OpenAI has released its largest study to date of how people are using its hallmark large language model AI chatbot, ChatGPT, offering an in-depth view of what is driving usership of the phenom.

ChatGPT came out in 2022 and marked the seminal rise of AI in the public eye. No longer confined to the machine learning models of coding pros, AI hit the mainstream with the chatbot’s riotous release, which promised (or threatened) to change society across the globe.

The study resulted in a National Bureau of Economic Research (NBER) working paper by OpenAI’s Economic Research team and Harvard economic David Deming, and draws from 1.5 million user conversations, tracking the evolution of user engagement since ChatGPT’s release.

OpenAI did stress that in an effort to preserve privacy, researchers did not read user messages, and automated tools were instead used to categorise usage patterns.

Demographic Shifts

The study found significant demographic shifts in usership of ChatGPT. Back in January 2024, users with ‘feminine names’ made up just 37% of usership. By July 2025, this rose to 52%, a more fair reflection of the general adult population than the launch of the chatbot.

ChatGPT is also more broadly accessible across the globe as well, which is reflected in the study. By May 2025, OpenAI found that the lowest income countries used ChatGPT 4x more than the highest income countries.

Recommended reading

• ChatGPT Search Inches Toward Strict EU Oversight as Users Surge
• ChatGPT Introduces Parental Safety Controls – Too Little Too Late?
• Is ChatGPT Making You Dumb? Possibly, MIT Says

Using ChatGPT

About 30% of consumer usage is related to work, the study found, while 70% is related to personal use.

The study found that about three quarter of user conversations revolve around practical everyday tasks, seeking information, and writing.

In work, writing was the most common task asked by users, while coding and ‘self-expression’ were generally more niche activities.

When split into categories of ‘asking’, ‘doing’, and ‘expressing’, OpenAI found that just about half (49%) of all messages could be classified as ‘asking’. Queries classed as ‘doing’ made up about 40% of queries, and about a third of work queries.

This means more people are looking to ChatGPT for information or as an advisor rather than a task completer, though many still seek out the chatbot directly for tasks.

‘Expressing’ only made up about 11% of usage, which reflects personal exploration and even play.