Shoppers are changing where and how they search and discover products. Instead of typing keywords into a search box, they describe a problem to chat interface and get a handful of products back. It looks like a small change of habit, but structurally it is the biggest shift in product discovery since search itself, because of what happens to everything that is not in the answer.
A search results page degrades gracefully. Position four still gets clicks, page two still exists, a determined shopper keeps scrolling. An AI answer does not degrade gracefully. It names three products, perhaps five, and the conversation moves on. There is no page two. For a retailer this changes the nature of the competition, as discovery stops being a ranking game and becomes a casting call. Either a product is a candidate, or it is invisible.
The contest is unfolding on two fronts at once. The first is on the retailer's own website, where an AI layer can do what a good salesperson does, understand the problem, ask the questions that matter, and guide a shopper from vague intent to a confident decision. That layer can be owned by retailer, the same way search, filters and the product page have always been owned. The second front are the answers of AI assistants that retailers don’t and can’t control. The only question then is whether a product's data earns it a place among the candidates. Different as they look, the two fronts reward the same investment, because both are decided by the depth and structure of the product data underneath.
Candidacy is computed from data
How does an AI decide which products make the answer? It reads product data and matches it against the shopper's constraints. When someone asks for a quiet dishwasher under 600 euros that fits a 45 cm niche, the model looks for candidates whose data actually carries a noise level, a price and a width. A product missing those attributes is not ranked lower; it is simply not in the running, no matter how good it is or how well it sells in the store.
The industry has a new name for competing in these answers: GEO, generative engine optimization. The rules rhyme with the last twenty years of search and SEO. Nobody can buy the ranking, and the shops with complete, machine-readable product data win the questions. If anything, the game is harsher than SEO ever was, because an answer has fewer slots than a results page, so the well-structured few take almost everything.
"AI ready" means two different things
Most of what retailers currently hear under the "AI ready" label is about transactions like checkout endpoints, manifests, payment rails, protocols like ACP and UCP that let a purchase complete inside the AI surface. This is real and it is arriving fast. It is also not the part that decides a product's fate. It is plumbing between the platforms, which e-commerce stacks and agencies will provide the same way they provide payment processing today.
The other meaning of AI ready is the part that decides whether the product is chosen or not. Whether an agent puts a product in the cart is settled before any checkout protocol fires, at the moment the model reads product data and picks its candidates. Eligibility and selection are different gates. The rails make products purchasable, but only data makes them chosen.
Why no plugin solves the second half
It is tempting to treat this as one more integration. Install something, check the box, be AI ready. But protocols only transport product data, they do not create it. And creation is where most catalogs fail, because the knowledge that would win the question is rarely in the catalog at all. It lives in spec PDFs, in size charts, in supplier portals, in the heads of category managers. A shopper who asks a good salesperson gets that knowledge. A machine reading a product feed usually does not.
Getting it into machine-readable form is unglamorous work, which requires structured attributes with real values, facts that are approved rather than scraped, and text in the languages shoppers actually use. Then comes the equally unglamorous work of making sure the machines can actually reach it, which requires structured markup rendered server-side, since most AI crawlers do not execute JavaScript, and crawler access through a CDN that probably blocks AI bots by default. Details like these are deciding who appears in answers.
The work pays twice
The same schema.org JSON-LD that gives an LLM the clarity to cite a product powers rich results on Google. The same structured feed depth serves classic Shopping listings and AI-driven surfaces alike. A catalog with incomplete structured data is not underperforming in one channel, it is underperforming in both.
Which is also why this is best treated as infrastructure rather than a campaign. The parallel to the early years of SEO is almost exact; the shops that did the structural work before it was fashionable owned the results page for a decade.
One more principle belongs here. Data stays healthy only if something uses it every day. A feed exported once and forgotten tends to quietly rot, while product data that also powers a live guided experience on the retailer's own site, where wrong or thin answers surface in real conversations and get fixed, keeps improving. This is where the two fronts reinforce each other, as the conversations on the first keep revealing exactly the data gaps that lose answers on the second.
The floor is rising
Since August, the EU AI Act's transparency rules are binding across Europe. Assistants must disclose themselves, answers should be grounded in real data, and retailers carry responsibility for the AI they deploy. Grounded answers stand on structured, verified product data. Regulation, in other words, is arriving at the same conclusion the market already reached, namely that the data layer is not optional.
Gem manufactures exactly this layer: a structured factsheet with approved facts for every product, exported as schema.org JSON-LD for your product pages and a Google Merchant supplemental feed, with a hand-off brief your web team can act on the same day. The same data powers the Gem Shopping Expert on your own site, so it is used, tested and corrected every day.
Book a demo to see what your own products look like to a machine shopper.