Google just published its first big look at how people actually use its AI assistant Gemini at work, and the findings quietly contradict a lot of the industry noise about AI wiping out office jobs.
The study, called ATLAS, examined 15 million anonymous conversations with Gemini across its app, its AI-powered search feature, and its developer tools. Researchers matched these conversations against official government job classifications covering more than 800 occupations in over 150 countries. Their conclusion: AI touches a huge number of jobs, but barely scratches the surface of what any single job actually involves.
About 7 in 10 occupations showed some Gemini use, and those jobs together make up nearly 9 in 10 of all U.S. employment. That sounds like AI is everywhere. But when researchers zoomed into individual jobs, workers were only using AI for around 21 out of every 100 tasks that make up their normal workload. And fewer than 1 in 10 of those work-related conversations were attempts to hand an entire task over to AI from start to finish. Most people were using it as a thinking partner, not a replacement.
This matters because it directly challenges the story some companies have been telling investors and employees. Over the past year, firms including Amazon, Salesforce, and several others have pointed to AI when announcing layoffs, framing job cuts as a natural result of AI making work more efficient. Google's own data suggests that, at least for how everyday workers are using its tools, that kind of full automation is still rare.
There is an important gap in the study worth noting. Google left out data from Workspace, Gemini for Google Cloud, and Gemini Enterprise, which are exactly the paid business tools that large companies use to automate internal processes. This is where a lot of the aggressive automation and job cuts are likely happening. So the study may be showing a rosier picture of individual AI use while missing the corporate side where AI is being built directly into workflows to replace tasks entirely.
A second finding worth paying attention to is who is using AI the most. Higher paid workers in wealthier regions use it more often. Google found that jobs paying more tend to show noticeably higher AI usage, which suggests the tool is currently helping people who already have leverage in the job market, not replacing people at the bottom.
The study also found unexpected use among blue collar workers. Electricians were using Gemini to pull up wiring diagrams, and auto mechanics were using it to look up engine specs, often using photos and video rather than text.
For business owners, the practical read is this: AI is spreading fast as a helper, but the jobs actually vanishing right now are more often a management decision dressed up as an AI story, not proof that software can run a department by itself. The tools are useful today for speeding up drafting, research, and troubleshooting. Betting on full automation of a role based on hype alone is still getting ahead of what the technology is actually doing in practice.