Google study: AI use at work is widespread, but automation remains limited
Kyiv • UNN
A new Google ATLAS study found that AI covers 70% of professions, but workers use it for only 21% of tasks. Less than 10% of interactions with Gemini are aimed at automating non-routine cognitive work.
The use of AI has spread across a wide range of professions, but it has not yet become the job automation force that many workers fear, according to a new Google study that covered nearly 15 million interactions with Gemini, writes UNN citing Axios.
Details
The report offers a broad overview of real-world AI usage, highlighting the gap between work that could be affected by AI, work where employees use AI, and tasks that are being automated.
Google found that people mostly use Gemini as a partner — for research, drafting, iteration, troubleshooting, and learning — rather than delegating entire tasks to it.
"AI adoption is very, very broad, as it spans a huge range of professions," said Scott Strand, an economist on Google's Technology and Society team.
Gemini usage covered 70% of professions, Strand said, adding that the professions where AI is used account for 90% of U.S. employment.
At the same time, he noted that AI usage is also "very shallow," as the average worker uses it for only 21% of tasks, with less than 10% of Gemini interactions apparently aimed at automating non-routine cognitive work.
This is similar to previous studies by Anthropic and OpenAI, although Anthropic reported a higher level of automation.
The report, known as ATLAS — Activity, Task, Landscape and Adoption Study — analyzed 14.65 million anonymous interactions in the Gemini app, the Google AI Mode search interface, and the Gemini API over two weeks in April.
Automated systems classified interactions as work-related or non-work-related, then mapped work-related activities to 4,000 tasks across 800 professions in 150 countries and 140 languages.
Non-work tasks were mapped to categories used in the Bureau of Labor Statistics' American Time Use Survey.
The study did not account for how people use business-oriented tools such as Gemini Enterprise and Google Workspace, as the company does not keep logs for those products. It is possible that this additional data could show a higher level of automation, the publication notes.
"We would obviously like to use that data if we could," Strand said. "For us, it wasn't even an option."
Workers in high-paying jobs and wealthier regions use AI more, the study showed.
For example, Google found that a 1% increase in median earnings by profession is associated with a 2.68% increase in AI usage.
The survey found more AI usage by blue-collar workers than researchers expected, including electricians looking up wiring diagrams and auto repair workers searching for engine maps.
"Workers tend to use a lot of what we call multimodal AI, that is, AI with images and video," Strand said.
Consumer usage was also broader than expected: Google found that non-work interactions accounted for 86% of conversations in Gemini.
Consumers used Gemini not only to speed up tasks like cooking and cleaning, but also to navigate government services and processes.
The open question is whether such studies can help predict where AI usage will go, or whether they are primarily snapshots of how people have already used the technology, the publication notes.
Google expects ATLAS to be the first in a series, although the company has not yet decided how often it will publish updates.