OpenAI Publishes Data on 165 Work Activities

OpenAI published new OpenAI Signals data on October 26 covering 165 intermediate work activities across nearly 100 countries. The dataset presents quarterly data from Q1 2025 through Q3 2026. It lets users explore activities by country, move from broad groups to more specific tasks, and see how their distribution changes over time.

What activities Signals covers

Intermediate work activities, or IWAs, are task categories shared across different occupations. The classification is based on O*NET, a standardized framework that groups specific job tasks into broader activities. Examples in the announcement include preparing instructional materials, developing marketing content, programming, and analyzing business data.

OpenAI reports that, across the 74 countries observed in every quarter, the ten most common activities accounted for an average of 51.1% of each country’s classified work-related messages in Q3 2026, the third quarter. In Q1 2025, the first quarter, that share was 56.0%. The comparison shows that work-related messages became less concentrated in the most frequent activities.

What the comparison with jobs means

The common framework makes it possible to compare activities found in ChatGPT messages with tasks in U.S. jobs. For this comparison, tasks are weighted according to employment in the occupations analyzed, so they do not all carry the same weight.

About 43% of that task mix is grouped under “Other IWAs,” the activities not published as individual categories. These include moving materials, operating vehicles, and maintaining medical records. That 43% refers to tasks in the mapped U.S. occupations, not to a share of ChatGPT messages. The categories were grouped because they did not meet the threshold for consistent representation in the original U.S. ChatGPT sample. OpenAI’s announcement explains the activities, results, and comparison with jobs.

Scope, privacy, and download

The analysis covers work-related messages from individual ChatGPT plans and excludes Enterprise messages and Codex use. The accompanying Signals page identifies those individual plans as Free, Go, Plus, and Pro. Signals documentation details the scope of the individual data.

According to OpenAI, researchers do not read individual conversations. Automated systems classify messages after identifying information has been removed. The company publishes aggregated statistics, omits results for small groups, and adds statistical noise, a modification of values that helps protect individual messages.

The data can be downloaded as CSV files, along with a dictionary explaining the fields and methodology. They are published under the Creative Commons CC BY 4.0 license, which allows them to be used, shared, and adapted with attribution or a citation. The download page provides the files and reuse terms.