The Hidden Story of ChatGPT: What 18 Billion Messages Reveal from OpenAI’s New Paper on User Behaviors
by Ara Zhang, September 22, 2025
OpenAI researchers, along with collaborators from Duke and Harvard, just published a deep dive into how hundreds of millions of people actually use ChatGPT. Beyond the obvious headlines about adoption, the paper reveals some surprising, lesser-discussed patterns that reshape our understanding of generative AI in daily life.
Full Paper: https://cdn.openai.com/pdf/a253471f-8260-40c6-a2cc-aa93fe9f142e/economic-research-chatgpt-usage-paper.pdf
Here are seven uncommon insights worth highlighting:
1. ChatGPT is increasingly about life outside of work
By mid-2025, over 70% of all ChatGPT messages were non-work related, up from 53% a year earlier.
“Non-work messages have grown faster and now represent more than 70% of all consumer ChatGPT messages” (p. 2).
This suggests that the biggest welfare gains from AI may come not from productivity at work, but from decision support, home tasks, and personal projects.
2. Writing is mostly about fixing, not creating
We often imagine people asking ChatGPT to generate essays or reports from scratch. But the data show the opposite:
“About two-thirds of all Writing messages ask ChatGPT to modify user text (editing, critiquing, translating, etc.) rather than creating new text from scratch” (p. 14).
That makes ChatGPT closer to a universal editor than a ghostwriter.
3. Coding isn’t as big as you think
Despite the hype around AI coding assistants, only 4.2% of ChatGPT messages were related to programming. That’s an order of magnitude smaller than for Anthropic’s Claude, where coding dominates usage.
“We find the share of messages related to computer coding is relatively small: only 4.2% of ChatGPT messages… compared to 33% of work-related Claude conversations” (p. 2).
This reflects ChatGPT’s broader audience: it’s less a dev tool and more an everyday companion.
4. The gender gap flipped
Early adopters were overwhelmingly male. But by June 2025:
“That number declined to 48% as of June 2025, with active users slightly more likely to have typically feminine first names” (p. 3).
Few tech products have shifted from 80% male → slight female majority in under three years.
5. Young people dominate, but not just for play
Nearly half of all messages come from users under 26. Still, older users skew more work-oriented:
“Work-related messages comprised approximately 23% of messages for users under age 26… with this share increasing with age” (p. 25).
So while youth drive overall volume, older professionals lean more on ChatGPT for workplace tasks.
6. Asking is beating Doing
The authors introduce a novel taxonomy: Asking (information/advice), Doing (task output), Expressing (sharing views). The surprise?
“Asking and Expressing grew much faster than Doing over the next year, and by late June 2025 the split was 51.6% Asking, 34.6% Doing, and 13.8% Expressing” (p. 17).
This means ChatGPT is becoming less of a task executor and more of a decision support system.
7. Usage patterns are the same across wildly different jobs
No matter the occupation—business manager, scientist, or sales rep—the same work activities dominate:
“Getting Information and Making Decisions and Solving Problems are in the top five of message frequency in nearly all occupations” (p. 20).
This suggests that ChatGPT is flattening knowledge work, converging on a common set of information-seeking and problem-solving tasks across fields.
Why this matters
Most analyses of AI focus on its impact on workplace productivity. But this study shows ChatGPT is just as transformative in non-work settings, especially in how people seek information, make decisions, and edit their words.
As the authors note:
“ChatGPT likely improves worker output by providing decision support, which is especially important in knowledge-intensive jobs where better decision-making increases productivity” (p. 3).
The real story is not just automation—it’s AI as an advisor, editor, and guide, across work and life.
Related Articles
- What is Fine-tuning in AI?
- AI Agent vs AI Chatbot - What’s the Difference, and Why We Don’t Talk About Chatbots Anymore
- What is RAG in AI?
