AI will reshape the global economy. Most people agree on that. What’s far less clear, however, is exactly how — and what it means for jobs, daily life, and long-term growth. Those outcomes aren’t guaranteed. In fact, they depend entirely on the choices we make. And to make good choices, we need good evidence.
Takeaways 🚀
✅AI use grows across work
✅ATLAS tracks real-world adoption
✅Most use supports daily tasks
✅Automation remains limited so far
✅Global patterns reveal digital divides
That’s precisely why Google launched the AI & Economy ATLAS — short for Activity, Task, Landscape, and Adoption Study. It’s an ongoing, large-scale study of how people use Google’s AI tools at work and at home. All data is aggregated and de-identified.
What Is ATLAS?
At its core, ATLAS draws from real-world AI use. Version 1.0 covers 15 million human-AI interactions across the Gemini App, AI Mode, and the Gemini API. Together, these tools reach more than one billion users every month.
The scope is massive. Specifically, ATLAS v1.0 covers:
- 150+ countries
- 140 languages
- 800 occupations
- 4,000 tasks
That makes it the most comprehensive study of real AI use to date. It’s also, notably, an early look at a fast-moving field. AI keeps improving. How people use it keeps changing. And the research tools needed to track all of this are still catching up. As a result, ATLAS will continue to grow and improve over time.
What We’re Learning From ATLAS v1.0
So, what does the data actually show? Here are the most important findings so far.
AI Use at Work Is Wide but Thin
First, AI has reached every industry. It shows up in 68% of all occupations — roles that make up around 90% of total U.S. employment. However, within any given job, people use AI for only about 21% of their tasks. In short, many people use AI, but only for a small slice of their workday.
Collaboration Beats Automation
Beyond sheer adoption, it’s worth looking at how people actually use AI at work. The answer, for the most part, is collaboratively. People use it for brainstorming, strategy, research, and learning. Creative and complex tasks — what ATLAS labels “non-routine cognitive” work — appear in AI interactions at a much higher rate than in the broader economy (65% vs. 35%).
Full automation, on the other hand, remains rare. Fewer than 10% of interactions let AI complete a task on its own. For now, AI helps people do their work — it doesn’t replace them.
It’s Not Just Office Workers
It’s also important to note that AI isn’t only for desk jobs. Workers in skilled trades use it too. Auto technicians and industrial mechanics, for example, use AI to interpret test results, debug wiring, and check machinery for wear — treating it as a real-time collaborator on the job.
What’s more, these workers are twice as likely to use multimodal AI, meaning tools that process images and video rather than text alone.
Most AI Use Happens Outside Work
Shifting focus from the workplace, another striking finding is this: more than 86% of AI interactions in ATLAS take place outside the workplace entirely. People use AI to research purchases, troubleshoot appliances, and navigate government services like taxes, licensing, and fines. This kind of value is difficult to capture through standard economic metrics — but it’s real, widespread, and significant.
Global Adoption Tracks Wealth — With Exceptions
Finally, zooming out to the global picture, AI has spread broadly. ATLAS captures activity in over 150 countries, covering 99% of the world’s population. Notably, English accounts for only about one-third of all AI conversations, and people tend to stick with their native languages even for complex tasks.
In general, AI adoption follows GDP per capita. Wealthier countries tend to use more AI per person, which points to a real risk of a growing digital divide. That said, there are meaningful exceptions. Several middle-income countries in South America and the Middle East are adopting AI at rates that rival far wealthier nations.
How ATLAS Was Built
To power ATLAS, Google DeepMind built a tool called OCTO (Observation Clustering and Taxonomy Organisation). Essentially, OCTO takes large volumes of unstructured text — like AI chat logs — and organizes them into clear, meaningful categories.
Throughout this process, privacy was a top priority. The team scrubbed all personally identifiable information and added several extra layers of protection. These steps included removing references to sensitive topics, cutting all links to underlying user logs, summarizing conversations, and grouping summaries so that no individual user can be identified.
What’s Next
As thorough as ATLAS v1.0 is, it’s really just the beginning. AI capabilities keep expanding. People keep finding new ways to use them. And research methods continue to improve. In response, Google’s AI & Economy Research Program is already working with academic and other researchers to tackle the many questions that still remain.
It’s also worth noting that ATLAS doesn’t yet capture the full picture. It doesn’t include widely used tools like Google Workspace, Google Translate, AI Overviews, or enterprise platforms like Gemini for Google Cloud — all of which reach billions of users and will be part of future research.
As ATLAS expands, the goal is to build a sharper, more complete view of how AI is reshaping the economy. Ultimately, the hope is to help researchers, policymakers, businesses, and workers make better, more informed decisions along the way.
