A new study by researchers from OpenAI, Duke University, and Harvard University offers one of the most extensive research to date on how ChatGPT is being used across the world. A National Bureau of Economic Research (NBER) working paper, the study analysed more than a million anonymised conversations from OpenAI’s consumer plans (Free, Plus, and Pro plans) between May 2024 and June 2025, revealing insights into adoption trends and how user behaviour has evolved.
Demographic Shifts
By June 2025, ChatGPT had been adopted by around 10% of the global adult population, with more than 2.6 billion messages exchanged daily. This is in comparison to the 451 million daily messages in June 2024.
The user base of ChatGPT has also diversified. Most notably, usage by gender has reached near parity. Compared to the early months of ChatGPT’s release, where about 80% of users had masculine first names, only 48% of users had masculine first names as of June 2025.
Moreover, the gaps between the age groups of users have narrowed, although the general age profile of ChatGPT users still skews young, with almost half of all messages being sent by adults under the age of 26.
In terms of geographical distribution, adoption is accelerating faster in low and middle-income countries, reflecting the growing demand for digital tools beyond traditional markets.
Distribution of ChatGPT Use Cases
While workplace use continues to expand in absolute terms, personal usage has grown faster. The proportion of work-related queries declined from 47% in June 2024 to 27% in June 2025 while non-work usage has grown from 53% in June 2024 to 70% June 2025, suggesting that users are increasingly exploring non-work applications of ChatGPT in daily life.
In terms of types of use, around 80% of consumer queries fell under 3 categories:
- Practical guidance: The most common use case, which includes tutoring, guidance on topics, and creative brainstorming.
- Seeking information: Similar to web search, where users search for information on current events, facts, products, etc.
- Writing: This includes drafting, editing, critiquing, summarising, and translating documents. 2/3 of writing-related queries focused on refining or editing user-provided text rather than generating new content from scratch. Writing also dominates professional ChatGPT use, accounting for 40% of work-related conversations.
Programming-related queries represent only 4.2% of messages, far lower than comparative studies of rival models such as Claude (33%). Social or companionship-related use cases are also minor, making up around 2.3% of messages.
User Intent
To examine user intent, the study also classified queries into Asking, Doing, and Expressing:
- Asking (49%): Seeking advice, clarification, or information.
- Doing (40%): Requesting ChatGPT to complete specific tasks (eg. Drafting emails, summarising, etc)
- Expressing (11%): Statements that are neither asking for information or completion of a specific task.
Between July 2024 and July 2025, ‘Asking’ and ‘Expressing’ was reported to have grown faster than ‘Doing’, suggesting that users increasingly treat ChatGPT as a decision-support tool. For work-related queries, however, Doing dominates, making up 56% of all professional queries.
Looking Ahead
ChatGPT is not only shaping professional tasks but also becoming deeply embedded in personal routines. The rapid expansion of non-work use points to significant welfare gains, with economists estimating that generative artificial intelligence provides at least $97 billion in consumer surplus each year.
In the workplace, ChatGPT’s role is clearest as an advisor and research assistant, helping professionals process information, refine communication, and support decision-making.
As adoption gaps continue to close, artificial intelligence models like ChatGPT are set to become a fixture in everyday life, changing both workplace norms and consumer expectations of information access.
Reference: National Bureau of Economic Research
