OpenAI Releases Global Data on How People Use ChatGPT at Work

OpenAI has released a new dataset examining how people use ChatGPT for work across nearly 100 countries, giving researchers a more detailed way to study the relationship between generative AI, workplace tasks and the global economy.

Published on October 26, 2026, by OpenAI Economic Research, the release expands the company's OpenAI Signals research initiative.

It maps work-related ChatGPT activity to 165 categories of work activities and provides quarterly data from Q1 2025 through Q3 2026.

The categories include tasks such as preparing instructional materials, developing marketing content, programming and analyzing business data.

Researchers can use the data to examine how work-related ChatGPT usage differs by country, changes over time and relates to broader economic indicators.

OpenAI says the release is intended to support further research into AI adoption and labor markets.

It provides evidence about the work people bring to ChatGPT, but it does not, by itself, establish whether AI has increased productivity, eliminated jobs or changed employment outcomes.

The data shows work-related AI use becoming more diverse

One of the main findings in the release is that work-related ChatGPT activity has become less concentrated in the most common categories.

Across 74 countries with data available for every quarter in the study period, the ten most common work activities accounted for an average of 51.1% of classified work-related messages in Q3 2026.

In Q1 2025, those same ten leading activities represented 56.0%.

The decline suggests that work-related usage is spread across a wider range of activities than it was at the beginning of the period.

OpenAI also reports that different countries increasingly show different patterns of specialization in the work people bring to ChatGPT.

However, the figures should be interpreted carefully.

A change in the distribution of ChatGPT messages does not necessarily mean that the underlying economy has changed in the same way.

Differences could reflect the types of jobs people perform, access to AI tools, or how workers choose to use ChatGPT.

The new dataset gives researchers a basis for investigating those possibilities rather than treating any single explanation as established.

Why OpenAI used an established occupational framework

To organize the data, OpenAI mapped ChatGPT usage to Intermediate Work Activities, or IWAs, drawn from O*NET, a standardized occupational information framework used to describe work and job-related tasks.

The framework groups activities that appear across different occupations.

This helps researchers compare the kinds of work people ask ChatGPT to support with established information about occupational tasks.

For example, a request to prepare instructional materials or analyze business data can be categorized as a work activity without assuming that the person making the request holds a particular job title.

The approach is designed to make comparisons between countries and periods more consistent.

The classification also creates an opportunity to compare observed AI usage with employment data and other economic indicators.

Researchers may be able to investigate whether the activities appearing in ChatGPT conversations resemble the work performed across different occupations, or whether usage patterns point to differences worth studying.

The dataset does not mean that OpenAI can determine every user's profession or infer the full range of tasks performed in a job.

It categorizes sampled, work-related messages according to the framework used in the research.

What the research can—and cannot—tell us about jobs

The new release contributes to a broader question facing employers, workers and policymakers: whether AI is simply helping people complete existing tasks or changing the range of work individuals can perform.

Earlier OpenAI Economic Research publications have explored how people use ChatGPT for activities associated with occupations outside their usual professional roles.

That research describes a pattern known as task crossover, in which workers use AI to help with tasks traditionally associated with another occupation.

The new international dataset complements that line of research by making work-activity patterns available across a broader set of countries and quarters.

It could help researchers examine whether particular activities become more or less common over time and how those patterns vary by geography.

Still, message categories are not equivalent to measures of job performance.

A worker asking ChatGPT for help with a task does not prove that the task was completed successfully, that it would otherwise have required another employee, or that the worker's job has changed permanently.

Establishing those effects requires additional evidence, including research on productivity, job responsibilities, employment, wages and the way organizations implement AI tools.

For businesses, the practical implication is that AI adoption should be evaluated at the task level as well as at the job-title level.

Understanding which activities employees use AI for can help organizations identify training needs, review workflows and establish appropriate controls.

The dataset can inform those investigations, but it is not a substitute for company-specific evidence.

Which ChatGPT activity is included in the release?

The newly released data covers work-related messages from individual ChatGPT plans.

It does not include messages from ChatGPT Enterprise accounts or usage of Codex, OpenAI's coding-focused product.

That distinction matters when interpreting the findings.

The dataset provides a view of consumer ChatGPT usage in work contexts, not a complete measurement of AI adoption across every workplace or business.

Enterprise use may follow different patterns because organizations can provide employees with dedicated accounts, internal guidance, company-specific workflows and centrally managed AI tools.

Codex usage is also outside the scope of this release, so the data should not be treated as a comprehensive measure of AI-assisted software development.

The dataset is therefore best understood as one source of evidence about how individuals use ChatGPT for work, rather than a complete picture of all AI activity in the global labor market.

How OpenAI says it protects user privacy

OpenAI says the research does not require researchers to read individual conversations.

According to the company, automated systems classify messages after identifying information has been removed.

The company also says it publishes aggregated statistics, withholds results for small groups and adds statistical noise to protect individual messages.

These measures are intended to reduce the risk that published results could reveal information about individual users.

The release is a dataset of aggregated usage patterns, not a public collection of people's ChatGPT conversations.

Nevertheless, readers should distinguish between OpenAI's stated privacy safeguards and an independent assessment of how those safeguards perform in practice.

Researchers using the data should also account for its scope and classification methods when drawing conclusions.

Privacy protections, sampling decisions and the categories used to classify work activity all affect how the results can be interpreted.

Where researchers can explore the data

OpenAI has made the new work-activity data available through its OpenAI Signals research initiative.

Researchers can use the dataset to explore activity categories by country, compare quarterly patterns and connect the findings with other economic information.

The release may be useful to labor economists, policymakers, business researchers and organizations studying how AI is incorporated into everyday work.

Its greatest value will likely come from combining usage patterns with independent evidence about employment, occupational responsibilities and measurable workplace outcomes.

The central takeaway is that OpenAI is providing a more detailed view of the work-related tasks people bring to ChatGPT.

The data shows that usage patterns vary across countries and that activity is becoming less concentrated in the most common categories.

Whether those patterns translate into sustained changes in productivity, job design or employment remains a question for further research.