Google publishes ATLAS v1.0, a first large-scale map of how people use artificial intelligence in daily life and at work. What exactly are people doing with these tools? How much does that change the real economy? The report points to answers based on 15 million aggregated and de-identified interactions between users and products like Gemini App, AI Mode and the Gemini API.
What is ATLAS and why it matters
ATLAS stands for Activity, Task, Landscape, and Adoption Study. It's an ongoing, large-scale study that analyzes real AI use across more than 150 countries, 140 languages, 800 occupations and 4,000 tasks. The sample comes from product interactions that together reach over 1,000 million people per month.
Why should you care? Because it offers empirical evidence about how AI integrates into concrete activities, not just theoretical forecasts. That helps researchers, companies and governments make better-informed decisions.
Key findings
-
AI at work is broad but shallow. Adoption spans most sectors and 68% of occupations in the U.S., but in a typical job AI is used for only about ~21% of tasks.
-
Collaboration and assistance dominate, not full automation. Work interactions with AI tend to focus on ideation, strategy, information search and learning. Less than 10% of interactions fully automate a task.
-
It’s not just for office jobs. It's also used by manual and technical trades like automotive and industrial mechanics for real-time diagnostics, troubleshooting and instant learning. When these professionals use AI, they are 2x more likely to use multimodal capabilities (images or video).
-
Much of the value happens outside work. Over 86% of recorded interactions are in personal contexts: shopping, help using appliances, or tedious bureaucratic tasks like taxes and licenses. Many of these benefits don’t show up in traditional economic indicators.
-
Global adoption generally follows GDP per capita, but with exceptions. Some middle-income countries in South America and the Middle East use AI at rates close to wealthy countries. Also, English represents only ~1/3 of conversations, indicating use in native languages for complex tasks.
Concrete examples
-
A car technician uses multimodal AI to interpret complex test results, debug wiring and inspect part wear from images.
-
Someone at home uses AI to navigate tax procedures or understand an appliance manual, saving time and friction.
Privacy and methodology
ATLAS was built with multiple layers of privacy protection. Beyond removing personally identifiable information, Google describes automated processes that strip sensitive references, separate data from original records, summarize text and aggregate results by user groups.
Text transformations and the organization of unstructured text are handled by a tool called OCTO (Observation Clustering and Taxonomy Organisation), which helps convert open conversations into entities and categories useful for analysis.
What’s next and why it matters to you
This is just the beginning. ATLAS v1.0 opens the door to long-term tracking of how uses, tasks and economic effects evolve. Research will continue in collaboration with academics and other stakeholders.
It matters because AI doesn’t act like an automatic force that changes everything overnight. Its effects depend on how it’s adopted, who uses it and for what. Will your job be complemented or transformed? What policies could maximize benefits and reduce risks? These are practical questions ATLAS helps answer with data.
Also, there’s a lot of economically relevant use that ATLAS doesn’t fully cover: products with billions of interactions like Google Workspace and Translate, enterprise platforms and emerging capabilities like agent programming and world models.
In the end, the lesson is simple: understanding the AI economy requires real data and public conversation so the technology serves the many, not just a few. Which finding surprised you most from these first results?
Original source
https://blog.google/innovation-and-ai/technology/research/understanding-the-ai-economy