How People Use ChatGPT
Chatterji, Cunningham, Deming, Hitzig, Ong, Shan, Wadman
2025NBER Working Paper Series32 citations
Adoption / usageInterdisciplinary
LLM / Generative AIWriting / contentSoftware / codingEducationCustomer serviceScience / researchJunior / entry-levelSenior / older workersGenderDeveloping economiesHuman-AI collaborationDecision-makingTraining / upskillingGeneral automation
AbstractDespite the rapid adoption of LLM chatbots, little is known about how they are used.We document the growth of ChatGPT's consumer product from its launch in November 2022 through July 2025, when it had been adopted by around 10% of the world's adult population.Early adopters were disproportionately male but the gender gap has narrowed dramatically, and we find higher growth rates in lower-income countries.Using a privacy-preserving automated pipeline, we classify usage patterns within a representative sample of ChatGPT conversations.We find steady growth in work-related messages but even faster growth in non-work-related messages, which have grown from 53% to more than 70% of all usage.Work usage is more common for educated users in highly-paid professional occupations.We classify messages by conversation topic and find that "Practical Guidance," "Seeking Information," and "Writing" are the three most common topics and collectively account for nearly 80% of all conversations.Writing dominates work-related tasks, highlighting chatbots' unique ability to generate digital outputs compared to traditional search engines.Computer programming and self-expression both represent relatively small shares of use.Overall, we find that ChatGPT provides economic value through decision support, which is especially important in knowledge-intensive jobs.
SummaryChatterji et al. use automated LLM-based classifiers on a privacy-preserving random sample of approximately 1.1 million ChatGPT conversations from May 2024 to June 2025 to document global adoption patterns, usage topics, user intent, and demographic variation in how people use generative AI chatbots
Main FindingNon-work messages grew from 53% to over 70% of ChatGPT usage between June 2024 and June 2025, with the three most common topics being Practical Guidance, Seeking Information, and Writing (collectively 78% of messages); educated users in highly-paid professional occupations are substantially more likely to use ChatGPT for work-related tasks and for Asking rather than Doing messages
Primary Datasets
Privacy-preserving sample of actual ChatGPT conversations
Secondary Datasets
WildChat public conversations (Zhao et al. 2024) for classifier validation; O*NET Database Version 29.0 for work activities taxonomy; World Gender Name Dictionary and Social Security popular names for gender classification; World Bank GDP and internet population data
- Key Methods
- Automated LLM-based classification pipeline applied to privacy-preserving random sample of ChatGPT conversations, with validation against public WildChat dataset; secure data clean room for matching to aggregated employment categories
- Sample Period
- 2024-2025
- Geographic Coverage
- US
- Sample Size
- Approximately 1.1 million classified messages from May 2024-June 2025; 700 million weekly active users and 2.6 billion daily messages by July 2025
- Level of Analysis
- Individual, Occupation, Industry, Country
- Occupation Classification
- O*NET (IWA and GWA taxonomies), SOC 2-digit codes
- Industry Classification
- SOC 2-digit codes
- Replication Package
- Partial
NotesNBER WP 34255; first large-scale analysis of real ChatGPT usage; non-work usage grew from 53% to 70%+
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year
[Claude classification]: NBER WP 34255; first large-scale analysis of real ChatGPT usage using internal data; novel privacy-preserving methodology with no human viewing of messages; validates automated classifiers against public WildChat dataset; uses secure data clean room to link usage to aggregated employment/education categories; introduces new Asking/Doing/Expressing taxonomy for user intent; non-work usage grew from 53% to 70%+ over one year