OpenAI Signals ChatGPT Work Data Maps 165 Activities Across Nearly 100 Countries

OpenAI has released new global data on how people use ChatGPT for work. The OpenAI Signals dataset groups work-related messages into 165 activity categories across nearly 100 countries, with quarterly observations from Q1 2025 through Q3 2026. The release is intended to help researchers examine how ChatGPT use varies by activity, country and time—not to announce a new ChatGPT feature.

The dataset offers a more detailed view of the kinds of tasks people bring to ChatGPT, from preparing instructional materials and developing marketing content to programming and analyzing business data. It also gives researchers a structured way to compare those patterns with other labor-market information. OpenAI describes the release in its October 26, 2026 announcement.

What the OpenAI Signals ChatGPT work data includes

The work-use data organizes messages into 165 Intermediate Work Activities, or IWAs. OpenAI says the categories draw on O*NET, a standardized framework that groups occupation-specific tasks into broader activities shared across occupations. Using a common taxonomy can make it easier to compare the types of work people ask ChatGPT to help with against occupational and employment data.

The observations are presented quarterly, beginning in Q1 2025 and ending in Q3 2026. The interactive view allows users to explore activity groups, drill down to more specific activities, and compare the mix of work-related messages across countries and quarters. OpenAI says the data covers nearly 100 countries; availability can vary because countries with insufficient data or where OpenAI does not operate are excluded.

This is a dataset about observed use of ChatGPT—not a complete inventory of work in each country. It records activities represented in the included messages and should be read as a view into one tool’s usage, rather than a direct measurement of national productivity, employment or total AI adoption.

The headline finding: work use appears less concentrated

One finding in OpenAI’s announcement is that the most common activities accounted for a smaller share of classified work messages over the period. Across 74 countries with observations in every quarter, the ten most common activities represented an average of 56.0% of each country’s classified work messages in Q1 2025. By Q3 2026, that average was 51.1%.

In practical terms, a smaller share of the classified messages in this group of countries fell into the top ten activities in the later quarter. OpenAI describes this as work-related use becoming less concentrated, with countries tending to specialize in different types of work. That is a useful pattern to investigate, but the figures alone do not explain why it happened. Differences could reflect the work people do, access to AI tools, changes in who uses ChatGPT, or other factors.

The finding is also about the distribution of messages across activity categories, not the number of jobs created or eliminated, the amount of time saved, or the economic value of the work. Those are separate questions requiring additional evidence.

Why the release may matter to researchers

Researchers studying AI and labor often need more than a broad measure of whether a tool is being used. Activity-level data can help frame questions about what people ask AI to do, whether usage patterns differ across countries, and how those patterns change over time. The O*NET connection provides a shared vocabulary for comparing AI-related messages with occupational task information.

For example, a researcher could examine whether the mix of activities changes over successive quarters, or whether countries show different patterns in a particular activity group. The dataset may also be combined with other economic indicators to explore relationships between work-related ChatGPT use and labor-market conditions. Such comparisons can point to questions for further study; they do not, on their own, establish that ChatGPT caused an observed economic change.

OpenAI also presents a comparison between ChatGPT work messages and activities performed across U.S. jobs. It warns that the distributions differ substantially. In particular, about 43% of the employment-weighted task mix for the mapped U.S. occupations falls into “Other IWAs,” outside the individually reported categories. The examples OpenAI gives include moving materials, operating vehicles and maintaining health or medical records. These categories did not meet the company’s threshold for consistent representation in the original U.S. ChatGPT sample.

That limitation matters when interpreting a chart or ranking. A category that appears prominent among messages does not necessarily occupy a similar share of work across the economy. Likewise, activities that are common in the economy may be missing from or grouped together in the ChatGPT comparison.

Important limitations to keep in mind

• Individual-plan usage only: The release covers work-related messages on individual ChatGPT plans. OpenAI says it does not include Enterprise messages or Codex use. The results therefore do not represent all workplace AI use and may understate business or technical use cases.
• Messages are not workers or jobs: The data describes classified messages. It should not be treated as a count of unique workers, businesses, tasks completed, hours saved or jobs affected.
• Country comparisons need context: Workforces, industries, language use, internet access and ChatGPT adoption differ. A difference in message shares does not necessarily mean workers in one country perform more of a particular activity overall.
• Some country and category data is unavailable: OpenAI excludes countries with insufficient data or where it does not operate. It also omits some activity categories from the U.S. comparison when they were not consistently represented in the original ChatGPT sample.
• Classification and aggregation shape the results: Messages are assigned to categories through automated systems. The published statistics are aggregated, and OpenAI applies privacy protections, including withholding results for small groups and adding statistical noise.
• Association is not causation: The dataset can help researchers describe patterns and develop hypotheses. It does not by itself prove that ChatGPT caused changes in productivity, wages, employment or the composition of work.

OpenAI says researchers do not read individual conversations for this release. According to the company, identifying information is removed before automated classification, and the published data uses aggregated statistics with small-group suppression and statistical noise. These are important privacy measures, while also meaning that researchers should consult the methodology before drawing conclusions from small differences.

How to use the data in practice

1. Start with the documentation. OpenAI provides downloadable CSV files and a data dictionary on its Signals data and methodology page. Review the definitions, fields and coverage before building a chart or analysis.
2. Choose a question before selecting metrics. Decide whether you are studying change over time, variation by country, or a specific work activity. Be explicit about whether your denominator is all classified work messages or a particular subset.
3. Check coverage for every comparison. When comparing quarters, verify that the countries and categories are present across the periods you analyze. The headline trend uses 74 countries observed in every quarter; a different country set may produce a different result.
4. Use external data carefully. If joining Signals data to employment or economic indicators, align dates, geography and category definitions. Explain any mapping choices and avoid treating a correlation as evidence of cause and effect.
5. Report the scope clearly. State that the dataset reflects individual-plan ChatGPT messages, not all enterprise use or the full set of work activities in a country. Include the time period and any exclusions in charts and summaries.
6. Cite and attribute the release. OpenAI’s data page says Signals data is licensed under Creative Commons CC BY 4.0, which permits use, sharing and adaptation with attribution or citation. Follow the suggested citation and license terms when reusing the data.

For employers and product teams, the dataset can be a prompt for questions about the kinds of tasks users are bringing to AI, but it is not a substitute for local user research or an organization’s own usage and outcome data. For journalists and analysts, the key is to distinguish the observed pattern in ChatGPT messages from broader claims about the labor market.

A research release, not a product launch

The main significance of this announcement is the availability of a more granular dataset for studying AI use at work. OpenAI is publishing categories, country-level views and a quarterly time series that researchers can explore and compare with other sources. The release does not describe a new ChatGPT capability, pricing change or workplace product.

That makes the story a better fit for AI research, labor economics and data coverage than for product-focused reporting. Its value will depend on how researchers use the data, whether the patterns hold up in other sources, and how future updates extend or revise the time series.

Frequently asked questions

What is the OpenAI Signals ChatGPT work data?

It is aggregated data about work-related messages sent to ChatGPT on individual plans. OpenAI maps the messages to 165 O*NET-based Intermediate Work Activities and provides quarterly views across nearly 100 countries from Q1 2025 through Q3 2026.

Does the dataset include ChatGPT Enterprise or Codex?

No. OpenAI says this release excludes Enterprise messages and Codex use. It therefore captures only part of ChatGPT use for work and should not be presented as a complete measure of workplace AI adoption.

What is the main trend OpenAI reports?

Across 74 countries with data in every quarter, the ten most common activity categories averaged 56.0% of classified work messages in Q1 2025 and 51.1% in Q3 2026. OpenAI characterizes the shift as less concentration in the most common activities.

Can the data show whether ChatGPT is changing jobs?

Not on its own. The data describes the activity mix in a sample of ChatGPT work-related messages. It does not prove that AI caused changes to jobs, wages, productivity or employment. Those questions require additional evidence and careful research design.

Where can researchers download the files?

OpenAI links the CSV downloads and data dictionary from its data and methodology page. The page also provides the recommended citation and license information.