A consumer picks up her phone and types: “Find me the best running shoe for wide feet under $150 available for delivery to my home by Thursday.” She doesn’t open a social-shopping app or launch a web browser. She doesn’t browse a retailer’s category page or scan sponsored listings. She simply asks her AI assistant, receives three options with a brief rationale for each, taps the one that fits, and checks out—all without visiting a single retailer site. Or, even easier, she lets an AI agent do all that for her.
For the brands that appeared in that answer, the transaction was a win. For the brands that did not, it never happened. And for the commerce media network (CMN) whose sponsored listings, category placements, and product detail pages went unseen throughout the entire journey—the shelf was full, but no one walked down the aisle.
This scenario is not a forecast. According to the February 2026 McKinsey Advertiser Survey, over 55 percent of consumers use AI for consideration and decision-making, and 60 percent say AI delivers better answers than traditional search. More than $3 trillion of global consumer commerce will likely be mediated by AI agents by 2030. The consumer in this scenario is quickly becoming the norm.
For most of the past decade, CMNs built their advantage on proximity to the transaction. Retailers built these networks by selling advertisers access to the shoppers already browsing their sites and apps—sponsored placement in search results and product pages, priced against the retailer’s own purchase data. Telecoms, travel companies, and other consumer brands with first-party data have since followed. Retailers discovered that their purchase data was worth more monetized than hoarded, and advertisers paid a premium to reach shoppers at the moment of highest intent. That model was sound because the shopper actually saw a product page or search result before buying—the digital shelf—but this is no longer the case.
The compression problem
Three forces are squeezing CMNs at once: Discovery is moving upstream; evaluation is shifting from pages to structured signals; and buying is automating on the same timeline.
Discovery is moving upstream
Nearly half of Google searches now include an AI-generated overview. Traffic from traditional search has declined 20 to 50 percent in some categories. The scarce resource is no longer ad inventory near checkout but rather eligibility inside the systems that decide what gets considered, compared, recommended, and bought. When an AI-generated answer contains three options, or one default, inclusion is everything and exclusion is a lost sale at the moment purchase intent was highest. AI-enabled shopping agents are already delivering up to 60 percent higher conversion rates than traditional browse-and-click flows.
For CMNs, this collapse has direct onsite and offsite implications.
For onsite, the risk is potentially dire. As search, clicks, eyeballs, and decisions move upstream, the traditional onsite advertising model of sponsored search and display is at risk of decline, even as more share of sales shifts online.
For offsite, the programmatic and search inventory that powers most offsite CMN campaigns—display retargeting, paid search extensions, open-web video—is built on the same traffic ecosystem that AI is draining. Offsite media that routes through the open web is losing the mid-funnel surface area it was designed to reach.
The pressure is particularly acute for advertisers without a native closed loop—categories like travel, telecom, and financial services that depend on open-web and programmatic inventory to drive consideration but have no first-party transaction signal to prove that the AI moment converted.
AI reads signals, not pages
More than half of marketers are already investing in AI-native ad formats on CMNs; 51 percent report purchasing AI-driven placements. But buying into AI formats without rebuilding the infrastructure underneath is a short-term fix. An AI assistant evaluating a purchase recommendation processes structured signals—attributes, pricing, availability, fulfillment terms, reviews, verified claims—not product pages or sponsored listings. CMNs whose data infrastructures make every product legible to AI at the moment of query are best positioned—a capability discussed in more detail below. Those whose infrastructures were built for a retailer’s own search index may find it insufficient for the AI layer above it.
Buying is automating, too
The buy side is automating on the same timeline. Some 82 percent of advertisers plan to buy AI ad formats directly in the next 12 months, bypassing traditional intermediaries. Networks that can be activated programmatically, expose optimization signals through clean APIs, and report performance in near real time will attract that spend. Those that can’t will require manual workarounds that erode efficiency and shift budgets elsewhere.
What CMNs actually sell is changing as a result of buy-side demand. CMNs are no longer just selling impressions or placements, but instead are selling inclusion in the shortlist an AI assistant presents to a shopper, and the ability to prove that influence translated into measurable outcomes. That proof matters more than ever: 42 percent of advertisers already cite reliance on black-box measurement systems as a key risk, and scrutiny of platform-reported results is rising. CMNs are among the best-positioned players in the advertising ecosystem to benefit from this move from browsable shelves to AI-curated shortlists—but only if they make the transition from media engines to decision infrastructures.
Five actions that define the transition
The path to CMN leadership is not a technology road map. It’s a set of strategic choices to make the network agent-ready while preserving the trust that makes retailer data valuable. Five key actions separate the networks that will lead this transition.
1. Make data a platform, not a database.
Data attributes, such as pricing, promotions, availability, fulfillment terms, ratings, reviews, and brand claims, need to be machine-readable, properly sourced, and fresh. If an AI assistant cannot verify a claim, it will not use it. If it cannot trust the price, it will not surface the product.
This is especially critical for offsite and new agentic onsite placements (such as sponsored prompt). An agent evaluating a product has no shelf to browse. It relies entirely on structured signals that need to be deterministic: verified attributes, pricing, availability, fulfillment terms. CMNs whose data infrastructure was built for onsite sponsored listings will find it insufficient to power eligibility in agent-led environments. Future value depends on whether the underlying product data is machine-legible wherever the query happens.
2. Set the rules of sponsored influence before someone else does.
What paid influence can legitimately affect—whether a product is eligible to appear at all and where it lands in a ranked comparison—is becoming both more commercially significant and more reputationally sensitive as AI intermediates more decisions. Networks that set clear limits on what sponsorship can and can’t change, with clear disclosure standards and trust controls, hold a structural advantage. Those that leave the rules ambiguous will face pressure from advertisers demanding transparency, regulators demanding accountability, and consumers demanding trust.
3. Build monetization products for decision moments, not page placements.
Sponsored answers and structured comparison feeds are the emerging inventory of agent-led commerce. These ad placements are purpose-built for environments where the buyer may be a machine and the ad may be a structured claim. First movers will set the pricing logic, the measurement frameworks, and the buyer expectations that late entrants will have to accept.
4. Make the network as easy to buy as a search keyword.
Advertisers should be able to activate a campaign and see optimization signals in near real time through APIs that human buyers and automated systems can both access without friction. The benchmark is the ease with which an advertiser allocates budget to a search keyword and sees results by morning.
5. Replace return on ad spend (ROAS) with proof of business impact.
As advertisers grow more skeptical of opaque platform measurement, the ability to demonstrate closed-loop outcomes beyond ROAS—verified sales lift and incremental revenue, not just clicks—is increasingly a competitive differentiator. Networks that make that case command better rates and longer commitments from CMOs and CFOs. For advertisers outside retail—those without a closed data loop of their own—this kind of independent outcome measurement is the entry ticket. CMNs that can supply verified outcome measurement become the infrastructure these categories cannot build themselves.
The infrastructure play
Future CMN leaders are building now. Agent-ready data infrastructure, governance architecture, and decision-moment monetization formats take time to build and are difficult to replicate once a competitor has them.
Commerce media built its first advantage on proximity to the transaction. The second advantage—harder to build, harder to copy, and worth more—is proximity to the decision. The networks investing in that infrastructure today are building to compete for the long haul.
Rewiring Advertising is a multipart series examining how AI is restructuring value across the advertising ecosystem and building on our recent research. Each article examines the same underlying pressure through a different lens—where value is moving, what the transition requires, and what leaders must decide now.
Jack Trotter is a partner in McKinsey’s Denver office, and Marc Brodherson is a senior partner in the New York office, where Quentin George is a distinguished partner and Aparna Srinath is an associate partner.
The authors wish to thank Arianna Sanchez and Veronica Retana for their contributions to this blog post.