Google launches ATLAS v1.0, a large-scale study of how people actually use Gemini

Google published the first edition of its AI & Economy ATLAS on July 23, 2026, a study built from 15 million de-identified interactions across the Gemini app, AI Mode, and the Gemini API that aims to map, in granular detail, how people are really using AI rather than how they say they use it.

What's new

ATLAS — short for Activity, Task, Landscape and Adoption Study — draws on aggregated, privacy-preserving data spanning more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks. Google says the analysis is powered by "Google DeepMind's Observation Clustering and Taxonomy Organisation (OCTO), a tool for transforming massive unstructured text data" into structured categories of activity.

The topline findings temper some of the more alarmist automation narratives:

• Gemini usage touches roughly 70% of professions, and those professions account for about 90% of U.S. employment.
• Despite that reach, Google reports that "at work, most AI use is focused on collaboration and assistance with tasks, and so far task automation is uncommon."
• The average worker uses AI for only about a fifth of their tasks, and Google estimates less than 10% of Gemini interactions are aimed at automating non-routine cognitive work outright.
• The bulk of usage — Google puts it above 86% — happens outside of work entirely, in personal contexts rather than professional ones.

Google frames this as the first in an ongoing series rather than a one-off report, though it has not committed to a publication cadence for future editions.

Context

ATLAS lands alongside a wave of competing claims about AI's labor-market impact, from predictions of mass displacement to reassurances that adoption is shallow and additive. It's also a direct answer to a methodological problem: most prior data on AI use at work has come from surveys, which are self-reported and prone to both over- and under-stating actual use. By working from real, de-identified interaction logs at the scale of a billion-plus monthly users across Gemini's surfaces, Google is positioning ATLAS as a harder empirical baseline than the survey-driven studies that have shaped the debate so far.

It also doubles as a competitive move. Google has been racing OpenAI and Anthropic on model capability; ATLAS shifts part of that competition toward whoever can most credibly claim to understand how AI gets used in the real world — data no outside researcher can replicate without access to Google's own logs.

Why it matters

The finding that automation remains rare — even where Gemini touches the majority of a profession's tasks — cuts against the more dramatic predictions of near-term job displacement, at least as measured through Google's own products today. That matters for policymakers and employers trying to calibrate workforce plans: broad reach without deep task automation suggests AI is currently augmenting work more than replacing it, though Google itself cautions this is an early, fast-moving snapshot rather than a settled conclusion.

It also sets a template other AI providers may face pressure to match. If Google can publish usage-pattern data at this scale and granularity, comparable disclosures from OpenAI, Anthropic, or Microsoft would let researchers and regulators compare real-world AI impact across providers rather than relying on each company's own framing.