Sponsored products now appear inside AI agent responses. The $165 billion retail media industry just found its way into the agentic commerce stack. Our own thesis needs updating.

We owe you a correction.

In February, we published an analysis titled "AI Agents Don't Click Ads. That Changes Everything." The argument was clear: agents parse data, not pages. They do not scroll, do not see hero banners, and do not respond to promoted listings. The $165 billion retail media industry, we wrote, faced structural disruption because its entire model depended on humans browsing. Agents query. They do not browse.

That was half right. The part about agents not browsing still holds. The part about advertising not reaching them does not.

On March 2, Criteo announced it had become the first ad-tech partner in OpenAI's advertising pilot inside ChatGPT. Sponsored product cards now appear below AI responses for users on the Free and Go tiers in the United States. The cards show a brand logo, product name, price, stock status, and estimated delivery. Labelled as "Sponsored." Visually separated from the organic answer. But right there, inside the conversation.

The mechanics matter here, because this is not a banner ad duct-taped onto a chatbot.

Criteo connects 17,000 advertisers and activates more than $4 billion in annual media spend. Through the OpenAI partnership, brands using Criteo's existing commerce media platform can now place ads inside ChatGPT responses with no new contracts, no new feeds, no new tags. If you are already a Criteo advertiser, you can go live in days.

That is the detail worth pausing on. This is not a speculative integration requiring months of technical work. It plugs into an existing demand-side pipeline serving thousands of brands. The supply of sponsored products that can appear inside AI conversations scaled overnight.

Criteo has also been building Model Context Protocol support for commerce use cases, piping product relevance signals, pricing data, and retailer performance into LLM environments. In late 2025, the company ran early experiments feeding these signals directly into large language models during inference. The aim: when an agent generates a product recommendation, Criteo's data is already in the room.

And the early numbers back it up. Aggregated data across Criteo's US clients shows users referred from LLM platforms convert at roughly 1.5 times the rate of other referral channels. High-intent users getting high-relevance results inside a conversation they initiated. That is a compelling pitch to any brand spending money on search ads.

Here is where we need to be precise about what broke in our original argument.

We wrote that AI agents would disrupt retail media because agents "don't click ads." That framing assumed advertising was a property of the website, something embedded in the browsing experience that an agent would simply bypass. Agents query APIs. They read structured data. They do not load web pages full of ad units. The ads, we reasoned, become invisible.

What we missed is obvious in hindsight. The ads do not have to live on the website. They can live inside the agent's own response.

OpenAI launched the advertising pilot on February 9, 2026, initially testing with agencies including Omnicom and WPP. Ads appear on the Free and Go tiers only. Plus, Pro, Business, Enterprise, and Education users remain ad-free. Users under 18 do not see ads. The format is cost-per-view, and advertisers do not get access to chat logs, only aggregated performance metrics.

The privacy safeguards are real. But the structural shift is what matters for our thesis. The middleman did not get cut out. The middleman moved upstream. Instead of the retailer paying to be promoted on a search results page that an agent bypasses, the retailer now pays to be promoted inside the agent's own answer.

The entire premise of agentic commerce rests on a simple proposition: the agent works for the buyer. It searches, evaluates, compares, and purchases on your behalf. It is your fiduciary in a transaction. That is the promise.

Sponsored placement inside the agent's response complicates that promise. Not destroys it. Complicates it.

As we explored in our analysis of the trust gap in agentic commerce, only 46 percent of shoppers fully trust AI recommendations. That number was already low. Now add the knowledge that some of those recommendations are paid placements. The trust problem just got harder.

BCG identifies a scenario where AI agents manage everything from research through purchase, shifting retail media toward "influencing agent defaults." Read that phrase carefully. Influencing agent defaults. That is not an agent optimizing for the buyer. That is an advertiser optimizing the agent.

And the scale is real. US retail media spend will hit $69.3 billion in 2026, up 18 percent year over year. Globally, the figure approaches $165 billion. By 2029, spending on AI search ads alone will reach $26 billion, roughly 14 percent of total search ad spend. That money is not going to disappear because agents arrived. It is going to follow the agents.

We wrote about how recommendation poisoning threatens the invisible shelf. We described how bad actors could manipulate what agents recommend through data poisoning and prompt injection. What we did not anticipate was that the manipulation would arrive through the front door, labelled "Sponsored," sanctioned by the platform itself.

Here is the tension nobody in the agentic commerce space has resolved.

If the agent is funded by the user (through a subscription, say), the agent's incentive is to serve the user. That alignment is clean. OpenAI preserves this for paid tiers. Plus and Pro users see no ads. Their agent answers are unsponsored.

But most ChatGPT users are on the Free tier. Their experience is now ad-supported. The agent's response includes sponsored products selected by whoever paid Criteo for placement. The user did not choose those products. The agent did not choose them on merit. Criteo's algorithm chose them based on advertiser bids and relevance signals.

This creates a two-tier system for agentic commerce. Pay, and your agent is independent. Do not pay, and your agent carries advertising.

We covered the search bifurcation in February. The split between ad-supported and subscription-funded AI was already visible. The Criteo deal makes it concrete. Discovery itself now depends on who is paying, and whether "who" is the user or the advertiser.

That is a structural conflict of interest at the foundation of agent-mediated commerce. It does not make the technology useless. It makes the governance question urgent.

The correction is not total. Two parts of our original thesis hold up.

First, agents still do not browse. The traditional retail media model built around banner impressions, promoted search results, and display advertising on retailer websites is still threatened. Criteo did not save that model. Criteo built a new one. The advertising infrastructure that depends on humans clicking through web pages has not been rescued.

Second, the retailers who win in agentic commerce are still the ones with clean data, transparent pricing, and reliable fulfillment. An agent evaluating products, even one that surfaces some sponsored results, still needs structured data to work with. The operational discipline argument we made holds. Criteo's own MCP integration relies on accurate pricing and availability data from retailers.

What changed is the power map. We assumed the agent would be a pure buyer's agent, cutting out intermediaries and optimizing for price, quality, and relevance. The reality is more complicated. The agent sits in the middle, influenced by the platform that hosts it, the data it has access to, and now, the advertisers who pay for placement within its responses.

The Criteo deal is a pilot. It covers the US market, Free and Go tiers, with cost-per-view pricing and privacy safeguards. But pilots scale. Criteo already has 17,000 advertisers ready to activate. OpenAI has hundreds of millions of users. The infrastructure is in place.

Watch for three things. First, whether other ad-tech companies follow Criteo into conversational AI placement. Google's Gemini, Perplexity, and Microsoft's Copilot are all potential advertising surfaces. Second, whether regulators engage with the conflict of interest question. An agent recommending sponsored products without the user fully understanding the commercial relationship is a consumer protection issue. Third, whether premium "ad-free agent" tiers become a standard pricing model across the industry, creating a permanent divide between ad-influenced and independent AI assistance.

We built our agentic commerce coverage on the argument that agents bypass traditional advertising. That was true of the old advertising. The new advertising found the agents. We should have seen it coming. The $165 billion retail media industry was never going to accept disruption quietly. It was going to adapt. And it did, faster than we expected.

If the agent carries advertising, whose agent is it, yours or the advertiser's?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.