OpenAI publishes data mapping ChatGPT work use across nearly 100 countries

OpenAI released data on October 26, 2026, mapping work-related ChatGPT messages to 165 activities across nearly 100 countries. The new OpenAI Signals release covers seven quarters, from early 2025 through September 2026. It gives researchers a common framework for comparing what people bring to ChatGPT with the work people do across occupations.

A shared language for different kinds of work

The release uses O*NET, a standardized occupational framework that groups job-specific tasks into activities shared across professions. Its 165 categories are called Intermediate Work Activities. Rather than treating every occupation as a separate case, the framework lets researchers compare ChatGPT messages with categories used in other occupational and employment data.

The activities include preparing instructional materials, developing marketing content, programming and analyzing business data. An interactive chart lets users choose a country, move from broad groups to specific activities, and follow changes over time. That offers a more detailed view than a single measure of work-related chatbot use.

OpenAI’s headline finding is that work messages became less concentrated in the most common activities. The comparison follows 74 countries present in every quarter, rather than the full set of nearly 100 countries. Across that consistent group, the ten leading activities accounted for a smaller average share by Q3 2026.

The top ten activities’ average share fell

Classified work messages

56.0%→51.1%

%

Average share represented by each country’s ten most common activities, across 74 countries observed in every quarter, according to OpenAI.

The country patterns were not uniform. OpenAI says different countries specialized in different types of work over time. It offers two possible explanations: differences in the work people perform and differences in access to AI tools. Those are questions for researchers to investigate, not established causes of the observed variation.

The chatbot sample and the economy do not line up

The dataset covers work messages on individual ChatGPT plans, not all use inside businesses. OpenAI’s individual-data page identifies those plans as Free, Go, Plus and Pro. Enterprise messages and Codex use are excluded, which OpenAI says likely underrepresents business and technical use cases. An employee’s individual account can appear; an enterprise conversation cannot.

There is another mismatch when comparing the messages with U.S. employment. About 43% of the task mix for mapped occupations, weighted by employment, falls into “Other” work activities rather than 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 its original U.S. ChatGPT sample. The company says this largely explains why its usage distribution differs substantially from the distribution of activities across U.S. jobs. The comparison therefore reflects both what users bring to ChatGPT and which activities the release reports separately.

Research access without reading conversations

The work-activity release sits within a broader individual Signals analysis based on sampled messages from July 2024 through September 2026. That longer observation window should not be confused with this release’s quarterly work-activity series, which begins in Q1 2025. Researchers can use the new categories to compare countries, track changes and connect usage with other economic indicators.

OpenAI says researchers do not read individual conversations. Automated systems classify messages after identifying information is removed. The company then publishes aggregated statistics, withholds results for small groups and adds statistical noise to protect individual messages. The public resource is therefore a set of protected statistical patterns, not a collection of workplace conversations for researchers to inspect.