For twenty years, brand strategy has rested on a stable assumption: there is a person who searches, compares and chooses.
Yet today that person has handed the choice to an agent.
Not the next generation of customers. This one.
Industry estimates indicate that as early as 2026 more than a third of online transactions will be initiated or completed by artificial agents.
More than half by 2030.
The point at which a brand meets those who choose it has changed in nature, while communication plans have stayed fixed on the previous model.
Here is how it happened, why, and with what consequences.
A concrete scene
In 2025 Walmart connected its catalogue to an OpenAI conversational assistant, making purchases possible even without the user reaching its website.
Over the same period Perplexity introduced a shopping experience with in-app checkout, and Google activated agent-executed purchasing directly on merchants’ sites. The resulting scene is concrete, and it is already here before our eyes.
A person asks the assistant for the option that suits a need. The assistant replies with three names, and acts. The brands outside those three names have remained invisible. Never assessed. Absent from the decision.
The numbers behind the break
The data accompanying this shift describe a clear-cut transition. Almost a break with the past.
Since mid-2024 referral traffic to websites generated by a major assistant has fallen by fifty-two per cent. The overlap between the top results of a traditional search engine and the sources cited in AI-generated answers has gone from around seventy per cent to under twenty. A brand can hold the top position on a search engine and be absent from the answer that steers the choice. That makes all the difference in the world.
Organisations that measure this phenomenon by click volume are watching the wrong indicator. What matters is share of mention. The percentage of relevant answers in which the brand appears. Harvard Business Review, in its March 2026 issue, recorded the same movement and described how brands are revising their structure to optimise it for an environment in which selection passes through the machine.
How machines choose
The question that follows is precise. When an artificial agent selects three names instead of thirty, on what basis does it choose them?
Selection favours not the most persuasive brand but the most legible one. The consistency with which it is described across different sources, the frequency with which it appears in authoritative contexts, the data structure that makes it accessible to retrieval systems. These are just some of the signals that speak the machine’s language. And a brand that does not manage them does exist as an experience for people, but remains invisible to the systems that precede it in the chain of choice.
The ground on which visibility is decided has changed.
Once (until not long ago) it was made of exposure. The more present you were, the more you were remembered. Now, however, that ground has become an entire territory of signals: of consistency across representations.
By now it is structure that makes a brand recognisable to retrieval systems even before it reaches the buyer’s eye. For this reason, covering the channels is necessary. Governing what systems read when they search for the brand in their own archives will soon be the condition that determines whether the brand makes it into the answer or stays out. All this, regardless of how much is spent downstream.
A brand exists where the system cites it
The lesson we can draw from this evolution runs through the individual case, and goes beyond it.
A brand exists to the extent that the system recognises and cites it. Building this recognition is a work of architecture. It rests on how the brand is structured, on how consistent its representation is and on how solid its authority is where the machine judges it.
Organisations that read all this as a technical adjustment will arrive when the space has already closed.
Those that read it as a governance decision will be present when the choice changes hands.
New Connections (FAQ)
We work B2B and don’t sell online. Does this concern us?
Purchasing is the visible tip of a broader mechanism. The same dynamic is at work in selecting a supplier, in choosing a professional partner, in the summary an agent produces of an entire sector. Wherever a system filters the options, the brand’s representation within that filter decides whether the brand enters the conversation. The scope is wider than it seems.
Our website has ranked well on Google for years. Isn’t that enough?
The overlap between the top results of a traditional search engine and the sources cited in AI-generated answers has fallen from around seventy per cent to less than twenty. A brand can hold the top position on a search engine and still be absent from the answer that guides the choice. These systems read, evaluate and cite according to their own logic. They require a structure built for that logic.
Can we wait a few more months before dealing with it?
Systems that form their own preferences tend to stabilise them. Positions gained while the ground remains open become hard to renegotiate once it consolidates. Industry estimates indicate that by 2026 more than a third of online transactions will be initiated or completed by artificial agents. Those who wait are waiting on shrinking ground.