The basic commercial job has not changed: attract someone, give them enough information to decide, make paying easy, and earn the next purchase.
Though maybe the homepage matters more than I want to admit, once the labels are clean.
Conversion work is no longer just the checkout button. Onsite search, recommendations, reviews, payment options, and the messages after the order (replenishment, abandoned cart, support) decide whether growth is a spike or a habit.
Why is selling online getting harder as it grows?
A customer can now leave with your product without ever seeing your homepage. They ask a chatbot which tent is right for a two-person weekend trip, get a shortlist, tap a link, and pay. Your ads did not create that visit. The product data did, or failed to.
Product data used to be treated primarily as something displayed on the product page. It now behaves more like infrastructure.
None of that rescues a bad product. Demand cannot invent inventory. But more systems will sit between the customer and the catalogue-assistants, shopping layers, marketplaces and all of them depend on the same underlying product data. Somewhere, a truck is pulling away from a dock and leaving behind a pallet of jackets that never got a size field.
The path to purchase is rarely a straight line, and privacy rules have made it harder to pin a sale on a single click. There is no clean answer here, and pretending dashboards still tell a complete story is how budgets get wasted. First-party data, information the retailer collects with consent from its own customers, becomes more important as third-party tracking becomes less reliable. Analytics that only celebrate last-click ads will miss a page that cannot convert because shipping looks expensive on a phone.
Is this still a pile of separate tools?
Product data now shows up in search listings, ads, onsite search, and AI answers at the same time. Fixing the underlying product data can improve several of those channels at once. Spending on one channel while the data is wrong just buys more of the same problem.
An incomplete title is not only an ugly page. The same gap can prevent an ad from serving properly, hide a shopping listing, confuse onsite search, and give an AI a thin or incorrect answer. Teams have started treating taxonomy, structured data (machine-readable markup that describes what a product is), GTINs, and feed quality as commercial work, not IT housekeeping.
A shopping campaign cannot fix missing attributes. An attractive homepage cannot fix an inaccurate feed. A CRM cannot retain customers if orders and returns live in different places. Recommendation engines, which suggest products from behaviour and catalogue data, only know what you tell them.
It is tempting to buy a platform, a feed tool, an ad account, analytics, a CRM (software that stores purchase history and contact details), and some automation, then treat them as separate parts of the business. In practice, they depend on much of the same underlying data.
People are asking software instead of searching
Search itself is changing in the same direction. Visibility Labs research reported by Search Engine Land found Google AI Overviews, the short generated summaries at the top of some results, on 14 percent of shopping queries. Fourteen. Many ecommerce teams had assumed shopping searches would be mostly spared. The data increasingly suggests otherwise.
The stores that do well from here will be the ones that label the crate before they pay for another truck.
Agencies used to be hired to fill the top of the funnel. For retailers looking for online store growth specialists, the changing technology landscape makes it important to look beyond a list of marketing services. The partners worth hiring can talk about feeds, and about why two platforms disagree on revenue, not only about traffic.
A campaign that dumps thousands of extra visitors on a weak product record is a cost. The useful questions are whether those visitors convert, what they cost, what they bought, and whether the customer kept the order or returned it. A partner who can only report clicks is grading the wrong test.
Generative AI, software that writes answers from large language models rather than returning a list of links, can summarise a product and narrow a purchase. Adobe reported that traffic from generative AI sources to U.S. retail websites rose 693.4 percent during the 2025 holiday season.
When did the product feed become marketing?
Traditional search still sends a lot of traffic. It is no longer the only place people research products.
The path to completing it has. Someone might find a product in Google, in an AI-generated answer, on a marketplace, or from an email. They compare a couple of sites and buy later by typing the URL. Each stop has its own data rules.
I have a bias here, and I will state it as one: most stores would earn more from cleaning their catalogue than from adding another AI widget.
Ads got smarter. They also got pickier.
The commercial point is blunt. Online sales are still growing. The work that produces them has moved behind the page.
A large catalogue can run to thousands of variants. Those need consistent names, categories, attributes, identifiers, prices, availability, and images across every system that consumes them. Product feeds, the files that push that catalogue to Google, shopping services, marketplaces, and ad platforms, are the trucks leaving the warehouse.
Shopify said AI-referred orders on its platform grew nearly 13 times year over year in the first quarter of 2026. Those merchants did not have to build the AI. They still benefited when it referred shoppers who went on to buy.
Ad platforms can now automate bids, audience targeting, and even creative variations with relatively little manual intervention. Automation still needs accurate inputs: the feed, conversion tracking, inventory, margins, and what you actually want the campaign to do.
Not that search died. It didn’t. Stores now have to be understandable in several kinds of discovery at once, which sends you back to the product data: titles, variants, materials, dimensions, images, reviews, price, availability, shipping, and returns.
What should a marketing partner even know now?
Marketing numbers have to sit next to commercial ones. Cost to acquire a customer. Average order value. Whether they come back. Impressions are not profit.
You can spend more and sell more while becoming less profitable if acquisition cost outruns order value and margin. You can also leave spend flat and get more from it by fixing your data or changing which customers you target.
Think of an online catalogue as a warehouse, not a shop window. Google, marketplaces, ad platforms, and AI assistants are distribution channels. Each one depends on product information it can understand: title, size, material, stock, price, and a product identifier such as a GTIN (Global Trade Item Number, which tells systems this is the same item everywhere). A missing size or stale “in stock” flag can leave the product invisible or unusable in that channel. Once a person is already on your site, the priorities change. Trust, usability, and whether checkout works properly on a phone matter more than any label.
U.S. retail ecommerce sales were 17.1 percent of total retail sales in the second quarter of 2026, and 12.4 percent higher than a year earlier, according to the U.S. Census Bureau. When that much of retail sits online, a sloppy product feed or a broken checkout is not a website nuisance. It is lost money.
By Glenn Blake
