What People Actually Ask AI About (vs. What They Actually Do)

2026-07-25 — business growth strategy India

I was looking at the numbers from Google's new ATLAS report the other day. It's a weird dataset. 15 million AI interactions cross-referenced against the American Time Use Survey. Nothing fancy. Just a comparison between what people ask Gemini about versus what Americans actually do with their time.

The gap is wild.

People ask about government paperwork roughly 20 times more often than they deal with it. Licenses, taxes, fines, voting. All of it. It's the biggest divergence in the data.

Professional services—doctors, lawyers, banks, salons—come up seven times more frequently in conversations than people actually use them. Education runs six to one. Shopping is about three to one. Actually, that one tracks for me. You ask a bunch of questions before you buy something, not after.

Then the data flips.

People spend roughly 18 times more of their actual day eating and drinking than they ask Gemini about it. Watching TV, dressing, cleaning, making meals—these are the stuff of actual human time, but AI barely touches them. They're the background tasks. The friction-free stuff.

What's happening here is slightly different from what consultants usually say when they quote this data. Most get hung up on the "people use AI for high-friction tasks" angle—which is true—but miss something else. People ask about the things that scare them or confuse them, not the things they're already good at.

Government paperwork doesn't come up 20 times more because people want government paperwork optimized. It comes up because government paperwork terrifies people. Same with doctors and legal things. Same with money.

The activities you spend most of your time on—eating, cleaning, watching things, dressing—you already know how to do. They don't need optimization. They're solved problems. Asking an AI about them would be like asking an AI how to walk.

Actually, that's not quite right.

People do ask Gemini about cooking, and they use it to complete cooking and cleaning tasks more quickly. But the scale is tiny relative to how much time these activities actually consume. The gap is still 18 to one in the other direction. What this probably means is that AI is still figuring out where it's useful for routine stuff. For now, it's finding its footing in the hard problems—the ones that make people reach out.

The data covers interactions from April 2026, so it's a snapshot of a moment. Google hasn't said yet how often it will update ATLAS, which means we won't actually know if these patterns shift or just entrench themselves. That's worth remembering before you read too much into trend lines.

What clients keep missing is that this isn't a use-case opportunity list. It's a map of human anxiety paired with AI capability. The gap isn't a bug to fix. It's probably a feature of how people think about new technology.

The stuff that scares us gets optimization first.