OpenAI releases ChatGPT work-use data across 165 activities
OpenAI publishes a new view of ChatGPT work use
OpenAI has released OpenAI Signals v3.0, a dataset connecting work-related ChatGPT messages to 165 Intermediate Work Activities (IWAs) across nearly 100 countries. The quarterly data covers the period from the first quarter of 2025 through the third quarter of 2026 and is intended to help researchers examine AI’s role in the economy.
The release categorises the tasks people bring to ChatGPT, including preparing instructional materials, developing marketing content, programming and analysing business data. It is built for comparison across countries and over time, and can be matched with other economic indicators.
OpenAI uses the O*NET framework, which groups occupation-specific tasks into broader work activities shared by different occupations. That common structure allows researchers to compare patterns in ChatGPT use with occupational and employment data rather than treating AI activity as a single undifferentiated measure.
Use is becoming less concentrated in leading activities
Across 74 countries observed in every quarter, the ten most common activities accounted for an average of 51.1% of each country’s classified work messages in the third quarter of 2026. That share was 56.0% in the first quarter of 2025.
OpenAI says the change indicates that work-related use became less concentrated in the most common activities. Countries tended to specialise in their use over time, while the types of work in which they specialised differed. The company notes that these differences may reflect variation in work performed across countries or differing access to AI tools, questions the dataset is meant to support rather than resolve.
The task-level view complements the product direction represented by ChatGPT Work task-completion capabilities as organisations assess systems that can carry work through to outcomes rather than only generate responses. Signals v3.0 provides a way to examine which activity categories people actually bring to ChatGPT in different markets.
Important boundaries for interpreting the dataset
The data covers work-related messages from individual ChatGPT plans. It excludes enterprise messages and Codex use, so it is not a complete measure of AI use inside businesses or of all OpenAI products.
OpenAI also compares the distribution of mapped ChatGPT activities with tasks across US jobs. The distributions differ substantially, in part because about 43% of the employment-weighted task mix for mapped US occupations is assigned to Other IWAs that sit outside the individually reported categories. Examples include moving materials, operating vehicles and maintaining health or medical records; those categories did not meet OpenAI’s threshold for consistent representation in the original US ChatGPT sample.
For privacy, researchers do not read individual conversations. Automated systems classify messages after identifying information has been removed, while published figures are aggregated, small groups are withheld and statistical noise is added. For businesses, the practical implication is to use the data as a task-level benchmark for AI adoption, while accounting for its product coverage and classification limits.