14.65M AI conversations analyzed
~3x purchase questions vs shopping time
~20x government-services gap, the largest
~5.8x education-topic gap

What happened

Google and DeepMind compared 14.65 million non-work US conversations — pulled from the Gemini app, AI Mode in Search, and the Gemini API between April 6 and 19, 2026 — against the American Time Use Survey, which tracks how Americans actually spend their days. The question: which topics do people ask AI about more than the time they spend living them?

Consumer purchases came in at roughly three times overrepresented — people ask AI about buying things about 3x more than they spend time shopping. Other gaps run wider: government services and civic obligations near 20x, professional and personal-care services above 7x, education around 5.8x. Eating and drinking flipped the other way — people spend vastly more time doing it than asking about it. And about half of the high-friction questions — medical, legal, money, government — arrived outside working hours.

One thing the report does not contain: click data. Google says nothing about whether any of those purchase conversations sent a single visitor to a website.

How much more people ask AI about a topic than time they spend on it
Government services
Professional & personal care
Education
Consumer purchases

Why this matters

This is Google quantifying its own surface, and the shape it reveals is the one ecommerce operators have been squinting at all year: the research phase of buying is migrating into AI conversations, and it's overrepresented there relative to actual shopping behavior. The gap categories tell you why — people take confusing, high-friction decisions to an assistant. Considered purchases live in exactly that zone.

Now the skepticism, because it's earned. Two weeks of data. Google measuring Google. And conversations-per-topic versus time-spent is an odd ratio — a thirty-second question isn't equivalent to an hour at the mall, so 3x overstates nothing and proves nothing by itself. Direction is what you should take from this, not magnitude. The direction matches every other signal: assistants are absorbing the questions that used to become search sessions and category-page visits.

Here's the uncomfortable arithmetic for a store owner. If purchase research happens in AI at triple the rate of shopping and most of it produces no click for anyone, the traffic you're losing to AI answers isn't going to a competitor — much of it is going nowhere. The only distribution that survives that math is being the source the answer quotes. That's a product-data and citability problem, not a rankings problem.

What to do about it

Run your own category through the assistants

Ask ChatGPT, Gemini, and AI Mode the five questions your buyers actually ask before purchasing. Record which brands and domains get named or cited. If competitors appear and you don't, you now have the gap this study predicts — measured on your own category, not Google's aggregate.

Make your product data quotable

Assistants cite sources they can extract cleanly. Ship Product schema with price, availability, and review ratings on every product page, and keep specs in real HTML — not images, not tabs that render empty to a crawler. This is the plumbing behind every citation you'll ever earn.

Publish the selection content assistants lean on

The overrepresented questions are situational — which option fits this budget, this use, this constraint. Write comparison and buying-guidance pages that answer them directly, with concrete numbers and honest tradeoffs. Generic category copy never gets quoted; specific answers do.

Segment AI referrals in GA4

Build a channel view for sessions arriving from AI surfaces. The click stream that does exist is small but arrives pre-researched, so track its conversion rate separately — it's your evidence for how much this shift is worth to your store.