Retail Media

AI Shopping Agents: The Label Is the Legal Minimum

By Paul F. Accornero, Architect of The Shopper Schism • October 7, 2026

Paid placement now sits inside the AI shopping agent, and the law already requires a label. One US rule says a label cannot cure a ranking that payment decides.

Most of the debate about ads in AI shopping agents asks one question: is the ad labeled? The law answered it years ago. Regulators in the US, the EU and the UK already expect paid ranking in search to be disclosed.

The harder question starts after the label.

The paid slot has moved inside the agent

Amazon's sponsored prompts for its shopping assistant, Alexa for Shopping (formerly Rufus), have been generally available in the US since 25 March 2026. Amazon's advertiser page, written before the rename, says that when clicked the prompts "may open a dialog in Rufus or respond to the customer directly on the page." Existing campaigns are "automatically enrolled," and the charge is "part of your CPC bidding." Third-party write-ups describe a small gray "sponsored" notice on them.

Amazon is not alone. Walmart has tested advertising formats tied to its Sparky agent, and Google has made ads eligible to appear inside AI Mode recommendation lists as a "Highlighted Answer," labeled "Sponsored." OpenAI said in September 2025 that its shopping results are "organic and unsponsored."

The paid slot used to sit beside the answer. In these products it is part of the conversation the shopper has with the agent.

The law already asks for the label, and in one case says it is not enough

The US Federal Trade Commission has asked for paid search ranking to be labeled since 2002. The European Union has banned undisclosed paid ranking in search results since 2022, under Annex I point 11a of the Unfair Commercial Practices Directive. The UK advertising code says marketing "must be obviously identifiable as such."

One US rule goes further. In the FTC's Endorsement Guides, 16 CFR 255.4, Example 3 describes a website whose product rankings are paid for. A disclosure that the operator is paid "would be inadequate," the Guides say, because the payments determine the rankings.

No statute I have found speaks directly to paid results inside an AI agent's answer. Example 3, written for a website's product rankings, is the closest text.

People miss the label inside a sentence

On a results page, the label works only if the reader sees it. Ofcom's 2026 survey found that 52% of UK search engine users correctly identified paid results.

Inside a conversation it is harder. In a preprint by Salvi, Cuevas and Horta Ribeiro (April 2026, two pre-registered experiments, 2,012 people), persuasion by a language model "nearly triples the rate at which users select sponsored products compared to traditional search placement (61.2% vs. 22.4%)," and "explicit 'Sponsored' labels do not significantly reduce persuasion." It is consistent with a peer-reviewed study by Tang and colleagues (2025), in which people struggled to detect ads in chatbot replies.

What the label cannot do

In a paper I have under review at California Management Review, I describe the disclosure gap: the distance between the commercial arrangements a platform maintains and those its agent discloses at the moment it recommends. A label narrows that gap for a person who reads it. When a household hands the decision to an agent, the agent reads the label at the moment of choice.

This is where retail media meets the Trust Paradox: consumers must trust AI agents to act in their interests, yet those agents are built and controlled by platforms with competing commercial incentives. It is also a case of the Shopper Schism®: once the shopper is an algorithm, whose algorithm it is shapes how a sponsored label is read.

I am not forecasting that retail media networks lose. Their revenue is growing. The trust cost appears where paid influence enters the agent's voice or its ranking, and it is largest where the retailer runs the agent and the household hands over the decision. The network spends that trust. The brand that bought the placement may pay for it.

What a brand controls

A brand cannot write a retailer's ranking rules. It can make its own claims checkable by an agent that works for the buyer. That is the work my glossary defines. Agent Intent Optimisation (AIO) is the discipline that runs alongside SEO when the buyer is a machine: engineering a product's data, claims, and commercial signals so that an AI agent can retrieve them, read them, and verify them independently. Where SEO competes for human attention and generative engine optimisation (GEO) competes for citation inside a generated answer, Agent Intent Optimisation (AIO) competes for the selection itself, and an agent selects on what it can check rather than on what it is told. Whether and how today's buying agents select that way is the question a pre-registered experiment is measuring now; nothing here reports its result. I set out the difference between being cited and being selected in an earlier essay on GEO and AIO.

Ask every retail media partner, in writing, how a sponsored item is labeled to the shopper inside its agent, whether paid terms enter the agent's organic ranking and whether a buyer's agent can read the label.

The label tells a person that money was paid. Verifiable product data gives an agent something to check. For the board, the question is no longer whether the ad is labeled. It is what an agent working for the buyer can check about the product.

Sources

Fetched 6 October 2026 unless marked. Preprints are marked.

Agent Intent Optimisation (AIO) and The Shopper Schism are registered trademarks in the UK and the EU.

© 2026 Paul F. Accornero. All rights reserved. · paulaccornero.com/legal-and-ip

Where to go next
Explore the Algorithmic Readiness AuditThe FrameworksThe Algorithmic Shopper
Back to Insights