Skip to content
Main Site News Console

ChatGPT's Ever-Expanding Work Landscape

· OpenAI Translated
OpenAILLM

October 26, 2026

OpenAI Economic Research is releasing data to enable researchers to delve deeper into questions about AI and the workforce.

Explore and download the data

Today, we are releasing OpenAI Signals data covering nearly 100 countries and 165 work activities, offering insights into how consumers use ChatGPT on the job. This data opens new avenues for researchers exploring AI’s role in the economy and provides fresh visibility into the variety of work activities people perform with ChatGPT—from drafting instructional materials and creating marketing content to writing code and analyzing business data.

Researchers can use this data to examine cross-country differences in work-related ChatGPT usage, track changes over time, and merge it with other economic indicators. This release maps ChatGPT usage to 165 work activity categories known as Intermediate Work Activities (IWAs)—a standardized task taxonomy used to categorize the types of work performed across occupations. The data is reported quarterly, spanning from Q1 2025 through Q3 2026.

Understanding how people use AI for work

Our new data draws from O*NET, a standardized framework that aggregates occupation-specific tasks into broader work activity categories applicable across occupations. Utilizing a unified framework helps researchers benchmark ChatGPT usage against other forms of occupational and employment data.

The interactive chart below lets users explore how these work activities shift over time within specific countries, spanning from high-level activity groups down to granular tasks. The data reveals substantial variation across countries in how ChatGPT is used for work.

Mapping this data to IWAs shows that work-related usage has become less concentrated among the most common activities. This suggests that, over time, countries are specializing their work usage—and doing so across different types of work. Such variations may reflect differences in the underlying work being done in each country or disparities in access and adoption of AI tools. Exploring these questions is just one of many research directions enabled by this new data.

Across the 74 countries observed in every quarter, the top ten most common activities accounted for an average of 51.1% of classified work messages per country in Q3 2026, down from 56.0% in Q1 2025.

How researchers can use the data

Combining ChatGPT usage data with IWAs allows researchers to compare ChatGPT messages to the actual activities performed across jobs in the US. The comparison visualization below contrasts how people use ChatGPT with the activities performed across all US jobs. This contrast highlights where ChatGPT adoption at work aligns—or diverges—from the work actually carried out across the US economy. These two distributions differ substantially, largely because roughly 43% of the employment-weighted task mix in matched US occupations falls into “Other IWAs”—activities not reported in separate categories. These include moving materials, driving vehicles, and maintaining health or medical records. These excluded categories did not meet our threshold for consistent representation in the initial US ChatGPT sample.

Select an activity to compare its share of ChatGPT usage with its share across the overall US economy.

Exploring the data

This release builds on OpenAI Signals—our ongoing effort to share empirical insights into how people use AI. The data reflects work-related messages from individual consumer ChatGPT plans and excludes enterprise messages or Codex usage. We encourage researchers to explore this data, combine it with other sources, and delve deeper into what these patterns mean for workers and businesses.

To protect user privacy, researchers do not read individual conversations. Automated systems classify messages only after removing identifying details. We publish exclusively aggregated statistics, suppress results for groups with small sample sizes, and introduce statistical noise to protect the privacy of individual messages.

Citation

BibTeX

@online{openaiecon2026signalsdata,
author = {Aaron Chatterji and Thomas Cunningham and David J. Deming and Zoe Hitzig and Drew Johnston and Alex Martin Richmond and Christopher Ong and Carl Yan Shan and Kevin Wadman},
title = {{OpenAI} Signals v3.0},
year = {2026},
organization = {OpenAI},
url = {https://openai.com/signals/data-download/},
urldate = {2026-10-26},
note = {Data download and methodology},
}