How People Are Using ChatGPT (PDF) Broken Down
The full white paper PDF to read is available here.
Usage Statistics and Insights
Weekly Active Users: 700M (July 2025), sending 18B messages per week.
Work vs. Non-Work: Work-related use declined from 47% (2024) → 27% (2025); non-work use rose from 53% → 73% in the same period.
Top Conversation Topics:
Practical Guidance: ~29%
Seeking Information: Grew from 14% → 24% (2024-2025)
Writing: Fell from 36% to 24% (2024-2025)
but still 40% of work-related use
Technical Help: Declined from 12% → 5%
Multimedia: Jumped from 2% → 7% after new features
User Intent (as of June 2025):
Asking (information, advice): 51.6%
Doing (output generation): 34.6%
Expressing (sentiments, chitchat): 13.8%
Demographics:
Gender: No longer male-dominated; likely even or slightly female-skewed
Age: ~50% of messages from under-26s; closing gap with older adults
Income: Faster growth in lower- and middle-income countries
Employment/Education Patterns:
Work usage is prominent among highly-educated, professional, and white-collar users
• Most work-related queries focus on information processing and writing tasks, mirroring modern knowledge work roles
Top 25 Takeaways
ChatGPT reached 700 million active weekly users (~10% of the world’s adults) by July 2025, with 18 billion messages sent weekly.
Early adoption skewed male, but the gender gap has essentially closed as of 2025.
Usage has increased more rapidly in lower- and middle-income countries over the past year.
Nearly half of all user messages come from adults under age 26, though age differences are declining.
Only 27% of messages in June 2025 were work-related, down from 47% a year earlier.
Non-work messages grew faster, increasing from 53% (mid-2024) to 73% (mid-2025).
Three top topics—Practical Guidance, Seeking Information, and Writing—comprise nearly 80% of conversations.
“Writing” (emails, editing, summaries, translation) is the dominant work-related use, making up 40% of work messages.
Around 2/3s of “Writing” requests ask ChatGPT to modify user-supplied text instead of creating new content.
Education (tutoring or teaching) is a significant use case: 10% of messages and 36% of “Practical Guidance” queries.
Requests for computer programming and self-expression remain a small fraction (4.2% and 2.4%, respectively).
Technical help (programming, math, data analysis) has declined as a share of total usage.
Multimedia capabilities (e.g., image generation) have seen dramatic spikes post-feature launches.
Most users ask for customised or personalised advice and plans, demonstrating flexibility beyond traditional search.
User intent falls into “Asking” (49%), “Doing” (40%), and “Expressing” (11%); work-related queries skew toward “Doing.”
The frequency of “Asking” (info-seeking, advice) is rising fastest, now constituting over half of all messages.
Quality ratings for “Asking” messages are higher (via automated and direct feedback) than for other intents.
The most frequent work activities include obtaining/interpreting information, providing decision support, offering advice, and engaging in creative problem-solving.
These knowledge-based activities (above) account for over 80% of work use-cases.
White-collar and highly-paid professionals use ChatGPT at work much more than other groups.
Collective consumer surplus in the US from ChatGPT was estimated at $97 billion in 2024.
Message classification and analysis used privacy-preserving methods—no researchers viewed raw user content.
Data on employment and education were aggregated, never directly linked to users, and subject to privacy controls.
Across jobs, information-seeking and writing-related use-cases dominate, regardless of occupation.
Core differentiator from web search: ChatGPT’s unique ability to generate, alter, and personalise digital output on demand.
