Writing for women, technical assistance for men: to varying degrees, some uses of ChatGPT appear gendered.
OpenAI doesn’t present things quite like that. But it explicitly draws correlations. For example, the fact that the majority of messages related to writing or editing text come from users with “typically female” given names. And the reverse for technical assistance.
These elements stem from analyzing conversations linked to 1.1 million accounts. For each, the user’s sex was inferred using public data (World Gender Name Dictionary, lists of popular names from the U.S. Social Security Administration, datasets of Brazilian and Latin American names).
Excluding names not present in these data or flagged as ambiguous, the original ChatGPT audience was decidedly male (about 80% of weekly active users). The gap gradually narrowed, until parity was reached in 2025.
A dataset of roughly 1 million conversations
These conclusions sit within a broader study that can be boiled down to one question: who uses ChatGPT and how?
To answer it, OpenAI focused on the public-facing versions of its chatbot (Free, Pro, and Plus). It limited itself to users who had not objected to having their data used to train models. It also excluded, among others, those who stated they were under 18 and those who used the service without logging in.
1.1 million messages drawn from as many conversations conducted between May 2024 and June 2025 were randomly selected. OpenAI added two sub-samples encompassing about 130,000 users and containing messages sampled respectively at the conversation level and the user level.
The whole set was processed without human intervention, using several classification models.
Declining professional use of the consumer versions
More striking than the gender difference is the age split. Nearly half of the messages (46%) came from users who reported being between 18 and 25 years old.
If we focus on messages classified as likely related to professional activity, nearly a quarter (23%) still pertain to the 18–25 bracket. Behind this, a growing personal use of ChatGPT: the share of non-work-related messages rose from 53% in June 2024 to 73% a year later. The trend is similar when grouping users by tenure (semi-annual intervals).
More research, less content creation
OpenAI defined 24 conversation categories and grouped them into seven broad themes: writing, practical advice, technical assistance, multimedia (creation/editing of images), information gathering, self-expression… and other/unknown (including, for instance, asking questions about the model).
The “practical advice” theme is the most common (28.8%), ahead of information retrieval (24.4%) and writing (23.9%). All others appear in fewer than one message in ten.
Over a year, the topic of “writing” declined (-12 points, to 24%). As did “technical assistance” (-7 points, to 5%)—OpenAI suggests that this usage shifted toward its API. In parallel, the “information retrieval” topic rose (+10 points, to 24%). Likewise for multimedia (up by 5 points, to 7%), with a peak in April 2025 when native image generation with GPT-4o was integrated.
A role as an assistant that asserts itself
Regarding professional uses, writing dominates (between 40 and 50% depending on the month within the period considered), ahead of practical advice (20–25%), information retrieval (11–14%; upward trend) and technical assistance (10–18%; downward trend).
Overall, two-thirds of “writing” messages do not involve creation but modification of content (editing or critiquing a text, translating, generating a summary).
In four out of ten cases, ChatGPT is asked to perform tasks. In five out of ten, to play an assistive role. This latter use overtook between June 2024 and June 2025. It appears in most professional conversations.
“Creative thinking” no more used in pro than in personal use
OpenAI linked messages to professional activities using the American O*NET (Occupational Information Network) taxonomy. This system classifies jobs according to the skills and tasks required to perform them. Tasks are aggregated across three levels: 2,087 detailed activities, 332 intermediate activities, and 41 generalized activities.
Within the generalized scope, nearly half of the messages (45.2%) concern three activities: information retrieval (19.3%), interpreting the meaning of information for others (13.1%), and documenting/recording information (12.8%). Followed by providing advice (9.2%) and creative thinking (9.1%).
The hierarchy differs if only professional messages are considered. The most represented activity then is “documenting/recording information” (13.2%). Next come “making decisions and solving problems” (10.6%), creative thinking (9.3%), “working with computers” (7.7%), “interpreting the meaning of information for others” (7.3%), etc.
Technical assistance and image editing, the least satisfying uses
Relatively to each message, OpenAI looked for signs of user satisfaction or dissatisfaction as the conversation progressed.
Over a year, overall satisfaction rose (from 48% to 57.8%). Writing, technical assistance and multimedia are the three themes that generate the most frustration.
The more educated you are, the more likely you are to use ChatGPT for work. Among users with a bachelor’s degree, the share of messages with a professional character reaches 46%. By contrast, 37% among those with lower education, and 48% among those with higher degrees.
The difference is less clear regarding the conversation themes. Except for writing, which is more regularly found among more educated populations.
The likelihood that messages pertain to professional work is higher among those employed in information technology or in management/business. If the former understandably favor the “technical assistance” theme and the latter the “writing” theme, information retrieval and practical advice are rather evenly distributed across the five professional categories OpenAI distinguishes.
