Large language models largely remain black boxes in terms of what is happening inside them to produce the outputs that they do. They have also been a bit of a black box in terms of who is using them and what they are doing with them. OpenAI, with some help from the National Bureau of Economic Research (NBER), set out to figure out what exactly its growing user base is getting up to with its chatbot. It found a surprising amount of personal use and a closing “gender gap” among its frequent users.
In an NBER working paper authored by the OpenAI Economic Research team and Harvard economist David Deming, the researchers found that about 80% of all ChatGPT usage falls under one of three categories: “Practical Guidance,” “Seeking Information,” and “Writing.” “Practical guidance,” which the study found to be the most common usage, includes things like “tutoring and teaching, how-to advice about a variety of topics, and creative ideation,” whereas “seeking information” is viewed as a substitute for traditional search. “Writing” included the automated creation of emails, documents, and other communications, as well as editing and translating text.
Writing was also the most common work-related use case, per the study, accounting for 40% of work-related messages in June 2025, compared to just 4.2% of messages related to computer programming—so it seems coding with ChatGPT is not that common.
Notably, work usage for ChatGPT appears to make up a shrinking share of how people are interacting with the chatbot. In June 2024, about 47% of interactions users had with the chatbot were work-related. That has shrunk to just 27%, which comes as other research shows companies largely failing to figure out how to generate any sort of meaningful return from their AI investments. Meanwhile, non-work-related interactions have jumped from 53% to 73%.
While users are apparently spending more time with ChatGPT in their personal time, OpenAI’s research found that a “fairly small” share of messages with the chatbot were users seeking virtual companionship or talking about social-emotional issues. The company claimed that about 2% of all messages were people using ChatGPT as a therapist or friend, and just 0.4% of people talked to the chatbot about relationships and personal reflections—though it’d be interesting to see if users who engage with a chatbot this way generate more messages and if there is stickier engagement.
For what it’s worth, other researchers seem to believe that this usage is far more common than those numbers might suggest. Common Sense Media, for instance, found that about one in three teens use AI chatbots for social interaction and relationships. Another study found that about half of all adult users have used a chatbot for “psychological support” in the last year. The teen figure is particularly of note, considering OpenAI’s research did find its userbase skews young. The NEBR study found 46% of the messages came from users identified as being between the ages of 18 and 25 (it also excluded users under the age of 18). Those users are also more likely to use ChatGPT for personal use, as work-related messages increase with age.
The study also found that there is a growing number of women using ChatGPT, which initially had a very male-dominated user base. The company claims that the number of “masculine first name” users has declined from about 80% in 2022 to 48% in June 2025, with “typically feminine names” growing to reach parity.
One caveat about the study that may give you pause, depending on how much you trust technology: OpenAI used AI to categorize all of the messages it analyzed. So if you’re skeptical, there’s an asterisk you can put next to the figures.