OpenAI releases ChatGPT work-activity dataset spanning 165 tasks across 100 countries
OpenAI Signals data maps ChatGPT usage to standardized work categories, enabling economic research on AI adoption patterns globally.
Last verified:
ChatGPT Work-Use Dataset Reveals Geographic and Task Diversification
According to OpenAI Blog, the company released Signals data tracking how 165 distinct work activities are performed using ChatGPT across nearly 100 countries during the eighteen-month period from Q1 2025 through Q3 2026. The dataset maps user queries to Intermediate Work Activities (IWAs)—a standardized taxonomy of work tasks derived from O*NET, the U.S. Department of Labor’s occupational information system. This release aims to provide researchers with empirical ground truth on where and how artificial intelligence is being adopted for professional tasks, moving beyond anecdotal adoption claims toward quantifiable patterns.
The core finding underscores a shift in how ChatGPT serves the global workforce. OpenAI reports that across 74 countries with continuous quarterly data, the concentration of work-related messages in the top ten activities declined from 56.0% in Q1 2025 to 51.1% by Q3 2026. This downward trend suggests two competing dynamics: users are experimenting with ChatGPT for a broader range of tasks, and different countries are developing specialized use-case profiles rather than converging on identical workflows.
Geographic Divergence in Work-Task Adoption
The dataset reveals substantial variation in how ChatGPT is deployed for work across regions. According to OpenAI Blog, this heterogeneity likely reflects both differences in occupational mix (countries with larger technical workforces may rely more heavily on programming assistance, for example) and unequal access to competing AI tools. The interactive visualizations released alongside the data allow researchers to track these country-level patterns over time and correlate them with external economic indicators—labor force composition, sectoral GDP, technology adoption rates—creating a bridge between microeconomic user behavior and macroeconomic modeling.
Divergence Between ChatGPT Use and Actual Employment Composition
A secondary but significant finding addresses a structural mismatch. OpenAI reports that ChatGPT work-use distributions differ substantially from the actual composition of U.S. employment. The gap is explained partly by methodology: approximately 43% of employment-weighted task mix for mapped U.S. occupations falls into “Other IWAs”—catch-all categories encompassing physical labor (materials handling, vehicle operation, equipment maintenance) that fall outside the individually tracked work activities. This discrepancy illustrates that ChatGPT adoption is concentrated in knowledge work and creative tasks, while routine physical and operational labor remains largely offline relative to the platform.
Why This Matters
This dataset addresses a critical gap in AI policy and economic research: until now, claims about ChatGPT’s impact on occupational displacement, skill demand, and geographic inequality have relied on surveys, vendor metrics, or inference from search behavior. OpenAI’s release of structured, anonymized usage data linked to a standardized occupational framework enables researchers to move from assertion to measurement.
For labor economists, the data surfaces whether AI adoption is concentrating risk in specific occupations or spreading diffusely across the workforce. For policymakers designing retraining or immigration policy, it offers evidence on which tasks are most exposed to automation and where that exposure varies geographically. For organizations deploying ChatGPT internally, the breakdown by country and activity type provides a reference distribution against which to benchmark their own usage—a validity check on whether their adoption pattern is typical or anomalous.
The ongoing quarterly updates (through Q3 2026, with the release suggesting continuation) also create a time-series foundation for detecting inflection points: when does a new ChatGPT capability (e.g., vision, voice, real-time web access) measurably shift the portfolio of work activities users bring to the platform? The data, in short, transforms ChatGPT from a black box with known endpoints into a partial window onto how labor markets respond to general-purpose AI.
Frequently Asked Questions
What are Intermediate Work Activities (IWAs)?
IWAs are standardized categories of work tasks drawn from O*NET, a U.S. Department of Labor database. They group occupation-specific tasks into broader work activities that span multiple jobs, enabling comparison of ChatGPT usage across occupational boundaries.
How does ChatGPT work-use differ from actual U.S. job composition?
According to OpenAI, the distributions differ substantially—ChatGPT users concentrate on activities like content creation and analysis, while U.S. employment includes significant physical tasks (materials handling, vehicle operation) that fall outside the tracked IWA categories.
What does 'work-use concentration declined' mean?
In Q1 2025, the ten most common work activities represented 56.0% of messages across 74 countries; by Q3 2026, this fell to 51.1%, indicating that users are diversifying their ChatGPT applications and countries are specializing in different task types.