment chosen from ChatGPT's very broad user base, with the reason for the choice - Pain points tied to how that segment uses ChatGPT, separating problems with the answers from problems with the workflow around them - A prioritized set of solutions with a first release, and what was cut - Success metrics that include a way to tell whether answers are actually useful, plus guardrails - Risks particular to a generative AI product, such as confident wrong answers, privacy of what users share and inference cost ### Follow-up Questions - Many people have tried ChatGPT once or twice and never became regular users. How would you find out why, and what would you change? - Thumbs-up and thumbs-down ratings cover only a small share of chats. How else would you tell whether answers are improving? - Your feature lifts usage but raises inference cost per user. How do you decide whether free users get it?