When Search Stops Being Something Humans Do

The largest habit on earth

Start with the size of the thing. In March 2025 Google said, for the first time since 2016, how much searching it handles: “more than 5 trillion searches on Google annually,” based on its internal data from January of that year. In 2016 the answer had been more than 2 trillion. That is Google alone, before Bing, Baidu, Naver, the search bar on YouTube, the search bar on Amazon, and the search box inside nearly every app on your phone.

How many people stand behind those searches is harder to say, and I want to be straight about it. The ITU counts about 6 billion people online in 2025, 74% of humanity, with 2.2 billion still offline. My working estimate is that roughly 5 billion of them search the web at least once a week. No one publishes that number. The closest measured figure I could find comes from GWI via DataReportal: about 80% of internet users aged 16 to 64 had used a search engine in the past month. Five billion weekly searchers is a reasonable estimate, and I am labelling it as one.

What is not an estimate is the money. In the three months to June 30, 2026, Google’s “Search & other” revenue line brought in $63.3 billion, up 17% on the year before. That is one company, one quarter. Global advertising across every medium is forecast by dentsu at $1.06 trillion for 2026, and search is the largest single piece of the digital share. All of it rests on one physical act: a human being looking at a results page and deciding where to go next.

That act is what this essay is about. Not search as a technology, which is going nowhere, but search as a thing people do.

Searches don’t fall. Searchers do.

That scene is not science fiction. It is the stated design of products being shipped now. When Google introduced AI Mode it described a “query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf.” By May 2026 Google said AI Mode had passed a billion monthly users and AI Overviews 2.5 billion. OpenAI reported more than 900 million weekly ChatGPT users in February 2026.

So here is the first correction to the premise I started with. If human searching drops 80%, total query volume probably goes up, not down. One human question becomes dozens of machine queries. Much of the web is already mostly machines talking to machines: Imperva’s 2026 Bad Bot Report put automated traffic at 53% of all web traffic in 2025, after it passed human traffic in 2024 for the first time in a decade.

This is why Gartner’s well-known February 2024 forecast, that “traditional search engine volume will drop 25%” by 2026, looks wrong on its own terms. Google told investors in April 2026 that queries were at an all-time high. But the forecast was pointing at something real. It was measuring the wrong unit. The unit that is falling is not the query. It is the human glance at the answer.

The query survives. The glance does not. And the glance was the product.

The decline has already started, and it has a shape

We do not have to speculate about the early part of the curve. It has been measured.

Pew’s finding is the cleanest early reading because it watched real behavior rather than asking people to describe it. When an AI summary appeared, people clicked through to the web about half as often, and the links inside the summary were almost never used.

The publishers downstream feel that directly. Chartbeat data compiled for the Reuters Institute show Google organic search traffic to more than 2,500 news sites fell 33% worldwide between November 2024 and November 2025, and 38% in the United States. Media leaders surveyed by the Institute expect a further 43% decline over the next three years. The losses are not spread evenly.

Wikipedia, one of the sources AI answers cite most, reported in October 2025 that human pageviews were down roughly 8% on the same months a year earlier, once bots built to look human were filtered out.

And yet Google’s search revenue grew 17% in its most recent quarter. Hold those facts side by side, because together they are the whole story in miniature. The value of search is not disappearing. It is being relocated, from the pages people used to visit into the answer layer that now sits above them.

What happens to the money

An 80% drop in human search does not delete a trillion-dollar economy. It moves it. Five movements matter most.

1. Advertising shifts from persuasion to eligibility

   Search ads work because a person is looking. An agent does not look. It reads, compares and selects against the constraints it was given. The job of marketing moves from catching an eye to qualifying for a shortlist: accurate product data, real availability, verifiable reviews, a clear return policy. An early signal: Google Network, the ads Google places on other people’s websites, slipped slightly year over year in the second quarter of 2026 while its own search line grew 17%. The money is following the answer, not the page.

2. The open web loses its payment system

   The web’s bargain was simple. You let a search engine read your page, it sent you readers, and you sold those readers something. Cloudflare’s own measurements show how far that bargain has frayed with AI systems: in early August 2025 it counted roughly 50,000 crawls by Anthropic for every referral sent back, and about 887 for OpenAI. Cloudflare notes that native apps often do not report where a visitor came from, so those ratios overstate the imbalance somewhat. The direction is not in doubt. When the agent reads and the human never arrives, the page has no way to get paid. What replaces the click is being negotiated right now: licensing deals, pay-per-crawl, paywalls. Whatever wins, the free, ad-funded middle of the web gets thinner.

3. Commerce gets an agent in the middle

   McKinsey estimates agentic commerce could orchestrate up to $1 trillion of US consumer retail revenue by 2030, and as much as $3 trillion to $5 trillion globally. Those are forecasts, not results. The mechanism underneath them is sound, though: when an agent compares every option every time, prices in commodity categories get pressed hard, and the premium for merely being findable disappears.

4. Brands become standing instructions

   The brands that do well in a delegated economy are the ones people name. “Get the usual.” “Book the place we like.” A brand becomes an instruction the person gives the agent, and everything without that status competes on specifications alone. The strong get stronger. The anonymous middle gets squeezed.

5. The default becomes the whole game

   Google paid Apple roughly $20 billion in 2022 to be the default search engine on Apple devices, according to court documents unsealed in 2024. In September 2025 Judge Amit Mehta barred exclusive default contracts going forward, while still allowing payment for placement, and ordered Google to share some search data with qualified rivals. That fight was over which search box you see. The next one is over which agent you have. When your agent does the searching, the default is no longer a slot on a screen. It is the thing that decides for you.

There is a whole labor market built on top of human search: SEO, affiliate publishing, comparison sites, local directories, content marketing, lead generation. None of it vanishes overnight, but all of it was priced on a human arriving at a page. I will not invent a job count; nobody has a credible one. What I will say is that the craft moves from getting found by people to getting chosen by machines, and those are not the same craft.

What happens to people

The economic story is about where the money goes. The social story is about what people stop seeing.

The shared page disappears

For twenty-five years, two people who typed the same question saw roughly the same ten links. That common page was a quiet piece of civic infrastructure. It meant a disagreement could at least start from the same evidence. A personal agent answers for you, from what it knows about you. The page stops being shared, and nobody notices, because nobody sees the page.

Surprise gets filtered out

The results page was where people ran into things they had not asked for: the tenth link, the contrary article, the forum post from someone who tried it and hated it. In the language of my Reality Equation, only Actuals retrain expectation. You cannot argue someone’s prediction into a new shape. It moves only when the world hands them something. An agent tuned to satisfy your request will tend to hand you the Actuals that fit it. Nobody touches your expectations directly, because nobody can. But whoever controls which Actuals arrive shapes, slowly, what you come to expect. At the scale of a society, that is the biggest change on this list.

Checking becomes a minority habit

Pew found that people clicked a source inside an AI summary in about 1% of visits. When the agent does the research, checking its work is an extra step, and most people will skip it, the way most people stopped reading the map once the phone started giving turn-by-turn directions. Errors do not have to be frequent to matter. They only have to go unexamined.

The commons starves the thing that feeds on it

Agents are good because they learned from, and keep reading, what people wrote on the open web. Wikipedia’s editors, forum answerers and small publishers mostly arrived as visitors first. Fewer visits means fewer of them. The agents are drawing down the well they drink from, and no one has yet built the pipe that refills it.

A new divide: whose agent works for whom

2.2 billion people are still offline, and that divide persists. A newer one opens on top of it. Some people will have capable agents they pay for, whose only job is them. Others will have free agents, and free has always meant someone else is paying. An agent funded by advertising has two masters at the exact moment it is choosing on your behalf. The class line of the delegated economy is not who can search. It is whose searcher is loyal.

The self that searched

Searching was also a way of thinking. Typing a question forced you to phrase it, and scanning the results showed you how other people framed the same problem. Hand all of that over and something in you stops being exercised. Not intelligence. Framing. That loss is worth naming even when the trade is worth making.

The ledger

A forecast is only useful if you can check it. Here is what I am claiming, and how it could fail.

Claim one: human search falls by 80%.

Already true

AI answers already roughly halve click-through on the searches where they appear. Google search referrals to news sites fell by a third in a single year. Over a billion people a month use AI Mode, which searches on their behalf.

What has to happen

Agents have to be trusted to act, not just answer: booking, buying, filing, and doing it reliably enough that people stop checking. Payments, identity and merchant systems have to be rebuilt for customers that are software. The routine searches go first: prices, hours, directions, reservations, comparisons.

Where I am probably wrong

Timing, first. This could take a decade rather than three years. And some searching is not a chore. People browse recipes, houses and old friends for the pleasure of looking, and nobody delegates a pleasure. If curiosity-searching survives while chore-searching disappears, the drop could stall well short of 80%. The economics above would still apply to the chore half, which is where most of the money is.

Claim two: the money relocates rather than disappears.

Already true

Google’s search revenue grew 17% in the second quarter of 2026 while its search referrals to publishers kept falling.

What has to happen

Agents need a revenue model that holds up: subscriptions, commissions on transactions, or advertising placed inside the answer. Whichever wins decides whose interests the agent serves.

Where I am probably wrong

Regulation or licensing could change the split. If paying sources for what agents read becomes the norm, publishers could end up better paid per use than clicks ever paid them, and the commons would refill instead of drain.

The question after the question

For a quarter of a century, the most common deliberate act of curiosity on earth has been typing a question into a box. It became so ordinary that we stopped seeing it, and an economy grew up around the seeing. When four out of five of those questions are asked by software on our behalf, searching does not stop. Something else stops: the moment a person looked at the world’s answer and chose.

That is not a reason to refuse the trade. I use an agent every day and I would not go back. It is a reason to ask the one question that matters more than what happens to Google: when your AI goes looking on your behalf, who else is it working for? Settle that, and most of this essay turns into a set of engineering problems. Leave it open, and the largest habit on earth quietly becomes someone else’s business model.