OpenAI has released new global data showing how people use ChatGPT for work across 165 work activities in nearly 100 countries. Published on October 26, 2026, the dataset is part of OpenAI Signals, the company’s research initiative designed to help researchers examine AI adoption, work patterns and the changing relationship between AI tools and the labor market. The data covers quarterly activity from the first quarter of 2025 through the third quarter of 2026.
One notable finding is that work-related ChatGPT use is becoming more diverse. Across 74 countries with data available for every quarter, the 10 most common activities accounted for 51.1% of classified work messages in Q3 2026, down from 56% in Q1 2025. The dataset includes activities such as preparing instructional materials, creating marketing content, programming and analyzing business data, giving researchers a more detailed way to study how people use AI in their jobs.
Key Takeaways
- 165 work activities: OpenAI has mapped work-related ChatGPT messages to 165 standardized activity categories.
- Nearly 100 countries: Researchers can compare work-related usage patterns across countries and quarters.
- Seven quarters of data: The dataset spans Q1 2025 through Q3 2026.
- More diverse usage: The 10 most common activities accounted for 51.1% of classified work messages in Q3 2026, compared with 56% in Q1 2025 across 74 consistently observed countries.
- Research focus: The data is intended to help researchers examine AI use alongside employment and economic indicators.
- Privacy safeguards: OpenAI says the release uses automated classification, aggregated statistics and additional privacy protections.
- Important limitation: The dataset covers individual ChatGPT plans, not enterprise messages or Codex usage, and does not directly measure productivity gains or job losses.
What Is OpenAI Signals?
OpenAI Signals is a research initiative that publishes data and analysis about how people use ChatGPT. The latest release expands its work-related data by organizing messages into standardized categories that researchers can compare across locations and time periods.
The new dataset uses 165 Intermediate Work Activities, or IWAs, drawn from the ONET occupational information framework. ONET groups tasks into categories that describe activities shared across different jobs, making it easier to compare AI use with established occupational data.
For example, a marketing professional might use ChatGPT to develop promotional content, while a teacher could ask it to prepare instructional materials. A software developer might use the same tool to write or debug code. Mapping these interactions to common activity categories helps researchers examine similarities and differences in how people use AI at work.
The release also provides quarterly data rather than a single snapshot. Researchers can therefore explore how the distribution of work-related activities changes over time and compare those patterns with other economic indicators.
The dataset is available through OpenAI Signals, where researchers can explore and download the data and review the methodology.
What the New Data Reveals About ChatGPT at Work
One of the most significant findings concerns the concentration of work-related usage.
Across 74 countries observed in every quarter, the 10 most common activities represented an average of 56% of classified work messages in Q1 2025. By Q3 2026, that figure had fallen to 51.1%.
| Metric | Q1 2025 | Q3 2026 |
|---|---|---|
| Countries included in the consistent comparison | 74 | 74 |
| Share represented by the 10 most common activities | 56.0% | 51.1% |
| Share represented by all other classified activities | 44.0% | 48.9% |
Source: OpenAI Economic Research. Figures refer to classified work messages across countries observed in every quarter. Correction, October 27, 2026: Q1 2025 through Q3 2026 spans seven quarters, not nine; the stated date range is unchanged.
The decline suggests that work-related ChatGPT use is becoming less concentrated in a small group of activities. More of the classified messages fall into other categories, indicating a broader distribution of tasks.
However, the figures should not be interpreted as proof that the total volume of AI-assisted work has increased or that particular jobs are becoming more productive. They measure the share of classified work messages, not the number of hours saved, tasks completed or jobs created.
OpenAI says different countries also exhibit different patterns of specialization. These differences may reflect variations in the work people perform, access to AI tools or other economic and social factors. The dataset gives researchers a way to investigate these questions, but it does not establish a single explanation.
Why OpenAI Is Using Standardized Work Categories
Without a common classification system, comparing ChatGPT usage across countries can be difficult. People may describe similar tasks in different ways, and the same job title can involve very different responsibilities in different industries.
O*NET provides a standardized framework for organizing occupational activities. OpenAI’s mapping connects work-related messages to 165 Intermediate Work Activities, allowing researchers to examine AI use through a structure that can also be compared with employment information.
The approach could support research into questions such as:
- Which work activities are most commonly supported by ChatGPT in different countries?
- How does the mix of AI-assisted work change over time?
- Do the activities people bring to ChatGPT resemble the tasks performed across occupations?
- How does AI usage vary alongside broader economic and labor-market indicators?
The framework is particularly useful because AI adoption does not affect every task in the same way. Writing a draft, preparing a presentation, analyzing data and operating physical equipment involve different requirements. A task-level view can help researchers study these distinctions more carefully than broad measures of AI adoption alone.
How Researchers Can Use the Dataset
OpenAI says the data can be connected with other economic and occupational sources to investigate how AI is being used in real-world work.
One application is comparing ChatGPT activity with the task distribution across U.S. occupations. This can help researchers identify areas where people frequently use ChatGPT and areas where AI-related activity appears less common.
OpenAI notes that about 43% of the employment-weighted task mix for the mapped U.S. occupations falls into the framework’s broader “Other IWAs” category. These activities include work such as moving materials, operating vehicles and maintaining health or medical records.
That figure is not the share of ChatGPT messages involving those activities. Instead, it represents a portion of the employment-weighted task mix that falls outside the individually reported categories in the comparison.
The distinction matters because AI usage data and the overall structure of employment measure different things. Some tasks may be common in the economy but rarely appear in ChatGPT messages. Others may be particularly well suited to conversational AI and therefore show up disproportionately in the dataset.
Researchers can use the difference to develop more precise questions about AI adoption, rather than assuming that the tasks most visible in ChatGPT usage represent the entire labor market.
Privacy and Data Limitations
OpenAI says the Signals release is designed to provide useful aggregate statistics without making individual conversations available to researchers.
According to the company, identifying information is removed before automated systems classify messages. OpenAI publishes aggregated results, withholds results for small groups and adds statistical noise to protect individual messages.
The data also has important scope limitations. It covers work-related messages from individual ChatGPT plans, including Free, Go, Plus and Pro accounts. It excludes enterprise messages and Codex use.
As a result, the release should not be treated as a complete picture of AI use across all businesses. Companies using enterprise deployments or coding tools may have different usage patterns that are not represented in this dataset.
The data also cannot, on its own, establish whether ChatGPT has improved productivity, reduced employment or changed wages. Those questions require additional evidence, including information about work outcomes, organizational practices and changes in the labor market.
The Bigger Picture
OpenAI’s latest Signals release reflects a broader effort to understand AI adoption at the level of individual work activities. Instead of asking only how many people use ChatGPT, researchers can examine what kinds of tasks people bring to the tool and how those patterns differ across countries.
That level of detail could help policymakers, economists and businesses develop more grounded assessments of AI’s economic impact. It may also help identify where workers are using AI to support existing responsibilities and where new workflows are emerging.
At the same time, usage data is only one part of the picture. Understanding AI’s full effect on work will require linking these patterns to measurable outcomes, including productivity, wages, hiring, training and the quality of completed work.
FAQs
1. What data has OpenAI released?
OpenAI released data mapping work-related ChatGPT messages to 165 Intermediate Work Activities across nearly 100 countries. The quarterly dataset covers Q1 2025 through Q3 2026.
2. What are Intermediate Work Activities?
Intermediate Work Activities are standardized categories used to describe tasks performed across occupations. OpenAI uses the O*NET framework to organize ChatGPT work-related messages into these categories.
3. What does the 51.1% figure mean?
Across 74 countries observed in every quarter, the 10 most common activities accounted for 51.1% of classified work messages in Q3 2026, down from 56% in Q1 2025. This indicates a less concentrated distribution of classified activities, not necessarily an increase in total usage.
4. Does the dataset include ChatGPT Enterprise and Codex?
No. The release covers work-related messages from individual ChatGPT plans and excludes enterprise messages and Codex usage. It therefore does not capture every way businesses use OpenAI technology.
5. Does the data prove that ChatGPT is replacing jobs?
No. The dataset describes patterns of AI use, not direct employment outcomes. Additional research is needed to determine how AI affects productivity, hiring, wages and job responsibilities.
Looking Ahead
Researchers will be able to use the new dataset to compare AI-assisted work across countries, track changes over time and connect usage patterns with other economic information. The most useful findings will likely emerge when these statistics are combined with independent evidence about how work is performed and how organizations measure results.
For OpenAI, the release adds more detail to its effort to document AI adoption beyond headline usage figures. The central question now is not only which activities people bring to ChatGPT, but how those interactions affect the quality, speed and organization of work. Answering that question will require broader datasets and careful analysis rather than usage statistics alone.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
