Google Publishes ATLAS Report: AI Touches 68 Percent of Jobs but Automates Less Than 10 Percent of Tasks

Google has published the first edition of the ATLAS report, the largest study to date of how people actually use artificial intelligence at work and in their personal lives. The findings are specific: AI has already reached two-thirds of occupations in the economy, but in a typical job it covers only about a fifth of the tasks performed, and full automation of work remains rare.

What the report shows

The report is called ATLAS, short for Activity, Task, Landscape and Adoption Study. Its authors are Zanna Iscenko, head of the AI & Economy team at Google, and Scott Strand, head of StratOps, along with fifteen other researchers from Google and Google DeepMind. The analysis used Google DeepMind's OCTO tool, which categorizes AI conversations by industry, occupation and specific task, while stripping personal data and sensitive information before the results are aggregated.

The biggest surprise is not the reach of AI itself, but how shallowly it cuts into the structure of individual jobs. The company stresses that fewer than 10 percent of work-related interactions end in full task automation, and that the average worker uses AI to handle only part of their duties, not an entire occupation.

AI as collaborator, not replacement

According to Google, most work-related interactions involve supportive tasks such as generating ideas, developing strategy, searching for information and learning new skills. The pattern that emerges from the data looks more like collaboration between a person and a tool than an algorithm replacing a worker.

The report highlights a surprising group of users: auto mechanics and industrial technicians, who reach for multimodal AI features, meaning photo and video analysis, twice as often as other workers, using them for real-time fault diagnostics and for learning to operate complex machinery.

Personal life dominates over work

Google's data shows that the vast majority, more than 86 percent, of all interactions with AI tools have nothing to do with professional work. People use Gemini and similar tools primarily in everyday life, for household tasks, planning, learning and entertainment, and only secondarily for job duties.

The report also shows a clear correlation between a country's wealth and the intensity of AI use: a 1 percent increase in GDP per capita is associated with a 0.9 percent increase in the number of AI conversations. Exceptions to this pattern include some middle-income countries in South America and the Middle East, where AI adoption is higher than GDP alone would suggest.

Earnings and the scale of the data

Google also provides data on the earnings of users who use AI at work: the median salary of the American workers covered by the study, weighted by employment, is $62,252 a year, but when weighted by the number of AI conversations conducted, it rises to $82,919. This means that people who earn more and perform more complex intellectual tasks hold proportionally more conversations with AI than lower-paid workers.

The scale of the study itself is unprecedented for this kind of labor market analysis: nearly 14.7 million interactions is far larger than the samples typically used in surveys or economic research based on respondents' self-reported answers. Google stresses that the data comes from actual tool use rather than questionnaire responses, which it says makes the findings more reliable than earlier forecasts about job automation.

What it means for companies and workers

For Polish companies and workers, the ATLAS report offers an argument against the most alarmist scenarios of mass job losses to AI, though it does not deny that work is changing. If even in the economy most advanced in AI adoption automation covers individual tasks rather than entire occupations, the key skill becomes weaving AI tools into existing duties rather than waiting for them to be replaced outright.

Google says ATLAS will be published on a recurring basis, allowing it to track how the depth of AI use in individual occupations changes in the coming months. The company has also released its full methodology and data as a research paper, opening it up for analysis by independent economists, including project collaborators Diane Coyle of the University of Cambridge and David Autor of MIT.