Google's ATLAS Study Finds AI at Work Is Broad but Shallow
Google's first ATLAS report, drawn from 15 million interactions, shows AI reaching most jobs while still touching only a slice of daily tasks.
What Google's ATLAS study looked at
Google is releasing the first report from ATLAS, short for Activity Task Landscape Adoption Study, a new ongoing research effort examining how people use the company's AI products in daily work and life. The v1.0 dataset behind the report is built from 15 million aggregated, de-identified human-AI interactions across the Gemini app, AI Mode, and the Gemini API. Google frames ATLAS as an early view of a fast-moving landscape, not a finished picture, and stresses that the dataset reflects use of its own AI products rather than a neutral cross-section of every AI tool in the economy. In scale, ATLAS v1.0's insights span more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks, which Google calls the most comprehensive look yet at how people are using AI at scale.
What the data shows
The headline finding is that AI use at work is broad but shallow. Workplace adoption spans effectively every industry sector, reaching 68% of the occupations Google's researchers tracked. But inside a typical job, people are selective about where they apply it: AI shows up in only about 21% of tasks within that job. Most of that use is collaborative rather than automated. Google's researchers describe the bulk of workplace interactions as ideation, strategy work, information retrieval, and learning, with fewer than 10% of those interactions fully automating a task outright. The report also notes that AI use is not confined to office work: workers in physical and technical trades are using conversational AI as a live collaborator for real-time diagnostics and troubleshooting, alongside more conventional knowledge-work uses.
Why the framing matters
Because the dataset comes entirely from Google's own products, it cannot tell us how people use AI tools from other companies, or how representative Google's user base is of the wider working population. Google itself flags this as an early, evolving picture rather than a definitive economic survey. Still, the scale of the underlying data, and the finding that most AI use today augments narrow slices of work rather than replacing whole jobs, gives researchers and policymakers a concrete data point to weigh against more speculative predictions about automation.
Key facts
- Google's ATLAS v1.0 report draws on 15 million aggregated, de-identified interactions across the Gemini app, AI Mode, and the Gemini API.
- ATLAS v1.0's insights span more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks.
- AI use at work reaches 68% of occupations tracked, but is applied to only about 21% of tasks within a typical job.
- Fewer than 10% of AI interactions at work fully automate a task; most are collaborative assistance.
- The dataset reflects usage of Google's own AI products, not a neutral sample of the whole economy.
Check your understanding
- What share of occupations does Google say use AI at work, according to the ATLAS report?
- Why does Google caution that the ATLAS dataset is not a neutral sample of how the whole economy uses AI?