What the latest NBER data reveals about How People Use ChatGPT, the gen AI technology that changed how humans think, write, and decide.
When ChatGPT launched in November 2022, few imagined that within three years, it would become humanity’s most widely adopted technology.
By July 2025, according to a new NBER Working Paper (“How People Use ChatGPT,” Chatterji et al., 2025), the platform had reached ~700 million weekly active users, about one in ten adults on Earth, sending 18 billion messages every week. That’s roughly 29,000 messages per second. Mind blowing facts.
The study offers one of the clearest pictures yet of how people use ChatGPT in real life- what they ask, what they create, and how their behavior is changing as the technology matures.
It draws on 1.1 million de-identified messages collected between May 2024 and June 2025, analyzed by GPT-5-based classifiers inside a secure, privacy-preserving environment. The largest study ever done on generative AI.
As the authors put it, this growth “has no precedent in modern technology diffusion.” And the story those messages tell? It’s less about automation and more about how humans are learning to think with AI.
🚀 The fastest diffusion in digital history
ChatGPT’s expansion has been steeper in low- and middle-income countries than in high-income ones, a reversal of the usual tech-adoption curve.
Early users were ~80 % male, but by mid-2025, women slightly outnumber men. Nearly half of all messages come from people aged 18–25 – digital natives now using large-language models as everyday thought partners.
InvuAI’s take: adoption followed not just access, but emotional utility. When AI starts answering your personal “how do I…” questions better than a search engine, it stops feeling like a tool and starts feeling like a habit.
💼 Work vs. life: the personal era of AI
The researchers tracked how people use ChatGPT across work and non-work contexts, revealing that between June 2024 and June 2025, total daily messages jumped from 451 million to 2.63 billion. Over that same year, non-work use grew from 53 % to 73 %.
Work use remains strongest among college-educated and high-income professionals, but the paper’s authors highlight a shift: the fastest growth now comes from personal guidance like tutoring, creative ideation, and decision support.
Economists Collis & Brynjolfsson estimate this created $97 billion in U.S. consumer surplus in 2024 alone.
Humans aren’t just delegating work, they’re outsourcing uncertainty.
💬 What we talk about with ChatGPT
Roughly 80 % of all messages fit three themes:
| Category | Share | Description |
|---|---|---|
| Practical Guidance | ~29 % | How-to advice, tutoring, creative ideation |
| Seeking Information | ~24 % | Current events, product questions, factual checks |
| Writing | ~24 % | Drafting, editing, translating, roughly 40 % of all work messages |
Technical help, once 12 % of all messages, is now just ~5 %. Education and tutoring hold ~10 %, while multimedia/image generation rose from 2 % → 7 % after April 2025 model updates. Self-expression like reflection, relationships, or role-play remains small (~2.4 %) but carries the highest satisfaction.
These shifts also highlight how people use ChatGPT not just to produce content but to structure thought but moving from writing to reasoning.
🎯 Asking, doing, expressing: the new hierarchy of intent
The researchers classify messages as:
- Asking (49 %),
- Doing (40 %), and
- Expressing (11 %).
By mid-2025, Asking had grown to 51.6 %, while Doing fell to 34.6 %.
At work, Doing still leads (56 %), and three-quarters of those Doing messages involve Writing tasks like reports, emails, documentation. But Asking messages earn the highest satisfaction.
We’re moving from using AI as a worker to using it as a thinking partner.
🧠 When AI enters the job description
Mapped against O*NET’s Generalized Work Activities, seven activities explain ~77 % of all work-related ChatGPT messages:
- Getting Information (19 %)
- Interpreting Meaning (13 %)
- Documenting or Recording (13 %)
- Providing Advice (9 %)
- Thinking Creatively (9 %)
- Making Decisions / Solving Problems (9 %)
- Working with Computers (5 %)
Together, they define what ChatGPT really is: a decision-support and information-handling system, not a coder or content factory.
👩💻 Who’s using it; and how
Work usage by field:
| Occupation | Work usage share |
|---|---|
| Computer | 57 % |
| Business | 50 % |
| Engineering | 48 % |
| Other professional | 44 % |
| Non-professional | 40 % |
In business and management, Writing makes up ~52 % of all work messages.
In computer roles, Technical Help = 37 % of work queries.
Across nearly every job type, “Making Decisions & Solving Problems” ranks in the top two work activities.
Gender parity has arrived as female users now slightly lead overall, with women writing more and men asking for more technical or factual help.
😊 Quality and satisfaction
By July 2025, “good” interactions were four times more common than “bad.” People rate Asking and Self-Expression topics highest; Technical Help and Multimedia lowest (perhaps that’s the reason why Technical help, once 12 % of all messages, is now just ~5 %, but also increase of other type of prompts changed the mix).
Quality improved alongside model maturity, classifiers used GPT-5-mini, while GPT-5 evaluated interaction quality.
💡 What this all means
ChatGPT has evolved into a mainstream cognitive companion, a place where decisions start, not just where text gets finished.
Its biggest economic value isn’t the cost it saves; it’s the clarity it gives.
InvuAI calls that cognitive leverage: when humans amplify their own judgment through conversation, not delegation.
“Writing dominates work-related tasks,” the authors note, but the true value is in “decision support… especially in knowledge-intensive jobs.”
InvuAI’s closing thought
A world typing 29,000 messages every second into ChatGPT isn’t just producing text, it’s teaching itself new ways to respond.
But we’re still only beginning to understand how people use ChatGPT … and how, in turn, it’s reshaping how people think. The way people use generative AI is evolving month by month, most recently from drafting and searching, toward reasoning, reflecting, and co-creating.
As Asking overtakes Doing, AI’s role is shifting from writer to co-reasoner, from tool to collaborator.
The next frontier isn’t just AI that writes for us, it’s seems AI that thinks with us, learns from us, and even helps us understand ourselves a little better every day.
Sources & further reading
- NBER Working Paper No. 34255 (Sept 2025) — Chatterji, Cunningham, Deming, Hitzig, Ong, Shan, Wadman: How People Use ChatGPT. (All figures and quotes above are from this paper.)
- OpenAI model/system docs are referenced in the paper for technical background on GPT-4/4o/o-series/5; see the paper’s Section 2 and Appendix timeline for launch milestones.
