Summary
OpenAI has released quarterly data mapping work-related messages on individual ChatGPT plans to 165 work activities across nearly 100 countries. The company reports that, across 74 countries with data for every quarter, the ten most common activities made up a smaller share of classified work messages in Q3 2026 than in Q1 2025.
OpenAI Economic Research has released OpenAI Signals data that maps work-related messages on individual ChatGPT plans to 165 work activities across nearly 100 countries. The quarterly dataset runs from Q1 2025 through Q3 2026 and is intended to help researchers examine how people use ChatGPT for work across countries and over time.
A task-level view of ChatGPT use
The categories are Intermediate Work Activities, or IWAs, drawn from O*NET, a framework that groups occupation-specific tasks into broader activities shared across occupations. Examples include preparing instructional materials, creating marketing content, programming and analysing business data.
Using a common occupational framework gives researchers a way to compare the kinds of work-related tasks brought to ChatGPT with other occupational and employment data. OpenAI says researchers can also connect the dataset with other economic indicators. The release includes an interactive chart for exploring activities by country and quarter.
The data describes classified work-related messages, not the proportion of workers performing an activity or the share of working hours it accounts for. It covers individual ChatGPT plans and excludes enterprise messages and Codex use, so it represents a defined part of ChatGPT activity rather than all workplace use of AI.
Work-related messages are spread across more activities
OpenAI reports that work-related ChatGPT use became less concentrated in the most common activities. Across 74 countries with observations in every quarter, the ten most common activities accounted for an average of 56.0% of each country’s classified work messages in Q1 2025. By Q3 2026, that average was 51.1%.
The figures indicate that classified ChatGPT use was distributed across a broader range of activity types in the later quarter. OpenAI says countries tended to specialise in different types of work-related use over time. The company suggests that differences in the work done across countries, or in access to AI tools, could help explain the patterns; the data release presents these as questions for further research.
OpenAI also compares ChatGPT activity with the employment-weighted task mix of mapped US occupations. It says about 43% of that occupational task mix falls into “Other IWAs” rather than the individually reported categories. These omitted activities include moving materials, operating vehicles and maintaining health or medical records. They were not consistently represented enough in the original US ChatGPT sample to meet the reporting threshold, a qualification relevant to interpreting that comparison.
For privacy, OpenAI says automated systems classify messages after identifying information is removed; researchers do not read individual conversations. The published data consists of aggregated statistics, with results withheld for small groups and statistical noise added to protect individual messages.