Google has launched the first iteration of its AI & Economy ATLAS (Activity, Task, Landscape, and Adoption Study), an ongoing, large-scale, de-identified study of how people are using its AI products and tools. It draws on 15 million aggregated and de-identified human-AI interactions across the Gemini app, AI Mode, and the Gemini API, which together are used by more than one billion people monthly. The report was published July 23, 2026.
What ATLAS measured and how
The dataset covers 14,653,926 interactions sampled from the Gemini app, AI Mode, and the Gemini API between April 6 and April 19, 2026. ATLAS v1.0 insights span more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks.
The report lists seventeen authors across Google and Google DeepMind, with Zanna Iscenko and Scott Strand as corresponding authors. Iscenko is AI & Economy Lead in Google’s Chief Economist’s Office; Strand is Head of Strategic Operations and Special Projects for Technology & Society at Google.
The underlying classification work was performed using Google DeepMind’s Observation Clustering and Taxonomy Organisation (OCTO), a tool for transforming massive unstructured text data into organized entities.
The current dataset does not include usage from several other Google products, including Workspace, Google Translate, AI Overviews, Gemini for Google Cloud, and Gemini Enterprise, meaning workplace adoption may be even broader than the findings capture.
Broad reach, shallow depth
Scott Strand described AI adoption as “Adoption of AI is very, very broad in that it touches a huge range of occupations,” noting that Gemini usage covered 70% of jobs and that those jobs represent 90% of U.S. employment. At the same time, he said AI use is also “very shallow,” with the average worker using it for only 21% of tasks, and less than 10% of Gemini interactions seemingly geared at automating non-routine cognitive work. The formal report figure is 68% of all occupations, collectively representing 88.4% of total U.S. employment.
Rather than automating entire occupations, Google’s researchers found that AI is typically used to assist with individual tasks. The vast majority of AI interactions at work focus on collaborative uses such as ideation, strategy, information retrieval, and learning.
Tasks like creative design and hypothesis testing, categorized in ATLAS as “non-routine cognitive,” show up in AI work interactions at a much higher rate than in the economy as a whole (65% versus 35%).
The study found more use by workers in blue-collar jobs than researchers anticipated, including electricians seeking wiring diagrams and auto repair workers looking for engine maps. This challenges assumptions that AI software tools remain exclusively white-collar, pointing instead to a wider spread across the app’s installed base.
Most usage happens outside the office
More than 86% of AI interactions in the dataset occurred outside the workplace. People were using AI for household activities such as researching purchases, operating appliances, and completing administrative tasks related to taxes, licenses, fines, and government services.
High-friction administrative tasks feature prominently, including navigating government services such as taxes, licensing, and fines. Google’s point is that this kind of value often does not show up in official economic statistics like GDP, so the real benefit of AI may be larger than the numbers suggest.
The finding carries implications for how policymakers and economists assess the productivity gains delivered by mainstream AI apps, an area where prior research has already flagged significant unmeasured value from widely used free digital tools.
Global spread and language patterns
On a per-capita basis, AI usage closely mirrors a country’s relative level of wealth, though some middle-income countries in South America and the Middle East are adopting AI at rates comparable to higher-income countries. ATLAS data shows AI usage in over 150 countries and territories.
English accounts for only about a third of global conversations, and users do not show signs of systematically abandoning their native languages for complex tasks. According to the report, Spanish follows as the second most common language, with Arabic, Portuguese, and Turkish also among the leading non-English languages.
Google confirmed that ATLAS will operate as a multi-year research program tracking AI’s macroeconomic trajectory. Google expects ATLAS to be the first in a series, though the company has not yet decided how often it will publish updates. The full report is available via Google’s official blog and as a primary research paper. For developers and businesses building on top of Gemini’s API, the data provides a baseline for understanding where real-world AI usage patterns diverge from theoretical projections of automation and displacement.
