Google’s own data just mapped where AI search hits first

Google published something this week that most people will read as a curiosity. It's not. It's a map.

The AI & Economy ATLAS report — 14.65 million real interactions across the Gemini app and AI Mode — lines up what Americans ask Google's AI about against what they actually spend their time doing. The mismatches are enormous. Government paperwork comes up in AI conversations roughly twenty times more often than it takes up people's day. Professional and personal care services — doctors, lawyers, banks — more than seven times. Education nearly six. Consumer purchases about three.

Meanwhile, people spend far more of their waking hours eating, cleaning, watching TV, and getting dressed than they do asking AI about any of it. That gap runs the other way at roughly eighteen to one.

If you're trying to work out which sectors are going to feel AI search first, and hardest, this is the closest thing to a public signal Google has ever released. And the categories that jump out are the ones every UK small business consultant already has clients in.

The shape of the demand shift, in one chart

Strip the report down and you get a rough hierarchy of where AI has already reset user behaviour versus where it hasn't.

The categories being eaten first are the ones with the highest research intent per unit of activity.

At the top — the categories where AI usage vastly outruns the time people spend on the underlying activity — sit the exact verticals I've spent 18 years building websites and running SEO for. Health. Legal. Financial services. Local professional services. Education and training. Considered consumer purchases.

At the bottom — where time spent dwarfs AI usage — sit categories that were already hard to monetise through search anyway. Passive entertainment. Household chores. Physical activities. Eating.

The middle sits roughly where you'd expect: hobbies, DIY, home repairs, comparison shopping.

The categories being eaten first are the ones with the highest research intent per unit of activity.

This isn't a surprise once you say it out loud. Nobody Googles "how do I eat lunch." People Google — and now ask Gemini — how to appeal a council tax band, whether a symptom is serious, what the difference is between an ISA and a SIPP, whether a particular solicitor is any good, which boiler to buy. Short, high-stakes, information-dense questions with real consequences attached.

Those are the questions where AI search is genuinely better than a page of blue links. Which means those are the questions AI search is going to keep for itself.

What this means if you're a UK service business

If you sell anything into the categories at the top of that ATLAS chart, the traffic composition of your website is already changing. It's just that most owners haven't realised yet, because absolute traffic numbers can look flat or even up whilst the underlying mix rots.

I see this pattern constantly in audits. A local firm looks at their Search Console dashboard, sees stable clicks year-on-year, and concludes the AI thing is overblown for their business. What they haven't checked is the query mix. The informational long tail — the "what is," "how do I," "is X worth it" queries — is quietly hollowing out. What's left is a smaller number of higher-intent, later-funnel queries where the user already knows what they want and is looking for a provider.

That sounds fine. In some ways it is fine — those are better clicks. But it has two consequences most people don't think through.

First, the informational content that used to feed the top of your funnel is now feeding Gemini's answer instead. You still need it to exist for the AI to cite you, but it will produce a fraction of the direct traffic it used to. Which means the ROI calculation on writing it changes.

Second, the intent that does still reach your site is more concentrated at the bottom of the funnel. That's more valuable per visit but also more brutal — you get fewer chances to build familiarity before someone is comparing you against two other quotes.

The verticals Google just flagged

Look at the top of the ATLAS gap chart and count the industries.

Government services and civic obligations (~20×). Anyone who does content or lead generation around tax advice, immigration, benefits, planning applications, licensing — the informational demand is already gone. What's left is enquiry-intent traffic.

Professional and personal care services (~7×). Doctors, lawyers, banks, salons. Every solicitor, accountant, IFA, dentist, physio, and clinic in the UK sits here. The most valuable historical SEO play in these verticals — write a definitive "what is [thing]" guide and rank for it — is precisely the play the AI has now automated.

Education (~6×). Course providers, training bodies, universities, tutoring businesses. Discovery and comparison happens in the chat window now, then people surface to book.

Consumer purchases (~3×). Retail and considered goods. The comparison-shopping traffic that used to land on affiliate blogs and product pages is being pre-digested.

Health, hobbies, financial services, home repair, appliance fixing. All listed further down the gap chart. All full of UK small businesses that historically built their acquisition on informational SEO.

Notice what's not on this list. Trades where the customer is already in a physical space with a broken thing. Emergency callouts. Purely local walk-in services. Anything where the query is "who, near me, right now" rather than "what, and is it worth it."

Those categories are fine. The rest need to think.

The categories being eaten first are exactly the categories most UK small businesses have been told to build content strategies around for the last decade.

What ATLAS doesn't tell you — and why that matters

Google shared usage patterns. They did not share what happens next.

The report is silent on whether any of these AI conversations produce a click to a website. Google's public statements about that — Nick Fox saying AI features send "billions of clicks weekly," Liz Reid saying total organic click volume is "relatively stable" with better average quality — are, as Search Engine Journal politely pointed out this week, unfalsifiable. There's no denominator. No breakdown by surface. No way to check.

So we have two data points that need to be held together carefully. Google is telling us confidently that certain categories dominate AI conversations. Google is also telling us — with much less specificity — that clicks to the web are fine, actually. Those two statements can both be true. But the first is auditable and the second isn't, and I'd want to be careful about basing planning decisions on the assurance rather than the data.

The honest read of ATLAS is: this tells you where user attention has moved inside Google's own products. It does not tell you what share of that attention ever makes it back out to your website. That second number is the one that would actually determine your marketing budget for 2027. And Google is choosing not to publish it.

The counter-read: why this might be less bad than it looks

Steelman time. There are two arguments for why the picture isn't as grim as the ATLAS chart makes it look, and both deserve to be taken seriously.

The first is that AI search may be expanding total demand rather than cannibalising it. Pichai's claim that "AI features are driving Search query growth" isn't nothing — Search & Other revenue is still up 17% year-on-year. If people are asking questions they wouldn't have asked before, some fraction of those will still convert into clicks, and the absolute click volume could hold up even as the click rate per query falls. This is a plausible reading of Reid's "stable click volume, higher quality" framing.

The second is that AI conversations concentrated in high-stakes categories may actually be pre-qualifying leads rather than replacing them. If someone spends fifteen minutes asking Gemini about ISAs before landing on an IFA's contact form, that IFA is getting a warmer lead than they used to get from a "what is an ISA" blog post. The traffic drops, the conversion rate rises, and the net economics look different.

I think both of these are partially true. But I don't think they get you to "nothing has changed." They get you to "the funnel has been compressed and re-shaped, and if your business relied on the wide informational top of it, you have a problem you need to address deliberately." That's the piece the industry commentary keeps missing.

The practical read

If your business sits in one of the categories Google's own data just flagged as high-AI-usage — and if you're a UK service business, it almost certainly does — three things follow.

AI systems cite brands they've heard of. If nobody's talking about you, nobody's citing you.

Stop measuring content ROI on direct traffic alone. The informational content still matters because it's the input layer AI systems cite. But its job has changed from "capture a click" to "influence a citation." Measuring the first will make you cut content that's doing the second.

Get honest about where you are in the funnel. If your traffic looks stable but your enquiry rate has quietly climbed whilst absolute session count has fallen, that's the compression happening in real time. It's not necessarily bad news — but it needs to be planned around, not stumbled into.

Build brand aggressively in the categories where AI is eating first. Health, legal, financial, education, professional services. These are the categories where being a recognised name in your local market has moved from "nice to have" to "the only durable acquisition channel." AI systems cite brands they've heard of. If nobody's talking about you, nobody's citing you.

Google just told us, using its own product data, which parts of the search economy are already inside the AI answer and which parts are still outside it. That's a genuinely useful map. It's just not the map most agency decks are drawing from, because most agency decks are still working from the framing that AI search is coming for everyone equally. It isn't. It's coming for specific categories first, and it's already arrived in most of them.

The businesses that will handle this well over the next 18 months are the ones who look at the ATLAS chart, find themselves near the top of it, and stop pretending it's someone else's problem.

The rest will keep publishing "what is" guides into the void and wondering why the numbers don't add up.