OpenAI Signals data maps how people use ChatGPT at work

What was released?

OpenAI Economic Research published OpenAI Signals 3.0 data on Monday (OpenAI). It classifies work-related ChatGPT messages into 165 work activities across nearly 100 countries. The categories are the Intermediate Work Activities (IWAs) from the U.S. O*NET framework, so the data can be matched with other employment statistics.

The series is quarterly and runs from Q1 2025 to Q3 2026. Examples include preparing instructional materials, developing marketing content, programming and analyzing business data.

What do the numbers show?

Work use is becoming less concentrated in the most common tasks. Across the 74 countries observed in every quarter, the ten most common activities made up an average of 51.1% of classified work messages in Q3 2026, down from 56.0% in Q1 2025. OpenAI says countries specialized over time, and in different directions.

In the U.S. comparison, ChatGPT use and the actual work mix differ substantially. The main reason is that about 43% of the employment-weighted U.S. task mix falls into activities outside the individually reported categories, such as moving materials, operating vehicles or maintaining medical records.

What to keep in mind

The data covers only work messages on individual ChatGPT plans; enterprise messages and Codex use are excluded. OpenAI says researchers do not read individual conversations: automated systems classify messages after identifying information is removed, results for small groups are withheld and statistical noise is added.

Why it matters: if you plan or study AI adoption, you get a downloadable, country-level baseline (data download). It is useful to compare with your own organization's usage, but remember it is consumer data, not enterprise data.