A team of researchers from OpenAI and Harvard University, including economist David Deming, today released the largest study ever conducted on how consumers use a generative AI chatbot. The NBER working paper, titled ‘How People Use ChatGPT,’ is based on a privacy-preserving analysis of roughly 1.5 million conversations drawn from the platform’s approximately 700 million weekly active users — about 10% of the world’s adult population.

The study finds that non-work-related messages have grown from about 53% of all usage in mid-2024 to more than 70% in June 2025, suggesting ChatGPT is increasingly embedded in daily life. Researchers classified conversations into three buckets: Asking (advice and information), Doing (task completion), and Expressing. About half of all messages fall under Asking, a share that has been growing. Writing accounts for 40% of work-related messages, while coding remains a relatively small share.

The gender gap among users has narrowed dramatically since early 2024, when 37% of users had feminine-coded names; today, female adoption is consistent with general adult demographics. Growth in lower-income countries now outpaces that in higher-income countries by a factor of more than four. Meanwhile, the paper notes that ChatGPT provides economic value through decision support, especially in knowledge-intensive jobs — improving workers’ judgment rather than replacing them outright.