The Agentic Layer: The Three Possible Futures for Commerce Media and Digital Advertising

That moment will never be the same.

Today, generative AI engines account for just 3.3% of online discovery time in the US, according to Comscore. ChatGPT, Claude, and Perplexity combined receive a fraction of Google Search’s traffic. But that will change. Google is integrating AI into search. AI platforms are adding agentic commerce experiences. Conversational AI already accounts for 15-20% of retailers’ referral traffic as of September 2025. By 2027, the distinction between search and agentic discovery will most likely have disappeared. And by 2030, according to McKinsey, US retailers could see up to $1 trillion in orchestrated revenue from agentic commerce.

Over the past few months, the infrastructure for this has begun to take shape. OpenAI open-sourced its Agentic Commerce Protocol (ACP), a technical standard designed to let merchants integrate checkout directly into AI conversations. Shopify and Etsy were the first to adopt it, enabling their merchants to sell through ChatGPT. Then Walmart announced a partnership with OpenAI to let customers shop conversationally through ChatGPT. Perplexity integrated shopping with PayPal. 

Collectively, these moves affirm the emergence of a new browsing layer between discovery and checkout. And with that comes a complete reorganization of how sponsored products, media campaigns, and advertising inventory are discovered and surfaced.

Fully understanding this shift requires clarity on who controls the browsing experience, how sponsored products flow through these new channels, and what happens when advertising moves from keyword-based to context-driven.

The curation question

For the past decade, retail media’s story was straightforward: Amazon, Walmart, and Shopify built proprietary media businesses because they owned the moment of highest commercial intent. Retailer data. Retailer audience. Retailer advertising real estate. This created what is now a $180 billion market.

Now imagine a layer above that. A consumer asks ChatGPT: “I need a new coffee maker. Compact, under $150, works on a small counter.” The AI understands the request, searches across retailers it has access to, surfaces options, and potentially completes checkout entirely within the chat. No click-through to Amazon. No redirect to Walmart.

This raises a fundamental question: Who decides which products appear in that response?

First scenario: retailers maintain control. Amazon, Walmart, and Shopify syndicate their product feeds and advertising inventory into AI assistants via expansions to protocols like ACP. Retailers push sponsored ads and product data into the protocol stack. Campaigns already running through retailer DSPs are automatically surfaced within agentic experiences. The retailer keeps the brand relationship and monetizes inventory the same way they do today. The AI platform becomes infrastructure – powerful, but infrastructure.

Second scenario: AI platforms build their own ad layer. OpenAI, Perplexity, and Google decide the moment of discovery within agentic responses is too valuable to hand over to retailers. They build their own ad networks. Brands bid directly for placements. The AI platform becomes the arbiter of which products, which brands, and which ads appear. This is the Google AdWords model applied to conversation; retailers lose control entirely.

Third scenario: a hybrid. Retailers push campaigns into AI platforms. Platforms reserve the right to monetize impressions natively as well. Shopify merchants submit product feeds through ACP, but OpenAI also accepts direct advertiser bids. Perplexity aggregates retailer data but inserts its own sponsored results. Control fragments and both streams coexist.

The structure you end up with determines everything about how advertising operates for the next decade. But the answer isn’t determined yet. That ambiguity is where strategy lives.

The market is already showing us which scenario each retailer prefers. Amazon is reticent to push curated product feeds to platforms that could become competitors, preferring instead to own the full customer journey from discovery to purchase. The company pulled out of Google Shopping entirely in July 2025, then blocked AI crawlers from OpenAI, Anthropic, and others from accessing its retail data in August 2025. Simultaneously, it’s building its own agentic features (Rufus for onsite discovery; “Buy for Me” to shop across third-party sites) to keep control of the discovery layer entirely in-house.

Other major retailers are taking similar defensive positions: Walmart launched Sparky, Lowe’s introduced Mylow, and Home Depot built Magic Apron. Shopify did the opposite: it integrated with OpenAI through ACP, allowing merchants to opt into Instant Checkout. The divergence isn’t subtle. It suggests retailers won’t uniformly syndicate advertising inventory into AI platforms. Some will protect their own agentic layer. Others will use AI platforms as distribution for their media business.

For advertisers, this means the next 18 months will be chaotic. Which retailers expose which inventory to which AI platforms remains unsettled. The media buyer’s job becomes significantly more complex. You’re no longer just optimizing across channels. You’re trying to predict which channels will even exist.

Sponsored products in the agentic era

Currently, brands run campaigns through a demand-side platform, targeting consumers based on search behavior, browsing history, and category interest. Sponsored products, which currently account for 70% of retail media revenue, appear next to organic results. The advertiser pays per click. ROI is measured against conversion data. 

This infrastructure exists across Amazon, Walmart, Shopify, and dozens of smaller retail media networks. It’s mature, optimized, and deeply integrated into how brands allocate budget. And it works.

In an agentic world, this will be different: A brand selling premium kitchen equipment runs a campaign through its retailer’s DSP today. Tomorrow, if that retailer exposes its advertising objects through protocols like ACP, those same ads could surface within an OpenAI query response. The same media object appears across multiple surfaces: traditional search results, retailer-owned property, and AI assistant conversational responses.

This creates both opportunity and tension.

The opportunity is obvious: single campaign, multiple surfaces, no new creative or targeting logic needed. Brands already optimizing across channels suddenly gain access to another channel.

The tension is more interesting. For this to work at scale, retailers must expose advertising inventory to AI platforms in structured, standardized ways: Amazon makes sponsored product objects available to OpenAI. Walmart allows Perplexity to surface its ads within conversational responses. Shopify’s advertiser base grants permission for campaigns to flow into third-party AI layers.

None of this can happen by default. It requires negotiated partnerships and technical standards. It also requires retailers to trust AI platforms with their most valuable asset: the moment of commercial intent. If AI-driven discovery displaces traditional browsing entirely, retailers risk being reduced to fulfillment networks layered with loyalty programs, with their onsite traffic losing much of its discovery-phase value. High-margin retail media revenue, built on controlling that discovery moment, would be directly threatened.

When Walmart partnered with OpenAI, the deal went beyond transactional checkout. Walmart and Sam’s Club members can now shop through ChatGPT by chatting about their needs (planning meals, restocking essentials, discovering new products) while Walmart handles the rest. Walmart calls this “agentic commerce”: AI that shifts from reactive to proactive, learning and predicting customer needs.

But here’s the complication: if Walmart is surfacing products through its OpenAI partnership, are Walmart’s sponsored products also flowing into those results? Should competing retailers get equal access to OpenAI’s response layer? Or does Walmart’s partnership give it preferential placement? If agents compress discovery and checkout into single conversations, traffic to retailers’ sites could crater. Without consumers browsing page after page, there won’t be enough impressions to serve.

If retailer-controlled supply dominates, the business model doesn’t fundamentally change. Retailers still own curation. The AI platform becomes a new surface on which to display retailer-mediated ads. The brand-media buyer-retailer relationship remains intact.

If AI platforms build independent ad networks, the power dynamic dramatically changes. Retailers lose control. Brands gain direct access to OpenAI or Perplexity. You’re back in the pre-retail-media era, when brands competed through Google and Facebook without retailer mediation. That would be a genuine market reorganization.

The answer will likely be messier: a mix of both. Different retailers will make different bets. But the mix determines who captures value. And that’s what actually matters.

Beyond keywords and clicks

If you learned digital marketing in the Google AdWords era, you know the core logic: identify keywords, bid on keywords, win the auction, and your ad appears. Measurement is simple: click or no click, convert or don’t. Was ROAS positive?

That model is becoming insufficient. Google’s AI Mode reaches over 100 million monthly users. When a consumer asks, Help me find a running shoe that’s good for marathon training and fits narrow feet, the query isn’t decomposed into keywords. It’s understood as complex and multi-dimensional: product category, use case, fit profile, and personal preference. The AI generates a narrative response – comparing shoes, discussing fit technology, mentioning brands – then surfaces ads.

But those ads aren’t triggered by keyword bids. They’re surfaced based on much richer signals: full query context, conversation history, inferred intent, user search history, brand mentions within the response, and the quality of advertiser product data. Google calls this “multi-dimensional targeting.” Brands using Performance Max and Shopping campaigns with broad match are automatically eligible for AI Mode placements. But eligibility doesn’t mean effectiveness.

To actually win visibility in agentic responses, brands need to optimize for signals that extend beyond traditional keyword strategy. Product feed quality becomes critical. Not for organic discoverability. For AI comprehension.

When an LLM parses thousands of product listings, it understands structured data: descriptions, specifications, materials, dimensions, and ratings. A brand with rich, organized product data gets surfaced. A brand with sparse data doesn’t. This is a hard constraint, not a soft suggestion.

Part of what makes agentic commerce compelling is how it solves a genuine consumer pain point: finding the right product, whether it’s an outfit for a specific event or a washing machine with particular features, can take an hour or more without AI assistance. Conversational AI compresses that research phase dramatically, which explains why adoption is accelerating even among consumers who still complete purchases through traditional channels.

In fact, retail businesses saw a 128% monthly increase in actions on AI tools and assistants throughout the first half of 2025, according to Salesforce. Consumers who regularly use AI agents are twice as likely to say their experience with retailers has improved.

Some forward-thinking brands are already building what amounts to proprietary mini-language models based on first-party data: product feeds, site content, customer reviews, and connected data sources. These mini-models are designed to be plugged into larger LLMs or shared with AI platforms as trusted data sources. The brand becomes a curator of its own narrative within the agentic layer.

This is fundamentally different from the keyword-bid model. Instead of “I want to show up when someone searches for X,” it becomes, “I want my product, my brand story, and my value proposition accurately represented when an AI agent answers a question in my category.”

Measurement has to evolve in lockstep. Clicks become a less reliable proxy for advertising impact in an agentic world. A shopper reads an AI summary mentioning your product, considers it against alternatives, and either clicks through to the retailer or continues the conversation to compare. Traditional analytics show zero direct clicks. But you influenced the consideration set. You shaped the narrative. You drove the consumer toward a decision point.

That influence matters. And it needs to be measured.

Forward-looking marketers are expanding KPI frameworks to capture this: engagement depth (how long a user spends reading the response, how many follow-up questions they ask, how many alternative products they explore), the number of agentic queries a brand is mentioned in, the sentiment of that mention, and context. All of these become performance indicators. Eventual conversion still matters, but the path to it is less linear and harder to attribute through traditional funnel analytics.

What happens next

A new browsing layer is reorganizing how commerce discovery and advertising operate: not an addition to existing infrastructure, but a new medium with different curation logic, different bidding mechanisms, and different performance signals.

The near-term outcomes are reasonably clear. Retailers with strong proprietary data will experiment with exposing inventory to AI platforms while simultaneously building their own agentic experiences. Amazon is doing this. Walmart is doing this. Google will layer ads into AI Mode as part of its Gemini monetization strategy. OpenAI and Perplexity will monetize their shopping integrations. 

Retailers’ defensive advantages will remain in customer experience, loyalty ecosystems, fulfillment networks, and omnichannel assets. But these won’t be enough. Retailers and brands will need to aggressively push for native agentic experiences to make sure their ads, agents, and commerce integrations are built into the new discovery pathways of AI.

But they can’t do it alone. They’ll need partners who can navigate the complexity they’re about to face. Specifically, partners who have genuine integration with both AI platforms and retailers. Partners who understand how to structure agentic campaign objects – it can’t be yesterday’s sponsored products in new packaging. Partners who can act as connective tissue, aggregating demand where AI platforms prefer fewer direct brand relationships. And critically, partners who are building measurement capabilities for agentic-specific metrics: engagement depth, conversation outcomes, and assisted conversions. The vendors that can credibly deliver on all four will become indispensable. The ones still optimizing for clicks will find themselves increasingly marginalized.

Looking further ahead, the medium-term structure is genuinely unsettled. Will AI platforms continue to serve as distribution channels for retailer inventory, or will they build independent ad networks? Will brands opt out of traditional retail media to bid directly with AI platforms? Will retail media networks evolve to wrap around agentic experiences, or will they be displaced?

For those responsible for advertising strategy, the key is recognizing this as an inflection point. The infrastructure, relationships, and measurement frameworks you’ve built aren’t obsolete. But they’re being tested by a new layer of commerce operating on different principles.

The winners will be those who recognize that agentic AI isn’t an addition to existing advertising channels but rather a new medium. Those who adapt their strategy and infrastructure will capture value. Those who treat it as a new inventory stream bolted onto yesterday’s playbook will find themselves marginalized.

The agentic era in digital advertising is already here. What remains to be determined is who curates that experience. And, in turn, who benefits from it.

---

Frequently Asked Questions