Picture the budget. A storefront redesign gets a line item, an agency, a launch date, and a celebratory all-hands when it ships. The homepage hero gets argued over in three meetings. The product feed, the structured file that lists every SKU with its price, availability, images, and attributes, gets a half-sentence in a ticket and an owner who is technically in marketing operations but mostly just whoever set it up.
That allocation is now backwards. The homepage is the asset you can see and the feed is the asset your buyers actually meet. As shopping moves into AI assistants, marketplaces, social apps, and comparison surfaces, the structured product feed has quietly become the highest-leverage thing a commerce team owns. This is the strategy argument for treating it that way. The mechanics of feed formatting are a separate subject; the point here is where the leverage moved and what it means for how a company is run.
The homepage was never the front door
Start with what the homepage was for. In the first two decades of ecommerce, the storefront homepage carried real weight. It was the address people typed, the page search engines pointed at, the canvas where a brand set tone before a shopper went deeper. A redesign moved numbers because the homepage genuinely was the entrance.
The data says it stopped being that. Yottaa's analysis of US retail traffic across the 2025 Cyber 5 period found that only 4.8 percent of search-driven sessions began on a homepage. Product pages took about 61 percent of search entrances and category pages roughly a third. The pattern is sharper for AI-referred traffic: in the same window, 77 percent of sessions from AI sources landed straight on a product page, and product pages drew 28 times more visits than homepages from those sources. The homepage was not killed by a competitor. It was bypassed by the way discovery works now.
This makes sense once you watch how people shop. A buyer rarely thinks "I will visit a brand's site and browse." They think "wide running shoes under 150 dollars" and type that somewhere. Search engines, marketplaces, and assistants all answer that specific intent by surfacing a specific product, not the broadest page on a site. The query lands the shopper deep, in front of one item, with the homepage skipped. The homepage still does useful work as a trust signal for returning customers and a navigational fallback. It is just no longer the entrance.
Discovery scattered, and the feed is what travels
If the homepage lost its primacy, where did the entrance go? It did not move to one new place. It shattered across many.
Marketplaces are a primary starting point. Salsify's 2025 consumer research found that 54 percent of shoppers discover new products through online marketplaces, with search engines at 65 percent and a fast-growing share starting on social platforms. eMarketer, citing IAB research, reported that 38 percent of US consumers already use AI for shopping and 52 percent plan to. None of these is a storefront. Each is a surface a brand does not control, with its own format, its own ranking logic, and its own idea of what a product record should contain.
Here is the part that reframes everything. A storefront cannot follow a shopper onto Amazon, into a TikTok feed, inside ChatGPT, or onto a comparison site. The designed page does not travel. The data does. Every one of those surfaces is populated by a structured representation of the catalog: a feed, an API response, an export. The marketplace listing, the shopping ad, the product an assistant reads, all of it is a feed. When discovery scattered across surfaces a brand does not own, the only asset that could scatter with it was the product data itself.
So the question "where is our front door" has a precise answer in 2026. It is wherever your feed reaches, rendered in whatever shell that surface wraps around it. The brand controls the contents, not the frame.
How the feed got here, quietly
The product feed is not new, which is part of why it was underrated for so long. It has been doing this job in the background since the early 2000s. Google launched Froogle in December 2002. Unlike the paid-submission comparison sites of the time, it let merchants submit inventory as a feed: a list of titles, descriptions, prices, and other fields. Froogle became Google Product Search in 2007 and Google Shopping in 2012, and Merchant Center arrived in 2009 to manage the feed. For most of that history the feed was treated as an advertising input. You maintained it grudgingly because Google Shopping ads needed it, and it sat with whoever ran paid search.
That framing, feed as ad plumbing, is why it never got strategic attention. It looked like a technical export, a means to one channel's end. What changed is not the file. It is the number of things that now read it. Marketplaces, social commerce, and AI assistants all read a feed. The same structured export that fed shopping ads in 2008 is now the input to nearly every place discovery happens. The feed did not change. The world rearranged itself around it.
The agent makes the feed non-negotiable
The trend had been building for years. Agentic commerce turned it into something a company cannot defer.
Through 2025 and into 2026 the buying flow itself began moving off the storefront. OpenAI shipped Instant Checkout inside ChatGPT in September 2025, starting with Etsy sellers and extending to Shopify's merchant base, so a shopper can complete a purchase without leaving the chat. In January 2026 Google announced the Universal Commerce Protocol at the National Retail Federation show, co-developed with Shopify and backed by Walmart, Target, Etsy, Wayfair, and major payment networks, and it went live in Merchant Center so retailers could enable it from the feed they already maintained. McKinsey projects that AI agents could orchestrate as much as 1 trillion dollars of US retail revenue by 2030, and 3 trillion to 5 trillion dollars globally.
When a machine is the shopper, the storefront stops being the interface. The agent never loads your homepage, never sees your hero image, never feels your typography. It reads structured data and reasons over it. McKinsey's framing of what agents optimize for is blunt: delivered value, meaning price, availability, fulfillment reliability, and the reversibility of returns. Merchants that expose clean inventory data and transparent policies become default suppliers without ever winning a traditional moment of consideration. The feed is no longer one channel's input. It is the whole surface a machine can perceive.
Two consequences follow, and both are operational. First, freshness becomes a live obligation. The ChatGPT product feed accepts updates as often as every 15 minutes. A price that was accurate this morning is a liability by lunchtime if the feed has not kept up, because an agent that quotes a wrong price or recommends a sold-out item learns to distrust the source. Second, attribute depth decides whether you are found at all. If a shopper asks an assistant for a machine-washable jacket under a certain weight and your weight field is blank, your jacket never enters the comparison. A human shopper would have inferred it from a photo. The machine does not infer. It filters on what the data explicitly says.
The vendor world has noticed. In April 2026 Feedonomics, the feed-management arm of the company behind BigCommerce, launched Agentic Catalog Exports, an enterprise service for syndicating catalog data to ChatGPT, Gemini, Copilot, Perplexity, and other agentic surfaces. Dell was an early adopter, preparing roughly 7,000 products for AI-driven discovery. When a feed company builds a dedicated product just to push catalogs into assistants, the feed has stopped being plumbing.
What a feed-first operating model actually means
If the feed is the highest-leverage asset, run the company as though that were true. A few things follow, and they are organizational before they are technical.
One source of truth, syndicated everywhere. The failure mode in most companies is many partial truths: the ERP knows cost and case pack, a spreadsheet holds the merchandiser's descriptions, the storefront has its own copy entered once and never reconciled, and each marketplace listing was written separately. Every divergence is a defect a machine can see. A feed-first model insists on one governed, authoritative copy of product data that flows to every surface, so a corrected dimension is fixed once and propagates, rather than fixed in six places by six people, three of whom forget. This is the discipline a product information management system exists to enforce, and it is why that unglamorous category has climbed the priority list.
Attribute richness as a deliverable, not an afterthought. The Salsify research found that 77 percent of shoppers rate product titles and descriptions as important to a purchase, and 84 percent cite price and availability. In a feed-first model, completing the catalog, the compatibility notes, the materials, the use-case context, the variant-level specifics, is a planned program with an owner and a deadline, not a thing someone gets to after the storefront launch calms down. Google's Universal Commerce Protocol even added Merchant Center attributes for answers to common product questions and for compatible accessories and substitutes, which tells you where the bar is moving.
Consistency treated as a trust signal. Salsify found that 54 percent of shoppers have abandoned a purchase because product information was inconsistent across channels, and 71 percent have returned an item because it did not match the listing. Humans punish inconsistency with a refund. Machines punish it by trusting the source less and ranking it lower. Either way, the same product saying the same thing on every surface is now a revenue variable.
Governance, because syndication multiplies mistakes. The moment one feed feeds twenty surfaces, a wrong price propagates everywhere at once. A feed-first model needs validation rules, ownership, and a clear process for who can change what, the same way a company governs its financial data. Speed without governance just distributes the defect faster.
The organizational problem this exposes
Here is the uncomfortable part, and the real reason this is a strategy piece and not a how-to. In most companies, nobody owns the feed.
The storefront has an owner: a head of ecommerce, a design lead, an agency. The brand has an owner: a marketing leader. The feed sits in the gaps. Merchandising owns pricing and assortment. Marketing owns the shopping-ad channels the feed historically served. Operations owns inventory. IT owns the export. The feed touches all of them and belongs to none, which is exactly how an asset gets neglected. That was survivable when the feed only powered shopping ads. It is not survivable now that the feed is the front door to every discovery surface a company has.
So the feed-first shift is, underneath, an ownership question. Someone senior has to be accountable for product data as a product: its completeness, freshness, consistency, and reach. That accountability cuts across merchandising, marketing, operations, and engineering, which means it has to be granted from above, not assembled informally from below. The companies that get this right will not be the ones with the best-looking storefronts. They will be the ones that decided, deliberately, that the catalog is a first-class asset and put a name against it.
The counterargument: brand did not die
A strategy piece that only argued one side would be selling something. The feed becoming the front door does not make brand worthless. It changes what brand has to do.
Brand still drives demand. An assistant returns three machine-washable jackets under a weight limit, all with clean data, and the shopper still has to choose. The name they recognize and trust carries real weight in that final step. Salsify found that 87 percent of shoppers will pay more for a product from a brand they trust, and that preference does not evaporate because a machine narrowed the list. Brand also shapes the query itself. A shopper who asks for "the new Glossier cleanser" by name has already been won before any feed was read. Demand generation still happens in culture, content, and reputation, and a feed cannot do it.
What changed is sequence. Brand increasingly creates demand, and the feed increasingly determines whether that demand can be fulfilled on the surface where it appears. A brand can spend heavily to make a shopper want its product, then lose the sale because the feed was stale, thin, or inconsistent and the assistant could not confidently recommend it. The honest formulation is not "feed instead of brand." Brand and feed are two halves of one system, and most companies have been funding the visible half and starving the half that does the delivering.
There is also a fair objection that this is overstated for some categories. A shopper buying a sofa or a wedding dress still wants to see, browse, and feel persuaded, and the storefront keeps more of its old role there. The feed-first argument is strongest for considered, comparable, attribute-heavy purchases, a large slice of commerce, and weaker for high-emotion ones. The shift is real and directional, not uniform.
Where this leaves a commerce team
The practical conclusion is not "stop investing in the storefront." It is to correct an allocation that no longer matches reality. For years, attention and budget flowed to the page a brand could see and admire, and the feed got whatever was left. The discovery data has inverted that logic. Most buyers, and increasingly most machines, meet the feed before they ever meet the storefront, if they meet it at all.
The move is to treat the feed as a first-class product. Give it an owner with real authority. Fund the unglamorous work of completeness, freshness, and consistency. Build the single source of truth and syndicate it deliberately to every surface that matters. Keep investing in brand, because brand is what makes anyone want the product, but stop pretending the storefront is the front door. The front door is now a file. The companies that reorganize around that fact will be the ones machines and shoppers can actually find.
Council summary
This post argues that the structured product feed, not the homepage, is the asset buyers and AI agents meet first, so commerce teams should fund and own it accordingly. The council checked every figure against primary sources, including the Yottaa, Salsify, McKinsey, OpenAI, Google, and Feedonomics claims, and corrected one overreach where search outranks marketplaces in Salsify's data. The takeaway: give product data a senior owner, keep it fresh, and syndicate one source of truth everywhere.
Comments