OpenAI Releases Deep Data on How ChatGPT is Changing Global Work

OpenAI Economic Research has released a massive and highly detailed dataset, dubbed 'OpenAI Signals,' designed to give academic researchers and economists an unprecedented window into how the public is integrating Artificial Intelligence into their daily professional lives. This data is not merely a usage counter; it is a sophisticated tool that maps consumer interactions with ChatGPT across the professional sphere, providing deep insights into the evolving relationship between AI technology and the global labor market.

The scope of this data is immense, covering 165 distinct work activities—known as Intermediate Work Activities (IWAs)—across nearly 100 countries. The dataset is structured quarterly, spanning a significant period from the first quarter of 2025 through the third quarter of 2026. The ability to track usage over time and compare it across diverse geographical locations makes this a powerful resource for comparative economic analysis. The core utility of the data lies in its ability to connect ChatGPT usage patterns directly to the standardized framework of IWAs.

To ensure that the data is comparable and meaningful across different industries and job types, OpenAI utilized ONET, which is a standardized framework. ONET groups specific, occupation-based tasks into broader, shared categories of work activities. By adopting this common framework, researchers can move beyond simply knowing that people use ChatGPT for work, to understanding what specific types of work tasks—such as preparing instructional materials, developing marketing content, programming code, or analyzing complex business data—are being augmented or replaced by AI assistance. This methodological rigor is crucial, as it allows for robust comparisons with existing, established occupational and employment data sets.

One of the most striking findings revealed by the analysis of these IWAs is the trend of 'de-concentration' in work-related usage. This means that, over time, the use of ChatGPT for work is becoming less reliant on the handful of most common, obvious tasks. Instead, the data suggests that countries are exhibiting a degree of specialization in their AI usage. Different nations are tending to specialize in different types of work tasks when utilizing ChatGPT. This observed specialization could be indicative of several underlying factors: it might reflect genuine, unique variations in the types of work performed across different national economies, or it could point to varying levels of access, adoption, or integration of advanced AI tools within specific national contexts. This suggests that the AI revolution is not a monolithic force, but rather a set of localized, specialized technological adoptions.

Further quantitative analysis highlights this shift. Across the 74 countries observed in every quarter, the ten most frequently reported work activities, on average, accounted for only 51.1% of a country’s classified work messages in Q3 2026. This represents a notable decrease from the 56.0% concentration observed in Q1 2025. This decline underscores the increasing breadth and diversity of how AI is being applied, moving beyond the initial, most obvious use cases.

When the data is paired with the established job descriptions of the U.S. economy, the comparison reveals substantial and critical differences. The visualization comparing ChatGPT usage to actual U.S. job activities demonstrates that the two distributions are not a simple mirror image. A significant portion—approximately 43%—of the employment-weighted task mix for mapped U.S. occupations falls into 'Other IWAs.' These are categories that encompass essential but less digitally visible tasks, such as moving physical materials, operating vehicles, or maintaining health and medical records. These crucial, yet often overlooked, activities were excluded from the original U.S. ChatGPT sample because they did not meet the necessary threshold for consistent representation in the individual user messages. This gap highlights a critical blind spot in current AI usage data: the underrepresentation of physical, logistical, and highly specialized, non-digital labor tasks.

It is vital to understand the limitations and the protective measures built into this data release. The data exclusively covers work-related messages originating from individual ChatGPT plans and explicitly excludes messages generated through enterprise-level accounts or the use of the Codex model. Furthermore, OpenAI has implemented stringent privacy protocols to protect individual users. Researchers are never given access to individual conversations; instead, automated systems classify messages only after all personally identifying information has been rigorously removed. To maintain user privacy, the platform publishes only aggregated statistics, with results for small groups withheld and statistical noise added to protect the integrity of individual messages. These robust privacy measures ensure that the research remains ethically sound while providing macro-level economic insights.

In summary, the OpenAI Signals data is a monumental resource that empowers researchers to conduct deep, multi-faceted studies on AI's economic impact. It allows for the tracking of evolving work patterns, the comparison of digital usage against physical labor realities, and the investigation of how AI adoption might accelerate or reshape national economic specializations. It serves as a foundational tool for understanding the future of work in an increasingly AI-augmented world.