Google Study Finds AI Use Broad but Shallow
Google’s analysis of 15 million Gemini interactions suggests that people are using AI for selected tasks rather than handing over entire jobs.
Topics
Google found that people are using artificial intelligence across a wide range of occupations, but usually for only a small share of the tasks within each job.
The first version of its ‘AI & Economy ATLAS’ study analyzed 15 million aggregated and de-identified interactions across the Gemini app, AI Mode and Gemini API. The data covered more than 150 countries, 140 languages, 800 occupations and 4,000 tasks.
Google described workplace AI use as “broad but shallow.” The products were used for work linked to every industry and 68% of occupations, which together account for about 90% of US employment. In a typical occupation, however, AI appeared in only about 21% of tasks.
Most work-related interactions involved assistance rather than full automation. People commonly used AI for research, learning, strategy and idea generation. Fewer than 10% of workplace interactions fully automated a task, Google said.
Non-routine cognitive work, including creative design and hypothesis testing, represented 65% of workplace AI interactions, compared with an estimated 35% of tasks across the broader economy.
The findings suggest that current AI adoption is “augmenting selected parts of jobs rather than replacing occupations in their entirety.” However, the study measures use of Google products and does not establish whether the interactions improved productivity, reduced employment or produced reliable outcomes.
Google also found that workers in manual and technical occupations were using AI for related tasks such as diagnosing equipment problems, interpreting test results and learning procedures. Users in those occupations were twice as likely to use multimodal tools involving images or video.
More than 86% of the interactions analyzed occurred outside work. Common uses included researching purchases, operating household equipment and navigating administrative processes such as taxes, licenses and fines.
Global adoption broadly tracked national income levels, raising the possibility that existing economic and digital divides may also shape access to AI. English accounted for about one-third of conversations, while some middle-income countries in South America and the Middle East recorded adoption rates comparable with wealthier economies.
Google said it removed personally identifiable and potentially sensitive information before aggregating the data.
The study does not include activity from products such as Google Workspace, Translate, AI Overviews or Gemini’s enterprise services, limiting its ability to represent the full scale of AI use across the economy.
