Google has put a map of AI usage on the table, and the result is less dramatic than launch-event language.
On July 23, Google released AI & Economy ATLAS v1.0, based on 14,653,926 de-identified interactions from Gemini App, Google AI Mode and Gemini API between April 6 and April 19. The team mapped those interactions to more than 800 occupations, 4,000 work tasks, 300 household activities, 150 countries and 140 languages.
Broad reach, shallow depth
ATLAS says workplace use appears across 68% of US occupations, representing just over 88% of US employment. But in the median occupation with observed AI use, AI touches only about 21% of tasks. Only 3% of occupations show usage across more than 75% of tasks.
Google calls ATLAS the "most comprehensive look to date at how real people are using AI at scale." In plainer terms, the company is trying to replace automation slogans with usage data.
Most use still looks collaborative
For non-routine cognitive work, end-to-end automation intent accounts for less than 10% of observed AI conversations. The heavier clusters are drafting, review, ideation, strategy, retrieval and learning.
The safer conclusion is that AI has entered many workflows, but mostly as a collaborator. It helps people get out of blank pages, search friction and low-level troubleshooting. It has not yet become a universal independent work station.
Manual work is part of the story
Where manual and technical trades use AI, multimodal use is more than 2x the overall work baseline. Google gives examples such as auto technicians and industrial mechanics using AI to interpret test results, debug wiring and inspect equipment wear.
Home use may be economically invisible
More than 86% of ATLAS interactions occur outside work. Mapped to time-use categories, they span activities covering about 98% of Americans' waking hours. Google estimates that if household AI use saves only 30 minutes a week on average, unpaid productivity gains in the US could be worth roughly $100 billion.
Globally, usage correlates with national wealth: a 1% rise in GDP per capita is associated with about a 0.9% rise in usage, while the lowest-adopting country quintile accounts for only 2% of conversations. English is only about one third of global conversations.
The map has limits
Google says ATLAS captures what people directly do with AI, not whether final output improves. It also leaves out Workspace, Translate, AI Overviews, Gemini Enterprise, Cloud platforms, coding agents and world models.
The blog says, "This is just the beginning." The useful part of the first edition is exactly that restraint. It gives measurable coordinates: how many jobs are touched, how many tasks are reached, how much use looks like automation, and where language or country gaps remain.
Sources: Google official blog, Google ATLAS v1.0 report, Axios, CocoLoop and Fox Business; checked the 14,653,926-interaction sample, source surfaces, occupation and task mapping, 21% task reach, under-10% automation intent, 86% non-work usage, and country and language metrics.