The most valuable real estate in the history of advertising was ten centimeters of white space next to a text box. Everything Google built, every quarterly earnings call, every agency retainer I have ever negotiated, sat on one behavioral assumption: when a human wants something, they will type it into a search box and tell you. That assumption held for a quarter century. It is now breaking, and most of the industry is still bidding on keywords like it is 2019.
I run campaigns for brands that used to live and die by search. What I see in the data, across my own clients and my own properties, is not a decline in intent. People want things as much as they ever did. What is declining is the declaration of intent. The query was a confession, freely given, timestamped, auctionable. When a customer asks an assistant to handle a task, when an agent researches products on their behalf, when a model answers before a browser ever opens, the confession never happens. Intent still exists. It just stops being spoken out loud in a place where you can buy adjacency to it.
This paper is about what advertising becomes when the query stops being the atomic unit of the industry. My core claim: the query is not being replaced by another surface you can buy. It is being replaced by a process, and you cannot buy adjacency to a process. You have to be inside it.
Key Findings
- The typed query was an artificial bottleneck that concentrated intent into an auctionable moment. That bottleneck is dissolving, and no equivalent chokepoint replaces it.
- Intent is shifting from declared to inferred. I call the always on, machine mediated version of this Ambient Intent: intent detected from context, delegation, and history rather than from a typed statement.
- The excess profit the industry earned simply because people typed their needs into a box, what I call the Query Dividend, is being competed away between 2026 and 2030. Budgets built on it will not survive contact with agent mediated buying.
- The winning position after the query is not adjacency but eligibility: being the answer a system retrieves, trusts, and executes against.
- Keyword research survives only as a proxy discipline. The real planning unit becomes the task: what the customer delegated, not what they typed.
The Query Was a Subsidy
Be honest about what search advertising actually was. It was not a triumph of targeting science. It was a toll booth placed on a bottleneck. Humans had needs, the web had answers, and the only bridge between them was a text box owned by one company. Every advertiser paid the toll, and the toll was cheap relative to the intent quality, because the customer did the hardest job in marketing for free: they told you exactly what they wanted, in their own words, at the moment they wanted it.
That is the Query Dividend: the margin advertisers and platforms harvested purely because declaration of intent was concentrated in one purchasable place. It was never a law of nature. It was an artifact of interface design. Change the interface and the dividend evaporates.
The interface is changing. Assistants answer directly. Agents execute multi step tasks without surfacing intermediate choices. I wrote about the mechanics of this shift in my analysis of where Google search is heading, and the short version is that the box is becoming a conversation, and the conversation is becoming a delegation. Nobody auctions the inside of a delegation. Yet.
From Declared Intent to Ambient Intent
Here is the frame I use with clients. Intent used to arrive in one shape: a declaration. "best crm for small agency." Ugly, explicit, gold.
Ambient Intent is what replaces it. It is intent that a system infers from everything except a declaration: the project you are working on, the contract renewal date it can see, the tone of the email you just received, the fact that your team doubled last quarter. An agent that knows all of that does not wait for you to search for a CRM. It knows you need one before you do, and when it goes looking, it does not go looking on a results page.
Ambient Intent has three properties that should terrify anyone running a keyword driven budget. It is continuous, so there is no auction moment. It is private, so there is no bid signal. And it is delegated, so the entity doing the choosing is not the entity doing the wanting. You are no longer persuading the customer at the moment of intent. You are persuading the system that noticed the intent, and that system reads documentation, structured data, reviews, and reputation. It does not read banner ads.
Where Intent Goes to Live
If the query disappears, where does the intent physically go? Follow it. It goes into three places.
First, into conversations with models. A meaningful share of commercial deliberation now happens inside chat sessions, where a person reasons through a decision with an AI over twenty turns. That deliberation used to be twenty searches. Now it is one context window, and your brand either exists inside the model's working knowledge or it does not.
Second, into agent workflows. When a person delegates a task, the intent is encoded in the instruction and the agent's toolchain, not in any public surface. I have watched agents comparison shop on behalf of users, and the brands that win those comparisons are the ones with clean structured data, parseable pricing, and machine readable trust signals. I laid out the playbook for this in my guide to marketing to AI agents.
Third, into context itself. Calendars, inboxes, documents, purchase histories. The systems that read this exhaust do not need the customer to declare anything. This is where Ambient Intent is born, and it is entirely upstream of anything the ad industry currently sells.
What Gets Bought When There Is No Keyword
I do not believe advertising dies here. Money always finds attention, and attention still exists. But the unit of purchase changes, and I will make this concrete with predictions I am willing to be wrong about in public.
By 2028, the dominant paid product in intent marketing will not be a keyword auction. It will be some blend of three things: paid eligibility, where brands pay to be retrievable and verifiable inside answer and agent ecosystems; paid data relationships, where brands supply feeds, inventory, and pricing directly to model providers under commercial terms; and outcome bids, where brands bid on completed tasks rather than impressions, because the agent can actually confirm completion.
Notice what all three have in common: none of them are media in the traditional sense. They are infrastructure positions. The question stops being "where does my ad show" and becomes "is my brand executable by the systems doing the choosing." I get into the measurement side of this, and what replaces search volume as the planning number, in Paper No.16.
The Timeline as I See It
Back of the napkin, here is how I expect this to sequence. Through 2026 and 2027, declared queries keep shrinking as a share of commercial decisions, but the auction still clears at rising prices because inventory contracts faster than budgets move. That rising CPC is not health. It is a crowd squeezing through a shrinking door.
From 2027 to 2029, agent mediated purchasing crosses from novelty to normal in a handful of categories first: travel, software, replenishable goods, business services. In those categories the Query Dividend goes to zero. Not down. Zero. There is no moment to buy.
By 2030, "search budget" will read the way "Yellow Pages budget" read in 2015: a legacy line item defended by the people whose careers depend on it.
What I Would Do About It
Concrete moves, in order of urgency.
Reframe your planning unit from keyword to task. List the twenty tasks your customers will delegate to agents in your category, and audit whether a machine executing each task can find, parse, verify, and transact with you. That audit will scare you into the right roadmap.
Build the machine legible version of your brand now. Structured data everywhere, entity clarity, pricing an agent can read without a human, claims backed by sources a model can corroborate. This is unglamorous work and it is the entire game.
Keep buying queries, but treat the spend as a melting asset. Harvest the remaining dividend, do not build headcount on it, and move a fixed percentage of that budget every quarter into answer eligibility, data partnerships, and owned relationships.
Own the relationship before the interface disappears. First party channels, memberships, direct utility. When an agent asks a customer "do you have a preference," the brands that installed themselves in memory win by default.
The query gave us the easiest targeting in advertising history, and we mistook a toll booth for a business model. The intent is still out there. It just stopped announcing itself. Go to where it lives now.