OpenAI has offered one of the clearest looks yet at what people actually do with ChatGPT. A first-of-its-kind National Bureau of Economic Research working paper, released by OpenAI’s Economic Research Team in association with Harvard economist David Denning, uses direct access to OpenAI’s internal user data rather than relying only on self-reported surveys.
The result is a snapshot of ChatGPT usage at enormous scale: consumer users, billions of daily messages, and a mix of personal, work-related, writing, and information-seeking tasks. It does not answer every question about large language models, but it does make one thing clear: ChatGPT is no longer just an experiment for early adopters.
ChatGPT has become a mass-market habit
The study focuses on ChatGPT’s consumer plans, including Free, Plus, and Pro tiers. Within that slice, the scale is striking. ChatGPT passed 100 million weekly active users in early 2024, moved past 400 million users early this year, and now has over 700 million users.
OpenAI described that current reach as “nearly 10% of the world’s adult population.” The company also notes that the count may be somewhat inflated because some logged-out users can be counted across multiple devices, and some logged-in users may keep multiple accounts under different email addresses.
Even with that caveat, the growth in message volume shows how quickly the product has expanded. OpenAI processed about 451 million messages per day in June 2024. By June 2025, that had risen to over 2.6 billion messages per day, averaged over a week near the end of the month.
For context, the source article notes that Google announced in March that it averages 14 billion searches per day. That comparison does not make ChatGPT equivalent to search, but it does show that conversational AI is now operating at a scale that belongs in the same broad discussion as the largest everyday internet services.
New users are driving much of the growth
The headline growth numbers tell only part of the story. OpenAI’s paper also breaks usage down by when logged-in users first created their accounts, which helps separate growth from new sign-ups and growth from existing users sending more messages.
For long-term users, average daily message volume rose sharply during two periods. The first ran roughly from September through December 2024 and coincided with the launch of the o1-preview and o1-mini models. The second came after the launch of the o3 and o4-mini models, when average usage increased again through June.
Since June, however, per-user message rates for established ChatGPT users have been mostly flat for three full months. In the source’s framing, established users are those who signed up in the first quarter of 2025 or before.
That matters because it suggests overall ChatGPT usage has recently been expanding through new user acquisition, not because older users are steadily increasing how much they use the product every day. The recent tumultuous launch of the GPT-5 model may change that pattern, but the source article treats that as an open question rather than a conclusion.
The user base is young, and the gender mix has shifted
The study also offers a look at who is using ChatGPT, though the source article is careful about how those figures were estimated. Among users who revealed their age in OpenAI’s study sample, 46 percent were between the ages of 18 and 25.
That makes younger adults a major part of ChatGPT’s visible user base. The source article also notes that people under 18 were not included in the sample at all, even though some are likely using the service.
OpenAI also estimated likely gender distribution for a large sample of users. To do that, it used Social Security data and the World Gender Name Registry’s list of strongly masculine or feminine first names.
According to that analysis, when ChatGPT launched in late 2022, roughly 80 percent of weekly active ChatGPT users were likely male. By late 2025, the balance had changed to a slight majority of likely female users, at 52.4 percent.
Most usage is not classified as work
One of the more important findings is that ChatGPT is not being used mainly as a business productivity tool, at least within the consumer-plan data set studied. OpenAI classified messages using an LLM-based classifier, and non-work tasks accounted for about 53 percent of all ChatGPT messages in June of 2024.
By June 2025, that share had grown to 72.2 percent. That is a major shift toward personal use inside the consumer product.
The source article points out an important limitation: Business, Enterprise, and Education subscription tiers were excluded from the data set. That means the study should not be read as a complete picture of all workplace use of ChatGPT. Still, within the consumer plans, the trend is clear: many newer users appear to be turning to ChatGPT for personal reasons rather than strictly for productivity.
Writing and information are the biggest clues
Writing remains one of the central ways people use ChatGPT. Across 1.1 million conversations from May 2024 to June 2025, OpenAI said 28 percent involved writing assistance in some form.
For work-related conversations, writing was even more prominent. It accounted for 42 percent of work-related conversations, making it the most popular work-related task in the sample. Among users with “management and business occupations,” it reached 52 percent of work-related conversations.
But the study also complicates the common idea that people mainly use ChatGPT to generate finished text from scratch. OpenAI found that 10.6 percent of all conversations involved asking the model to “edit or critique” text, compared with 8 percent for generating “personal writing or communication” from a prompt. Another 4.5 percent involved translating existing text to a new language, while 1.4 percent involved “writing fiction.”
Information seeking is rising too. In June 2024, about 14 percent of ChatGPT conversations were tagged as “seeking information.” By June 2025, that share had climbed to 24.4 percent, slightly ahead of writing-based prompts in the sample.
That rise helps explain why ChatGPT usage increasingly overlaps with habits once associated mainly with search engines. At the same time, the source article notes that even if recent GPT models seem better at citing relevant sources, OpenAI is no closer to solving the widespread confabulation problem. For users, that means the service may be increasingly useful as an information tool, but the trust question remains central.