OpenAI has published Signals data that maps how people use ChatGPT for work to 165 Intermediate Work Activities (IWAs), spanning nearly 100 countries and covering quarterly observations from Q1 2025 through Q3 2026. The Signals data connects user messages on individual ChatGPT plans to a standardized occupational framework so researchers can study task-level patterns of AI use across countries and over time.
The release classifies work-related ChatGPT messages into 165 IWAs derived from O*NET, a widely used framework that groups occupation-specific tasks into broader activities common across jobs. OpenAI highlights examples from the dataset including preparing instructional materials, developing marketing content, programming, and analyzing business data. Data are presented on a quarterly basis and include interactive charts that let users inspect the mix of activities within individual countries and track how that mix changed between Q1 2025 and Q3 2026. The public Signals dataset covers messages on individual ChatGPT plans only; it does not include enterprise-plan messages or usage of Codex.
OpenAI reports several notable patterns from the classified data. One key finding is a decline in concentration among the most common activities: across 74 countries observed in every quarter of the series, the ten most common activities accounted for an average of 51.1% of each country’s classified work messages in Q3 2026, down from 56.0% in Q1 2025. OpenAI interprets this shift as evidence that countries tended to specialize in different types of work brought to ChatGPT over the period. The release also shows substantial cross-country variation in the mix of activities, a feature OpenAI points to as an area researchers can probe to understand whether differences reflect the nature of national workforces, variation in access to AI, or other factors.
To make ChatGPT usage comparable with other employment statistics, OpenAI mapped messages to IWAs that are derived from O*NET. That common framework enables crosswalks between ChatGPT activity and occupation-level task distributions in U.S. labor statistics and other sources. When OpenAI compared ChatGPT activity with the activities performed across U.S. jobs, it found notable differences: roughly 43% of the employment-weighted task mix for the mapped U.S. occupations falls into Other IWAs that were not reported individually in the ChatGPT data. These omitted categories include tasks such as moving materials, operating vehicles, and maintaining health or medical records, which did not meet the company’s threshold for consistent representation in their original U.S. ChatGPT sample.
OpenAI positions the Signals release as a resource for researchers seeking to explore how AI integrates with economic activity. The dataset supports cross-country comparisons, temporal tracking of activity mixes, and linkage to other economic indicators. Researchers can use the interactive visualizations to move from broad activity groups down to specific tasks, examine which tasks users bring to ChatGPT, and compare usage patterns with the broader U.S. economy or other labor statistics. The combination of standardized IWAs and quarterly country-level data is intended to create a common input for studies of AI’s economic impacts.
Privacy protections and methodological limits are documented alongside the dataset. OpenAI says it removes identifying information before automated systems classify messages, that researchers do not read individual conversations, and that published outputs are aggregated statistics. To mitigate re-identification risks, the company withholds results for small groups and adds statistical noise to the published aggregates. OpenAI also cautions about constraints that affect coverage: the dataset excludes enterprise-plan messages and Codex usage, and some categories of work did not meet the threshold for consistent reporting in the ChatGPT sample. Those limitations shape which activities appear in the public release and the extent to which the data represent all work-related use of the tool.
By making this quarterly, occupation-aligned dataset available, OpenAI invites independent analysis of observed patterns and their implications for workers and businesses. The Signals data provides a new, standardized way to examine how people bring different tasks to ChatGPT, while the accompanying documentation and visualizations aim to help researchers interpret results in light of the dataset’s coverage and privacy safeguards. As researchers explore cross-country differences and shifts in activity mixes between Q1 2025 and Q3 2026, the release is likely to become a reference point for empirical work on AI’s role in the world of work.
Source: Read the original source
Leave a Reply