The next customer is an agent

A working definition of agentic advertising—and why brand means something different to a model

“The next customer is an agent” is deliberately provocative. The human still has the need and sets the preferences. In the commerce systems discussed here, the human also authorizes the purchase. The merchant still has customers, responsibilities and a brand to protect.¹

What changes is the route between intention and choice.

AI systems can already interpret a request, compare products and, in some services, help complete checkout. The human may receive a shortlist or finished cart without reviewing all the underlying product pages.¹

The next gatekeeper, then, is an agent.

A working definition

Agentic advertising is paid communication designed to appear while an AI helps someone make a decision or complete a task.

That definition needs clear boundaries. Helping an AI understand a product is not automatically advertising. Neither is appearing in an organic citation or recommendation. Within this framework, payment marks the boundary between organic selection and advertising. Paid influence should be clearly disclosed, especially when it could otherwise appear to be an independent recommendation.²

There are three distinct layers:

Layer	The brand’s job	The platform’s job	Paid?
Machine legibility	Supply accurate, structured, current facts	Retrieve and interpret them	Not by itself
Organic selection	Earn inclusion on relevance, fit and trust	Compare and recommend independently	No
Agentic advertising	Offer a useful, clearly identified paid message	Match, label and keep it separate from the answer	Yes

OpenAI already makes this separation. It says ChatGPT shopping results are chosen independently and are not ads. Paid ads are separate from answers and clearly labeled. OpenAI also says its ad system looks at conversational intent and other allowed context, not just keywords.³

Agencies should not collapse these layers into one promise to “make the AI recommend your brand.” That would mix technical optimization, earned visibility and paid media. It would also create a serious trust problem.

What changes

Machine consideration can precede human attention

Advertising has long competed for a person’s limited attention. That scarcity does not disappear. But a new selection event can happen before it.

Missing or unclear information may make a product harder for an agent to compare. The item may be unavailable, its returns policy may be unclear or an important specification may be missing. It may also fall outside the user’s budget or delivery deadline. This is an inference from the inputs platforms say they use, not a published ranking rule. No headline gets the chance to work if the product fails that first comparison.⁴

Clear, verifiable information can affect whether an agent is able to evaluate a product. Relevant facts may include its identifier, options, price, availability, total cost, delivery time, return policy, warnings and support terms. OpenAI says ChatGPT shopping may rank merchants using factors such as availability, price, quality and whether the merchant is the maker or primary seller. Google’s product guidance also emphasizes price, availability, shipping and returns.⁴

A media plan can no longer begin with the ad impression. It must also account for the product information that determines whether an offering can be found and compared.

Brand becomes evidence a model can retrieve

A model does not “feel” a brand as a person does. Its output should not be treated as one stable opinion of a company. The result can change with the user’s request, the sources retrieved, the available product data and the platform’s rules.⁴

For this essay, it is useful to think of brand as a body of evidence. That evidence may affect whether a company appears relevant, reliable and safe to recommend.

Potential evidence includes the company’s claims and policies, product catalogs, independent reporting, reviews, expert sources, complaint records, recalls and fulfillment history. Which sources matter will depend on the system, the request and what information it retrieves. Conflicting sources make the picture less clear.

For a human, brand can compress experience into a feeling: “I trust them.” For an agent, brand is more likely to be reconstructed at the moment of a task: “The available evidence suggests this option meets the constraints.”

Brand building still matters because much of that evidence comes from human experience. Reputation, coverage, reviews, loyalty and cultural meaning do not disappear. They become inputs to the agent’s decision.

Creative expands from message to proof system

A traditional creative brief asks what people should remember. An agentic brief must also ask what a system can verify and use.

That does not mean replacing emotion with a spreadsheet. Human desire still creates the brief. Strong ideas still generate demand, and visual identity still makes a product recognizable. The difference is that objective product claims need supporting evidence.⁵

“Built for small spaces” becomes dimensions, clearances, installation requirements and photographs in context. “Arrives in time” becomes inventory by location, handling time and a delivery commitment. “More sustainable” becomes a defined boundary, method, certification, date and comparison baseline.

At least one current ad platform is moving in this direction. OpenAI’s guidance favors specific, benefit-focused copy that explains what an offering does, who it is for and when it is useful, rather than relying on a generic slogan. It also requires the landing page to be relevant, reachable and consistent with the ad.⁶

The strongest creative idea may still be emotional. The supporting facts must be easy to check.

From discovery to transaction

The change goes beyond search results. Commerce systems are being redesigned so that a platform or agent can move from discovery to checkout without a custom integration for every merchant.

OpenAI and Stripe’s Agentic Commerce Protocol connects buyers, agents and businesses during checkout. Google and an industry group are developing Universal Commerce Protocol as a common language for platforms, agents, businesses and payment providers. Microsoft describes product feeds as the discovery layer and protocol-based checkout as the transaction layer. In Microsoft’s Copilot Checkout, the merchant remains the merchant of record.⁷

These systems are still changing. The Agentic Commerce Protocol, for example, describes its current specification as beta. Agencies should not assume that one protocol will become the universal standard. The direction matters more than the acronyms: agents need current data, clear permissions and secure payment. Users, platforms and merchants also need a record of what happened.⁷

The current checkout examples discussed here also keep a human control point. Google says its agentic checkout asks for permission and purchase confirmation. Microsoft says the merchant remains the merchant of record and continues to use its existing payment, fraud, tax, fulfillment and reconciliation systems. People may talk about an agent “buying,” but agencies should still ask four questions: Who approves the purchase? Who pays? Who fulfills the order? Who is accountable?⁸

What agencies will be asked

Clients will not begin with a neat request for “agentic advertising.” They will arrive with practical questions:

Can an agent correctly understand what we sell?
Why is a competitor recommended when we are not?
Which claims survive comparison with independent sources?
Are our price, inventory, shipping and returns consistent everywhere they appear?
When is inclusion organic, and when can media buy access?
What should creative look like when matching is based on conversational context rather than one keyword?
Can an agent complete the task without breaking the customer experience?
How do we measure influence when the decision begins—or ends—inside the interface?

These questions cross the usual boundaries between media, SEO, commerce, experience design, data and reputation. Another dashboard on top of disconnected systems will not solve them.

From questions to an operating model

These questions do not fit neatly inside an SEO, media or commerce team. Agencies will need to connect five things: the brand promise, the evidence behind it, the product record, the path to action and the rules for measurement and accountability.

That framework—and a practical first-quarter plan—belongs in its own discussion. I develop it in the companion piece, An operating model for the agentic customer.

The strategic point comes first: becoming easier for a machine to read is not the same as earning a recommendation, and neither is the same as buying an ad. Agencies should keep those activities separate even when they coordinate the work.

Brand after the interface

The most tempting conclusion is that brands should now market to machines instead of people. That is exactly backward.

Agents act on human goals. They use human preferences, follow platform rules and rely on evidence created by companies and communities. A brand that manipulates the machine while disappointing the customer may eventually produce reviews and other evidence that make it harder to recommend.

The durable strategy is this: give the person a clear promise, give the agent reliable proof and make the path between them accountable.

The next customer is still human. But the next decision may pass through a model first.

Sources and further reading

Want to talk this through for your clients?

The first conversation is free.

Book a Scoping Sprint

More research