The largest research study of 700 million users has revealed essential patterns of ChatGPT usage.
The research study NBER 34255 provides an in-depth examination of how people use ChatGPT. 700 million users send 18 billion messages weekly according to this study which establishes itself as the most comprehensive research on AI adoption.
What you'll learn in this comprehensive analysis:
- • How 700 million people actually use ChatGPT
- • The difference between "Asking" and "Doing" - and why it matters
- • Why 73% of usage is non-work related
- • The surprising dominance of Practical Guidance over Programming
- • Demographic trends and global adoption patterns
Table of Contents
- 1. Study Overview: The World's Largest AI Usage Study
- 2. Privacy-First Methodology
- 3. The Scale of the Phenomenon: 700 Million Users
- 4. Work vs. Personal: The Great Shift
- 5. Anatomy of a Conversation: What ChatGPT Is Used For
- 6. The Asking vs. Doing Framework
- 7. Global User Data and Demographic Trends
- 8. Business Implications: ChatGPT as Decision Support Engine
In September 2025, researchers from Harvard, Duke, and OpenAI published what is arguably the most important study on practical Generative AI usage: NBER Working Paper 34255 "How People Use ChatGPT". This groundbreaking study based on internal OpenAI data provides the first scientifically rigorous insights into the actual user behavior of 700 million ChatGPT users.
The results correct many widespread assumptions about AI usage and show: ChatGPT is not primarily an automation tool, but a sophisticated decision support system. This insight has fundamental implications for how businesses develop AI strategies and how we assess the economic impact of Generative AI.
Study Overview: The World's Largest AI Usage Study
Unprecedented Data Scale
The study analyzes data from 700 million weekly active users and over 18 billion messages per week - equivalent to approximately 29,000 messages per second.
- • Nearly 10% of the world's adult population uses ChatGPT weekly
- • Data collection over 13 months (June 2024 - July 2025)
- • Analysis of over 1.1 million conversation samples
Scientific Rigor
The researchers developed innovative methods to analyze sensitive data without ever seeing the content of user messages - setting a new standard for responsible AI research.
- • Complete anonymization of all user data
- • Automated LLM-based classification systems
- • Secure Data Clean Room (DCR) environment for demographic analysis
Groundbreaking Insights
For the first time, we can demonstrate with hard data how Generative AI is actually used - and the reality differs significantly from expectations.
- • 73% of usage is non-work related
- • ChatGPT is primarily used for decision support, not automation
- • Programming accounts for only 4.2% of total usage
Key Numbers at a Glance
Privacy-First Methodology
A New Standard for Responsible AI Research
The study sets a new benchmark for privacy protection in large-scale data analysis. The research team developed innovative methods to gain insights into user behavior without ever seeing the content of messages.
Privacy Principles of the Study
Zero Human Access Principle:
- • No researcher ever saw message content
- • Complete automation of analysis
- • Use of other LLMs for classification
- • Contextual analysis with 10 previous messages
Secure Data Linking:
- • Data Clean Room for demographic data
- • Minimum group size of 100 users
- • Aggregated queries without export capability
- • External validation via WildChat dataset
The Scale of the Phenomenon: 700 Million Users
Historically Unprecedented Adoption
The numbers are breathtaking: ChatGPT reached over 350 million logged-in weekly active users just two years after launch and grew to 700 million by July 2025. This adoption speed surpasses all previous technologies.
| Time Period | Weekly Active Users | Daily Messages | Growth Rate |
|---|---|---|---|
| Nov 2022 (Launch) | ~100M | ~500M | - |
| July 2024 | 350M+ | ~500M | +250% |
| July 2025 | 700M | 2.5B | +100% |
Deepening Usage Across All Cohorts
Particularly noteworthy: Growth comes not just from new users, but also from more intensive use by existing users. The earliest users (Q4 2022) are using ChatGPT more than ever today.
Model Improvements
The underlying LLMs (GPT-4, GPT-4o, GPT-5) have become more capable, providing more utility.
User Discovery
Users progressively discover new and valuable applications for the technology, integrating it deeper into their lives.
Work vs. Personal: The Great Shift
The 70% Event: Dominance of Personal Usage
One of the study's most surprising findings: While early AI narratives focused on workplace transformation, the data shows that 73% of ChatGPT usage is non-work related. This shift has massive implications for our understanding of AI value creation.
| Time Period | Non-work Related | Work Related | Daily Message Volume |
|---|---|---|---|
| June 2024 | 53% (238M) | 47% (213M) | 451M |
| June 2025 | 73% (1.91B) | 27% (720M) | 2.63B |
Economic Implications of Personal Usage
This massive personal usage points to an enormous, often overlooked source of economic value - consumer surplus and household production worth billions.
Anatomy of a Conversation: What ChatGPT Is Used For
The "Big Three": Practical Guidance, Information, and Writing
77% of all ChatGPT conversations fall into three main categories: Practical Guidance (29%), Seeking Information (24%), and Writing (24%). This concentration shows that ChatGPT primarily functions as a knowledge mediator and advisory tool.
Practical Guidance (29%)
The largest and most stable application area. ChatGPT functions as a personalized advisor for customized, adaptive advice on an enormous range of topics.
- • Tutoring/Teaching: 10.2% of all messages
- • How-To Advice: 8.5% of all messages
- • Personal advice on life decisions
- • Customized solution approaches
Seeking Information (24%)
Rapidly growing area (from 14% to 24% over the year). ChatGPT positions itself as a direct alternative to traditional search engines - but with conversational interaction.
- • Direct competitor to Google/Bing
- • Growth from 14% to 24% in one year
- • Contextual, personalized responses
- • Follow-up questions possible
Writing (24%)
While the share has decreased from 36% to 24%, writing remains a core function. Critical: Two-thirds of all writing tasks involve editing user-provided text, not creation from scratch.
- • 67% of writing tasks are text editing
- • Editing/critique of provided text
- • Translation and summarization
- • ChatGPT as co-pilot, not ghostwriter
Correcting Widespread Narratives
The data corrects several widespread assumptions about ChatGPT usage:
Myth: Programming Dominates
Reality: Only 4.2% of all messages concern programming. Serious developers likely prefer dedicated APIs and tools like Copilot.
Myth: Therapy/Companionship is Common
Reality: Only 1.9% for Relationships/Personal Reflection and 0.4% for Games/Role Play. Therapeutic use is a niche.
The Asking vs. Doing Framework
The Critical Distinction: Decision Support vs. Task Execution
The study's most important conceptual innovation is the "Asking vs. Doing" framework. This distinction shows that ChatGPT primarily functions as a cognitive co-pilot, not as an automation tool.
"Asking" - 49% of All Interactions
Users seek information or advice to be better informed or make better decisions. This is decision support in its purest form.
- • Example: "What should I look for when choosing a health plan?"
- • Growing faster than "Doing" interactions
- • Higher quality ratings from users
- • Particularly pronounced among highly educated professionals
"Doing" - 40% of All Interactions
Users ask ChatGPT to perform tasks - rewrite emails, write code, create content. This is direct production support.
- • Example: "Rewrite this email to make it more formal"
- • Dominates in work tasks (56% vs. 35% "Asking")
- • Heavily concentrated in writing tasks
- • Output can be directly fed into production processes
"Expressing" - 11% of All Interactions
Statements that neither ask for information nor request task execution. These have little to no economic content.
- • Self-expression and reflection
- • Low economic value
- • Mainly in personal conversations
Temporal Evolution: The Rise of "Asking"
Over the past year, the share of "Asking" messages has grown significantly relative to "Doing". This suggests users are finding increasingly sophisticated applications for the technology.
Why "Asking" Receives Higher Ratings:
- Personalization: Responses are tailored to specific user circumstances
- Interactivity: Follow-up questions enable deeper exploration
- Contextual Value: Solutions consider individual constraints and goals
Global User Data and Demographic Trends
The Democratization of AI Usage
ChatGPT is undergoing a remarkable demographic transformation. What started as a tool for tech-savvy, male early adopters has become a globally diversified platform.
The End of the Gender Gap
The originally heavily male user base (~80% masculine first names in late 2022) has completely rebalanced. By June 2025, users with typically feminine first names achieved a slight majority.
- • Late 2022: ~80% masculine first names
- • June 2025: 48% masculine, 52% feminine first names
- • Complete rebalancing in just 2.5 years
- • Shows broad societal acceptance
Global Spread in Developing Countries
The most dramatic growth is occurring in low- to middle-income countries (GDP per capita $10,000-$40,000), indicating rapid, global democratization of advanced AI technology.
- • Strongest growth in emerging markets
- • Bridging the digital divide
- • Equal access to AI technology
- • Potential for leapfrog innovation
Education Drives Work Usage
Highly educated professionals are the most intensive work users. Their activity is disproportionately classified as "Asking," confirming ChatGPT's value in augmenting high-level cognitive tasks.
- • Higher education = more work-related usage
- • Professional occupations: 50%+ work usage
- • Computer Science: 57% work-related
- • Management: 50% work-related
Generational Usage Patterns
| Age Group | Share of Messages | Work-Related Usage | Characteristic |
|---|---|---|---|
| 18-25 | 46% | 23% | Power users - volume drivers |
| 26-35 | 28% | 35% | Balanced mixed usage |
| 36-45 | 18% | 42% | Professional integration |
| 46+ | 8% | 45% | Highest work integration |
Business Implications: ChatGPT as Decision Support Engine
The Transformative Thesis: From Automation to Augmentation
The NBER study provides a clear, data-driven answer to the question of ChatGPT's economic value: It is not primarily an automation tool, but a sophisticated decision support system. This insight fundamentally changes how businesses should develop AI strategies.
Cognitive Co-Pilot, Not Replacement
The dominance of "Asking" interactions (49%) and their higher quality ratings show: ChatGPT amplifies human intelligence rather than replacing it.
- • Augmentation > Automation
- • Higher user satisfaction with advisory functions
- • Particularly valuable for knowledge work
- • Supports complex decision processes
O*NET Analysis: Universal Applicability
The connection to O*NET work activities shows: 81% of all work-related messages fall into just seven categories focused on information processing and cognitive support.
- • Documenting and Recording Information: 18.4%
- • Making Decisions and Solving Problems: 14.9%
- • Thinking Creatively: 13.0%
- • Remarkable similarity across occupations
Strategic Business Implications
Companies should view ChatGPT not as a cost-cutting tool, but as a productivity and quality enhancer for knowledge work.
- • Investment in employee training for AI augmentation
- • Focus on decision support rather than automation
- • Integration into strategic planning processes
- • Measurable improvement in decision quality
The Future of Work: Three Core Principles
1. Humans + AI > Humans or AI Alone
The highest value creation comes from combining human creativity and AI support, not through replacement.
2. Advisory Beats Automation
The most valuable AI applications support complex decisions rather than automating simple tasks.
3. Universal Applicability Requires Tailored Implementation
While ChatGPT is used similarly across all professions, integration must be specifically tailored to industry and role.
Conclusion: A New Era of AI Adoption
NBER Working Paper 34255 marks a turning point in our understanding of Generative AI. With scientific precision, the study shows that ChatGPT has become an indispensable tool for 700 million people - not as a replacement for human intelligence, but as its amplifier.
The Five Transformative Insights
- 1. ChatGPT is primarily a personal tool (73% non-work related)
- 2. Decision support dominates over automation (49% "Asking")
- 3. Practical Guidance is the top use case (29%)
- 4. Programming is a niche, not mainstream (4.2%)
- 5. AI democratization is happening globally and gender-neutrally
For businesses, this means a fundamental realignment: ChatGPT should not be viewed as a cost-cutting tool, but as an investment in the quality of human decisions. The future belongs to organizations that deploy AI as a cognitive co-pilot, not as a replacement for human expertise.
The NBER study doesn't just provide data - it provides a roadmap for the next phase of the AI revolution. It's not about replacing humans, but enabling them to make better decisions. In a world with 700 million ChatGPT users, this is no longer a vision, but already reality.
Sources and References
This analysis is based on the complete NBER study and supplementary research materials:
- Primary Source: Aaron Chatterji, David J. Deming, et al. (2025). "How People Use ChatGPT." NBER Working Paper 34255. National Bureau of Economic Research.
- Data Validation: WildChat Dataset für Klassifizierungsvalidierung, O*NET-Datenbank für Arbeitsaktivitäts-Mapping
- Methodology: Privacy-preserving LLM-based classification, Data Clean Room for demographic analysis
All data and statistics are current as of publication date (September 2025).
Author
Falco Schneider
Founder, Ultra Relevant
Published
January 16, 2025
18 Min Read
Related resources
Keep exploring to deepen your AI visibility strategy.
5-Step GEO Playbook for AI Visibility
Follow five steps to get cited in ChatGPT, Perplexity and other AI search engines.
AI Content Framework That Ranks
Blend AI writing with editorial QA so Google rewards your work.
Pricing & Plans
Compare AI Brand Monitoring packages and launch with the plan that fits you.
How relevant is your product in AI searches?
68% of purchase decisions are influenced by AI recommendations today. Test your visibility for free now.