Google study finds AI used across 70% of jobs but minimal automation

Artificial intelligence tools have reached workers across a remarkably broad swath of the economy, but they're not yet automating jobs at the scale many feared or predicted, according to new research from Google analyzing nearly 15 million real-world interactions with its Gemini AI assistant.

The study, called ATLAS (Activity, Task, Landscape and Adoption Study), examined 14.65 million de-identified interactions across Gemini's consumer app, Google's AI Mode search, and the Gemini API over a two-week period in April. The findings paint a picture of widespread but shallow adoption: AI touches many jobs without fundamentally transforming how most work gets done.

Why it matters

This research provides concrete data on the gap between AI's theoretical capabilities and its actual deployment in workplaces. While speculation about job displacement continues, the evidence shows workers are primarily using AI as a collaborative tool for research, drafting, and troubleshooting rather than handing over complete tasks. For business leaders evaluating AI investments, this suggests the technology's near-term value lies in augmentation rather than replacement.

Broad reach, limited depth

Scott Strand, an economist on Google's Technology and Society team, said Gemini usage covered 70% of job categories, representing approximately 90% of U.S. employment. However, the average worker used AI for only 21% of their tasks, and fewer than 10% of Gemini interactions appeared designed to automate non-routine cognitive work.

The methodology involved automated classification systems that sorted interactions into work-related and non-work categories, then mapped work activities to 4,000 tasks across 800 occupations in 150 countries and 140 languages. Non-work tasks were categorized using frameworks from the Bureau of Labor Statistics' American Time Use Survey.

These findings align directionally with earlier research from Anthropic and OpenAI, though Anthropic reported somewhat higher automation levels in its studies.

Income and geography drive adoption

The data revealed clear patterns in who uses AI most intensively. Workers in higher-paid occupations showed significantly greater AI adoption: a 1% increase in an occupation's median earnings correlated with a 2.68% increase in AI use. Geographic wealth also played a role, with workers in wealthier regions using the technology more frequently.

Researchers found unexpected adoption among blue-collar workers, including electricians seeking wiring diagrams and auto repair technicians looking for engine maps. These workers particularly leveraged multimodal AI capabilities that process images and video, according to Strand.

Consumer use dominates

Non-work interactions accounted for 86% of Gemini conversations, a higher proportion than researchers anticipated. Consumers used the AI not only for routine tasks like cooking and cleaning but also to navigate government services and bureaucratic processes.

Study limitations

The research excluded data from business-focused products like Gemini Enterprise and Google Workspace because the company doesn't maintain logs for those services. This gap means the study may undercount automation in enterprise settings where AI adoption could be more advanced.

Google plans to make ATLAS the first in a series of reports tracking AI adoption over time, though the company hasn't determined the publication frequency for future updates.

The details were first reported by Axios.