The store was built for browsers. The next customer does not browse. And most retailers are optimizing the wrong experience for the wrong user.
A note on perspective: This article is written from the CEO and Chief User Experience Officer chair simultaneously. Retail and ecommerce UX has been optimized around one assumption for 25 years: a person sits in front of a screen, searches or browses, reads product pages, and decides. That assumption is breaking in a specific and measurable way. The CEO who reads this as a web optimization problem will invest in better product pages, faster load times, and improved search. The CEO who reads this as a structural shift in who the customer actually is will make a different set of organizational decisions entirely. Those two sets of decisions produce incompatible products, and only one of them positions the organization for the commerce environment that is building right now.
The product page has not fundamentally changed since Amazon built the template in the late 1990s.
Product image. Product title. Price. Description. Reviews. Add to cart. The layout has been refined. The photography has gotten better. The reviews have gotten more sophisticated. Mobile has forced responsive design. But the structural logic of the page, a human being arrives, reads information, and makes a decision, has been the unchanging foundation of ecommerce UX for a generation.
That structural logic was correct for the user it was built for. It is increasingly incorrect for the user that is arriving now.
By 2027, a significant and growing portion of purchase decisions in categories from consumer electronics to apparel to household goods will not be made by a person sitting in front of a product page. They will be made by an AI shopping agent acting on behalf of a person who stated a goal, authorized the agent to find the best option, and received a recommendation or a completed purchase without ever seeing the product page at all.
That is not a prediction about a distant future. Google AI Overviews already intercept a measurable share of product search traffic before it reaches a retailer’s page. Amazon’s AI shopping assistant selects products based on criteria rather than surfacing browse results. Apple Intelligence is building shopping assistance into the operating system itself. The agent-mediated purchase is not arriving in 2027. It is already here, and 2027 is when the volume reaches the threshold where retailers will no longer be able to treat it as an edge case.
Most retailers are spending 2026 optimizing the experience for the human browser. That work is not wasted, because human browsers are not disappearing. But it is insufficient, because the most valuable customer segment, the customer whose purchase intent is high, whose cart value is high, and whose decision timeline is short, is increasingly the one whose agent completes the transaction without a browsing session at all.
What the AI Shopping Agent Actually Does
The AI shopping agent is not a smarter search bar.
A search bar surfaces options. The human browser evaluates them. The agent does not surface options for human evaluation. It evaluates them itself, against the criteria the person stated or that the agent inferred from prior behavior, and produces a recommendation or completes a purchase directly.
This requires a completely different kind of product information than the product page delivers.
A product page is designed for human reading. It uses language and imagery that creates desire, context, and narrative around a product. It surfaces social proof through reviews. It creates urgency through scarcity signals. It guides attention through visual hierarchy. All of these are sophisticated design tools aimed at a human cognitive process.
An AI agent reading a product page encounters a problem: most of the information it needs to make an accurate comparison is either absent, inconsistently structured across different retailers, buried in dense review text, or formatted for visual parsing rather than machine extraction. The agent that is trying to compare the energy efficiency of seven air conditioners across five retailers is working against page designs that were never intended to support systematic, structured comparison.
The retailers whose product data is clean, structured, complete, and machine-readable will be the ones whose products get recommended by agents. The ones whose product data lives primarily in image files, marketing copy, and inconsistently formatted specification tables will be the ones whose products the agent skips, not because the product is inferior but because the agent cannot reliably extract the information needed to include it in the comparison.
This is a data architecture problem that expresses itself as a UX problem. And it requires a CEO-level decision to fix, because it is not a web team optimization. It is a fundamental rethinking of what a product catalog is for.
The Two Retail Strategies for 2027
There are two ways a retail organization can respond to the arrival of AI shopping agents as a primary customer type. They produce different products, different competitive positions, and different organizational structures.
The first strategy is page optimization.
The organization continues to invest in the human browsing experience. Better product photography. Improved page load performance. More sophisticated personalization for returning visitors. Enhanced search and filtering. Stronger review collection and display. AI features added to the page itself, a chatbot that answers product questions, a style advisor that suggests complementary items.
This strategy is not wrong. Human browsers still represent the majority of sessions and will for years. It is also a strategy that is running on a diminishing return as the share of agent-mediated purchase decisions grows. Every dollar invested in the browsing experience is a dollar not invested in the agent-ready product data infrastructure that determines whether the product exists at all in the agent-mediated purchase flow.
The page optimization strategy produces an increasingly beautiful experience for the customer type whose share is declining, while the product becomes progressively less visible to the customer type whose share is growing.
The second strategy is infrastructure redesign.
The organization makes a different bet: the product catalog is now a dual-use asset. It serves human browsers through the page experience and AI agents through structured data APIs and machine-readable product information. Building for both users simultaneously requires treating structured product data as a first-class design and business asset rather than as a backend technical concern.
Practically, this means every product in the catalog has complete, accurate, consistently structured attribute data. Not marketing copy about the attributes. The attributes themselves, in a format that an agent can read, compare, and reason about. Dimensions, materials, compatibility, energy ratings, care requirements, return conditions, shipping timelines, whatever a buyer in this category needs to make an informed decision.
It means the product catalog has an API layer that authorized agents can query, not just a web page that humans visit. It means the trust and authorization architecture for agent-mediated purchases is designed: how the customer authorizes an agent to complete a purchase on their behalf, what confirmation the customer receives, what the return and dispute process looks like for a purchase the customer did not personally complete.
None of this eliminates the browsing experience. It extends the product’s reach to include the agent-mediated interactions that are becoming a structurally important share of purchase volume in high-consideration categories.
Why This Is a CEO Decision, Not a Catalog Decision
The product catalog has historically been owned by a combination of merchandising, e-commerce operations, and technology, with design responsible for the page presentation on top of it. That ownership model produces a catalog that is optimized for the page experience and treats the underlying data as the technical substrate that powers the page.
The strategic shift I am describing requires treating the product data itself as the primary customer-facing asset, with the page as one of several interfaces through which that data is delivered to different user types. That is a philosophical inversion of how most retail organizations think about their catalog, and it cannot be executed without organizational authority behind it.
Specifically, it requires resource allocation to data quality and structure work that will not show up in page conversion metrics in the near term. The agent-mediated purchase does not appear in the same analytics as the browser session. It may not appear as a website visit at all if the agent queries the product API directly and completes the purchase through an integrated checkout. Measuring the value of the investment requires tracking purchase volume that originates outside the traditional session funnel, and most retail analytics stacks are not built to capture it.
It requires a decision about API access: which agents, under what authorization model, with what data access, can query the product catalog and initiate purchases on behalf of customers. That is a legal and product decision, not a technology decision, and it involves choices about brand representation in agent-mediated contexts that have no precedent in the page-based commerce model.
And it requires the CEO and the CUX officer to agree on what the customer relationship means when a significant portion of purchases are not preceded by a browsing session. The brand touchpoints, the experience design, the trust signals that have been built into the page experience for decades, most of them do not exist in an agent-mediated transaction. The relationship the customer has with the retailer in an agent-mediated world is determined primarily by the accuracy and reliability of the product data the agent uses, the smoothness of the checkout and fulfillment, and the experience when something goes wrong. That is a stripped-down version of the brand relationship that most retail organizations have never designed for intentionally.
The Three Design Decisions That Determine the Outcome
If I am leading this transformation, these are the decisions I make before any technology selection, any API specification, any catalog migration project.
The first decision is what structured product data actually means for this category.
Every retail category has a different information architecture for agent-mediated comparison. The attributes that determine a purchase decision for a laptop are different from those for a mattress, which are different from those for a running shoe. The design work here is not technical. It is research: what does a buyer in this category need to know to make a confident decision, and is that information present, accurate, and structured in the catalog in a form that an agent can retrieve and reason about?
Most retailers who have done this audit find that their product data is 60 to 70 percent complete for the attributes that matter most in agent-mediated comparison. The missing 30 to 40 percent is the gap between being included in an agent recommendation and being skipped. Closing that gap is not glamorous work. It is also the highest-leverage investment a retailer can make for the agent-mediated commerce environment.
The second decision is what happens after the agent recommends.
The agent-mediated purchase does not end when the agent makes a recommendation. It ends when the customer confirms, the transaction completes, and the product is delivered and accepted. Every point in that sequence where the experience is confusing, opaque, or inconsistent with what the customer understood when they authorized the agent creates a trust problem that affects both the retailer and the agent platform.
The design of the post-recommendation experience, the confirmation the customer receives, the tracking and delivery experience, the return and dispute process for agent-initiated purchases, needs to be designed specifically for the context in which it arrives. A customer who did not browse and decide is arriving at the confirmation screen without the context a browser session would have provided. The design has to supply that context: what was purchased, why the agent selected it, what the customer’s options are, and how to contact a human if something is wrong.
The third decision is what the human browsing experience becomes when it is no longer the primary channel.
The human who visits a product page in 2028 is not the same as the human who visits a product page today. A growing portion of quick, low-consideration, well-understood purchases will be handled by agents. The human who browses is increasingly the one making a high-consideration decision where the browsing session itself is part of the decision process, the person who wants to see the product, read the reviews carefully, compare options personally, and feel the act of choosing.
That is a different user with different needs than the one the current page optimizes for. The page experience that serves the deliberate browser in a world where the agent handles the routine purchase is richer, more editorial, more experience-focused, and less conversion-optimized than the page built for the volume browser. Designing for that user is a creative and strategic choice that most retail UX teams have not had to make explicitly because they have always been designing for everyone.
What the Organizations That Move First Will Have
By 2028, the retailers with clean, complete, machine-readable product catalogs and authorized agent purchase flows will have a structurally different position in the commerce ecosystem than the ones still optimizing for the human browsing session.
They will capture purchase volume from agent platforms that surfaces their products to customers who would never have found them through traditional search. They will have lower customer acquisition costs for high-intent purchases because the agent-mediated flow bypasses the entire top-of-funnel browsing session that traditional conversion optimization is built around. And they will have built the data infrastructure that positions them to participate in the next generation of commerce channels, whatever form those take, because a clean, structured, machine-readable product catalog is the foundation of every agent-mediated commerce model currently being built.
The retailers that wait for the agent-mediated purchase volume to become undeniable before they invest in the infrastructure will find that the investment required to catch up is larger than the investment required to lead, because the agent platforms will have already built their recommendation models around the retailers whose data was available when the models were trained.
The product page is not dying. The assumption that it is the only experience that matters is.
Sources: Google, “AI Overviews and Search Commerce Impact Report,” 2026; Forrester Research, “The Agent-Mediated Purchase: Retail Strategy for 2027,” 2026; McKinsey, “The State of AI in Retail and Consumer Goods,” 2026; Shopify, “Commerce Trends 2027,” 2026; Nielsen, “AI Shopping Assistant Consumer Adoption Survey,” 2026; Gartner, “Hype Cycle for Retail Technology,” 2026.