Search Engine Journal published an analysis on July 25 that lines up conversation topics from Google’s AI products against how Americans actually spend their non-work hours, and the two data sets rarely agree. Matt G. Southern, writing for Search Engine Journal, built the comparison on top of AI and Economy ATLAS v1.0, a report Google and Google DeepMind released on July 23. The gap between what people ask AI about and what they spend their day doing is the story, not the underlying report.
ATLAS stands for Activity, Task, Landscape and Adoption Study, built on a taxonomy DeepMind calls OCTO (Observation Clustering and Taxonomy Organisation). Search Engine Journal’s benchmark rests on 14.65 million logged exchanges spanning three Google surfaces: the Gemini app, AI Mode in Search, and the Gemini API. That is a slice of a wider report covering roughly 15 million. The publication then set each topic’s share of non-work conversation beside the share of waking non-work hours Americans actually spend on it, drawing the time figures from the American Time Use Survey and the conversation figures from the window of April 6 to 19.
Government services and civic obligations show the widest divergence in the data. People discuss licenses, taxes, fines and voting in AI conversations roughly twenty times more than the time-use survey shows they spend handling that paperwork. Professional and personal care services, doctors, lawyers, banks and salons, run more than seven times higher in conversation share than in time share. Education sits at 5.8 times, and consumer purchases run close to three times.
The pattern reverses for activities people carry out rather than plan for. Eating and drinking absorbs roughly eighteen times more of Americans’ active time than it generates in AI conversation share. Travel, sports, caring for household members, house cleaning, cooking and watching television skew the same direction. People act on these categories far more than they ask about them.
That reversal is the useful part for a search team. Categories where conversation share outruns time share are exactly the ones where a person needs guidance before acting: filing paperwork, choosing a lawyer, comparing insurance, understanding a fine. Categories where time share outruns conversation share are things people already know how to do. The talk-to-time gap functions as a content-opportunity map, and it points squarely at high-consideration topics that most publishers treat as an afterthought.
Google’s own report labels the medical, legal, financial and government queries high-friction, and about half of those queries arrive at night, before dawn, or on weekends, when human offices are closed. Read against the talk-to-time gap, the timing matters as much as the topic. A person asking a licensing question at eleven at night has no support desk to call, and Google’s AI is the interface answering in that gap, whether or not a published page ever gets credit for it.
None of this is traffic data. Every figure in the ATLAS report, and therefore every figure in the analysis built on top of it, originates inside products Google itself operates, and nothing in the dataset records a click. That means the report cannot say whether any of these high-friction conversations ever reach a publisher’s website, only that the conversations are happening. Google shared a related claim in May, when it placed health, food and travel among AI Mode’s top subjects without disclosing referral data either. The pattern holds: Google describes what people ask, not what happens after they ask it.
Search teams serving licensing, legal, medical or financial niches should treat the twenty-to-one and seven-to-one gaps as a prioritization signal, not a confirmed traffic opportunity. Build the after-hours, high-friction answer content first, and measure its performance in Search Console and AI Mode citation reports rather than assuming Google’s own usage numbers will translate into visits.
Search Engine Journal, reporting by Matt G. Southern, published July 25, 2026.
