Noticia
Google’s ATLAS maps how people are using AI across apps
Google’s new ATLAS study examines 15 million AI interactions across more than 150 countries, revealing how people use AI for assistance, learning and everyday tasks.
Google has launched ATLAS, a research project designed to show how people are using artificial intelligence in everyday life and at work. Unlike a product announcement focused on a new feature, the project examines real interactions with Google’s AI services and places them in a wider economic context. The company’s full ATLAS announcement describes the first dataset as a large-scale, de-identified study rather than a forecast of what AI users might do in the future.
ATLAS v1.0 is built from 15 million aggregated human-AI interactions across the Gemini App, AI Mode and the Gemini API. Google says those three services together are used by more than one billion people each month. The dataset covers more than 150 countries, 140 languages, 800 occupations and 4,000 tasks, giving the project a distinctly international scope.
Assistance still dominates workplace use
The study’s clearest finding is that AI is spreading across work, but it is usually being used selectively. Google says AI activity touches 68% of occupations that collectively account for 90% of total employment in the United States. Within a typical job, however, AI is used for roughly 21% of tasks rather than across the entire role.
Most workplace interactions fall into collaborative categories such as ideation, strategy, information retrieval and learning. ATLAS also records a higher share of non-routine cognitive work, including creative design and hypothesis testing, than the economy as a whole: 65% of AI work interactions in this category compared with 35% in the wider baseline. Fully automated tasks remain uncommon, representing less than 10% of workplace interactions in the dataset.
That distinction matters. The figures do not show that AI has automated 10% of jobs, nor do they measure whether a particular task was completed successfully. They describe the type of help people request from Google’s systems. The picture that emerges is closer to an always-available collaborator than to a system running an entire job without supervision.
Mobile apps are becoming everyday problem-solving tools
More than 86% of the interactions in ATLAS take place outside work. Google says people use its AI tools for activities such as researching purchases, learning how appliances and tools work, and navigating administrative tasks involving taxes, licences and fines.
For smartphone and app users, this is one of the report’s most relevant signals. The Gemini App and AI Mode are consumer-facing services that people can reach while moving between messages, searches and other mobile tasks. ATLAS does not claim that every interaction happened on a phone, so it would be wrong to treat the dataset as a direct measurement of smartphone usage. It does show, however, that conversational AI is being used for practical decisions well beyond the office.
That broad use also helps explain why mobile AI is moving from a novelty into a layer that sits across search, communication and daily administration. When users ask for help comparing a purchase, understanding a device manual or completing a complicated public-service process, the value comes from reducing friction rather than replacing an entire workflow.
AI is not limited to desk jobs
ATLAS also challenges the idea that AI adoption belongs mainly to office-based professionals. Google reports that people in manual and technical occupations, including automotive technicians and industrial mechanics, use conversational AI for live diagnostics, troubleshooting and on-the-fly learning.
When workers in these fields use Google’s AI tools, they are twice as likely to use multimodal capabilities. That can include asking an AI system to interpret images or video alongside text. In the examples cited by Google, multimodal interactions can help inspect machinery, interpret test results or debug electrical wiring. These findings do not establish that AI is safer or more accurate than an experienced professional, but they do show how a camera-equipped phone can become part of a practical field workflow.
A global picture with a digital-divide warning
The data spans more than 150 countries and territories representing 99% of the world’s population, according to Google. English accounts for only about one-third of conversations, suggesting that users do not universally abandon their native languages when handling complex tasks.
At the same time, AI usage per person broadly follows relative GDP per capita. Google identifies exceptions in parts of South America and the Middle East, where adoption rates can resemble those of higher-income countries. The pattern points to a familiar risk for mobile technology: access to capable devices, affordable data and useful local-language services may determine who benefits first.
What ATLAS can and cannot prove
Google says it built ATLAS with several privacy protections. Personally identifiable information is scrubbed, possible references to sensitive information are removed, links between the de-identified dataset and underlying user logs are removed, and text is summarized and aggregated into groups representing multiple users.
The project is still limited in scope. It only covers the Google products and tools included in this first release. Google notes that services such as Workspace, Translate, AI Overviews, enterprise platforms and some advanced capabilities are not yet reflected. ATLAS should therefore be read as an early view of how people use Google’s AI ecosystem, not as a census of all AI activity.
That caveat does not make the research irrelevant. It makes the methodology easier to interpret. The report’s value lies in documenting how AI is being used today: mostly for assistance, learning and practical problem-solving, across many languages and occupations. For smartphone and app developers, the message is straightforward. The next phase of mobile AI will be judged less by how impressive a demo looks than by whether it helps people complete ordinary tasks with useful context, clear limits and appropriate privacy safeguards.