The Pulse
OpenAI Maps 165 ChatGPT Work Activities Across Nearly 100 Countries
OpenAI released data tracking 165 work activities in consumer ChatGPT messages across nearly 100 countries from Q1 2025 to Q3 2026. The dataset shows work use spreading across more activities, but excludes enterprise messages and Codex and
AI.info Team · Written with AI · fact-checked ·
OpenAI’s new global dataset offers researchers a detailed view of work people bring to ChatGPT, but it does not cover the workplace use most likely to sit inside company accounts. Published on October 26, the release tracks 165 types of work activity across nearly 100 countries; OpenAI says it excludes enterprise messages and Codex, leaving a consumer-focused picture rather than a census of AI use on the job.
Work use spreads across more activities
The data groups ChatGPT work messages into 165 Intermediate Work Activities, or IWAs, using O*NET, a framework for organizing tasks shared across occupations. The categories range from preparing instructional materials and developing marketing content to programming and analyzing business data. The quarterly series runs from the first quarter of 2025 through the third quarter of 2026, and lets researchers compare activity patterns across countries and over time.
One trend in the release is a decline in the share of work messages captured by the most common activities. “work-related use became less concentrated in the most common activities,” write Aaron Chatterji and colleagues in the OpenAI Signals v3.0 materials. Across 74 countries with data in every quarter, the ten most common activities made up an average 51.1% of each country’s classified work messages in Q3 2026, compared with 56.0% in Q1 2025.
The US comparison leaves out a large slice of work
OpenAI also compares ChatGPT activity with the mix of tasks performed across U.S. jobs. The two distributions differ substantially, the company says, partly because about 43% of the employment-weighted task mix for the mapped occupations falls into “Other IWAs,” outside the activities reported individually. Those less-represented categories include moving materials, operating vehicles and maintaining health or medical records.
OpenAI says the omitted categories did not meet its threshold for consistent representation in its original U.S. ChatGPT sample. That boundary matters: the comparison is not a direct measure of how much ChatGPT has changed the tasks workers perform. It shows how classified consumer messages map against occupational tasks, with a substantial portion of the U.S. task mix grouped outside the detailed categories.
Consumer messages are not company-wide usage
The release covers work-related messages sent through individual ChatGPT plans, not enterprise accounts or Codex. OpenAI’s accompanying Signals documentation says the wider dataset draws on classifications of one million consumer ChatGPT messages per month, from July 2024 through September 2026, associated with accounts declaring an age of at least 18. It also excludes users who deleted their accounts or opted out of having their data used for training, and says those limits likely understate business use.
OpenAI says researchers did not read individual conversations: automated classifiers processed de-identified data, and the published statistics aggregate results, withhold small groups and add statistical noise. The company makes the data available for researchers to download and use under a Creative Commons Attribution 4.0 license.
A research resource, not a workforce survey
The release gives researchers a way to connect observed ChatGPT activity with standard occupational categories and other economic indicators. Its country-level patterns may help identify where work-related use is becoming more varied, while the U.S. comparison exposes gaps between a consumer chatbot sample and the full range of work in the economy.
But the dataset describes messages classified as work-related, not all workplace AI use or the share of workers using ChatGPT. Its clearest measured shift is narrower: among 74 countries observed throughout the period, the ten most common activity categories accounted for 51.1% of classified work messages in Q3 2026, down from 56.0% in Q1 2025.