For most of the last two decades, a store's visibility was a search problem. You ranked on Google, you bought ads, you optimized for keywords, and shoppers found you in a list of ten blue links they sorted through themselves. That world is ending faster than most merchants realize. Shoppers have started asking AI assistants what to buy, and the assistant does not return a list. It returns an answer. It names a store, or a few stores, and the shopper acts on the recommendation. The intermediary is no longer a ranked page of options. It is a single confident suggestion.

This changes the central question every online store has to answer. It is no longer “where do we rank?” It is “when a shopper asks AI for a product like ours, does the AI name us, or someone else?” Nobody had measured that at scale. So we did.

Between April and the publication of this report, we ran the largest study of its kind. We took 9,720 ecommerce store domains, spanning global brands and small independent shops, and for each one we generated six purchase-intent queries matched to its products, the kind of question a ready-to-buy shopper actually types into an AI assistant. We submitted all of them to Google Gemini and recorded what happened: was the store recommended, was a competitor recommended instead, or did the assistant decline to name anyone. That produced 58,320 head-to-head recommendation tests across 58,043 distinct queries.

The findings are not incremental. They describe a market that looks almost nothing like search.

The first and largest finding is that most stores simply do not exist to AI. Sixty percent of the stores we tested were recommended in none of their six queries. Not ranked low. Not on the second page. Absent. Across all 58,320 tests, stores were recommended only 14 percent of the time. For six in ten of them, the number was zero.

The second finding overturns the obvious assumption about who wins instead. You would expect the recommendations to flow to the giants, to Amazon and Walmart and Etsy, the names with the budgets and the brand recognition. They do not. Across every loss in the study, we counted 50,287 different brands that AI recommended. The single most-recommended brand on the entire internet, Etsy, accounted for just 1.3 percent of recommendations. The hundred most-recommended brands combined made up only 11 percent. Seventy-one percent of the brands that won anything won exactly once. There is no incumbent that owns AI search. The field is wide open in a way search has not been for fifteen years.

The third finding is the one most merchants never think about, and it turns out to matter more than anything else they can control. The single biggest predictor of whether a store gets recommended is not its size, its age, or its ad budget. It is its product category. Win rates run from 17 percent in categories like sports, pets, and hobbies down to 9 percent in home and living. The share of invisible stores runs from 52 percent in food and beverage, the most visible category, to 74 percent in home and living, the least. Before a store does a single thing, its category has already set the odds it is playing against.

The remaining findings explain the mechanism. Changing the type of question barely changes the outcome, which means the problem is not how shoppers ask but what the catalog says. Winning is a head-to-head event: in most wins, AI named a specific competitor it chose the store over, which means visibility is a contest the engine resolves in the moment between a small set of stores it can actually describe. AI only answers specific questions: when it declined to name anyone, 80 percent of the time the query was too broad, which tells merchants exactly where the battle is. And the small group of stores that won every single query share one trait above all others. They are focused. They are single-category brands an AI engine can describe in one sentence, and that legibility is what gets them named.

Taken together, the data points to a single conclusion, and it is good news for any store willing to act on it.

That sentence is the difference between the 60 percent and the 0.8 percent. The rest of this report is the evidence behind it, category by category, and what it means for your store.

The study was built to answer one question with as little ambiguity as possible: when a shopper asks an AI assistant for a product, which store does it name?

We began with a sample of 9,720 ecommerce store domains sourced from BuiltWith. The sample was deliberately broad rather than curated. It spans global brands with hundreds of millions in revenue and small independent shops with a few hundred products, because the goal was to describe the whole landscape, not a flattering slice of it. A note on counting: some brands operate multiple regional storefronts, and we treated each domain as a separate store, since each one competes for recommendations on its own.

For each store, we generated six purchase-intent queries matched to its product category. These were not brand-name searches, which would be trivial to win. They were the questions a shopper asks when they know what they want but not where to buy it: “best wireless charging stand for iPhone and Apple Watch,” “where to buy cruelty-free nail paints,” “compare premium athleisure brands with performance wear for women.” We submitted each query to Google Gemini and classified the response three ways. If the assistant named the store, we recorded a win. If it named a competitor instead, we recorded a loss, and we captured which competitors. If it declined to recommend any specific brand, we recorded a no-recommendation, and we captured the reason. Across the full sample that produced 58,320 head-to-head comparisons spanning 58,043 distinct queries.

The first number in this study is the one that should reframe how every merchant thinks about AI. When we ran six purchase-intent queries against each of 9,720 stores, 60 percent of those stores were recommended in exactly none of them.

It is worth sitting with what that means, because “invisible” is easy to read past. These are not stores ranking on page two. In search, a store that ranks tenth still appears; a determined shopper scrolls to it. There is no page two in an AI recommendation. The assistant names a store or it does not, and for six in ten stores, across all six chances, it did not. The shopper asking for exactly what these stores sell was handed a competitor every time, and never knew the store existed.

Across all 58,320 tests, the aggregate breakdown was stark: stores were recommended 14 percent of the time, a competitor was recommended 79 percent of the time, and 7 percent of the time the assistant named no one. A 14 percent win rate across the whole market is low. But the average hides the more important structural fact, which is how those wins are distributed.

Visibility in AI is not a gentle gradient that most stores sit somewhere along. It is close to all-or-nothing. Sixty percent of stores won zero queries. Another 28 percent won only one or two of their six. Just 11 percent won three to five. And a thin 0.8 percent, seventy-eight stores out of 9,720, won all six. Put differently: of every store that managed to win anything at all, 70 percent won only once or twice. The middle is nearly empty. Stores are either largely invisible or, in a small number of cases, dominant, with very little in between.

Consider a concrete case. Anine Bing is a well-known Los Angeles fashion label with real brand recognition. In our scan it won zero of its six queries. When a shopper asked “where can I buy Los Angeles-based women's denim,” a near-perfect description of the brand, Gemini recommended AGOLDE and Nordstrom instead. The store was not obscure. It was simply not legible to the assistant as the answer to its own defining question. That is the texture of the 60 percent: not unknown businesses, but businesses the AI does not connect to the questions they should own.

This distribution is the spine of everything that follows. A market where 60 percent win nothing and 0.8 percent win everything is not a market where visibility is earned gradually through effort and budget. It is a market where something specific separates the stores that get named from the stores that do not. The rest of this report is the search for what that something is.

Most conversations about AI visibility focus on things a merchant controls: the content, the schema, the product descriptions. Those matter, and later findings show exactly how. But the largest single factor in whether a store gets recommended is decided before the merchant does anything at all. It is the category they operate in.

We classified the queries in the study into product verticals and measured how each one performed. The spread is wide enough that it dwarfs most of the variables a store can influence. In sports and outdoors, in pets, and in hobbies and crafts, stores won 17 percent of the time. In apparel and accessories, the figure was 10 percent. In home and living, 9 percent. A store in the best-performing categories is winning recommendations at nearly twice the rate of a store in the worst, before accounting for anything it does well or badly.

The invisibility figures tell the same story from the other side. The share of stores in each category that won nothing ranged from 52 percent in food and beverage, the most visible category in the entire study, to 74 percent in home and living, the least. Three out of four home and living stores were invisible. That is not because home and living merchants are worse at their jobs. It is because of how the category interacts with the way AI recommends.

The pattern underneath the numbers is consistent and, once seen, intuitive. AI does well in categories where products have clear, describable attributes and distinct use cases, the things a language model can latch onto and match to a specific query. A supplement has a measurable composition. A bike has a frame type and a use case. A coffee has a roast level and an origin. These give the assistant concrete hooks. AI does poorly in categories where products are visually driven and differentiated by taste rather than specification. Apparel and home decor are the clearest examples: a dress or a side table is chosen largely on look, and look is exactly what a text-based engine cannot evaluate, so it falls back on a few large, safe, generic retailers and leaves the independent stores unseen.

This finding reframes what a realistic goal looks like. A home and living store competing in a category that is 74 percent invisible is playing a fundamentally harder game than a food and beverage store in a category that is 52 percent invisible, and the two should not measure themselves against the same benchmark. But the finding is not a counsel of despair for the hard categories. Within every category, some stores win, and the later findings show that what separates them is legible, specific, well-structured product information, which is precisely the thing the difficult categories most often lack. Category sets the starting odds. It does not determine the outcome. The next section goes inside the major categories one at a time to show what winning and losing actually look like in each.

A reasonable hypothesis, after seeing how much category matters, is that the type of question also matters. Maybe stores lose the broad comparison questions but win the narrow product ones, or vice versa. If that were true, a merchant could pick their battles by targeting the query types they tend to win.

The data closes that door. We tested three kinds of purchase-intent query. Direct product queries, which name a product and ask where to get it (“best wireless charging stand for iPhone and Apple Watch”). Comparison queries, which ask the assistant to weigh options (“compare premium athleisure brands with performance wear for women”). And scenario queries, which describe a situation with multiple requirements (“which store offers installation services and a wide selection of kitchen appliances”). If query type were a lever, the win rates would diverge.

They barely move. Direct product queries won 14 percent of the time. Comparison queries, 13 percent. Scenario queries, 13 percent. Across three structurally different ways of asking, the outcome is effectively flat. There is no query format a losing store is quietly winning on, and no format a winning store is quietly losing. Stores win or lose across all of them at nearly the same rate.

This matters because of what it rules out. The problem is not how shoppers phrase their questions. A store cannot fix its invisibility by hoping shoppers ask in a more favorable way, because there is no more favorable way. The flatness points the cause firmly back at the store itself, specifically at the catalog, the product information the assistant draws on regardless of how the question is worded. Whether a shopper asks directly, comparatively, or by scenario, the assistant is reaching into the same well of what it knows about the store, and if that well is shallow, the phrasing of the question cannot save it.

One detail in this finding is quietly encouraging. Stores did win head-to-head comparison queries, the hardest type, 1,462 times. Being chosen explicitly over named rivals, when the assistant is directly weighing options, is not a fluke that happens by accident. It happens often, to stores whose information gives the assistant a clear reason to prefer them. Which raises the question the next finding answers: what does winning actually look like at the moment of the comparison?

There is a third outcome besides winning and losing, and it is more instructive than it first appears. In 7 percent of all tests, the assistant declined to name any store at all. It returned a generic, informational response, or asked the shopper to be more specific, without recommending a brand.

When we examined why, the reason was overwhelmingly consistent. In 80 percent of these no-recommendation cases, the query was simply too broad for the assistant to name anyone. Questions like “where can I find a wide variety of products” or “compare online retail stores with diverse offerings” gave the model nothing specific to match, so it named no one. The pattern is clear and it is a finding about how AI recommendation fundamentally works: vague, informational questions produce no winner, and specific, intent-rich questions produce a named store.

For merchants, this is one of the most directly actionable findings in the study, because it tells them exactly where the battle is and is not. The battle is not for broad category terms. “Home goods,” “skincare,” “men's clothing”: these broad terms, the keywords merchants spent fifteen years fighting over in search, often produce no recommendation at all in AI, which means there is no winner to be and no point optimizing for them. The battle is entirely for the specific, high-intent questions a ready-to-buy shopper actually asks, because those are the only questions AI answers with a store name. “Earthing sandals with Vibram soles,” “open-ear headset for office use,” “LED hair growth helmet”: these are the queries that resolve to a store, and they are the queries worth winning.

The strategic shift this implies is significant. Search rewarded breadth, ranking for as many keywords as possible, including broad ones. AI rewards the opposite. It rewards being the unambiguous answer to specific questions, and broad terms are not where recommendations are won or lost. A catalog optimized for broad category coverage is optimized for queries that produce no winner. A catalog optimized to answer specific, concrete questions is optimized for the only queries that name anyone at all.

Seventy-eight stores won all six of their queries. In a study where 60 percent won nothing, this group is worth understanding closely, because whatever separates them is the closest thing the data offers to a formula.

They have one trait above all others in common. They are not generalists. They are overwhelmingly focused, single-category brands with one clear product identity, companies known for one thing whose entire catalog reinforces what that thing is. The group includes brands defined by a single product world: sleep systems, sparkling drinks with a specific health angle, multi-tools, rucking gear, drum education, wine preservation, earthing footwear, at-home hair-growth devices. Each is a brand you could describe in a single sentence, and that is precisely the point.

The pattern, stated plainly, is this: the stores AI recommends most reliably are the ones an AI engine can describe in one sentence. When a catalog has a sharp, consistent identity, the assistant knows exactly which question it answers, and names it confidently and repeatedly. When a catalog is sprawling or vague, spanning many loosely related categories, the assistant has nothing clean to recommend it for, and it falls into the invisible majority. Coravin is the answer to wine preservation. Drumeo is the answer to learning drums. Bartesian is the answer to cocktail capsules. The assistant does not have to guess what they are for, so it names them.

Focus reads as authority to a machine the same way it does to a person. A specialist is trusted on their specialty. A store that sells one category exceptionally well, and whose product information consistently reinforces that one thing, gives the assistant both the clarity to know what to recommend it for and the confidence to do so over broader competitors. This is the trait the difficult categories most often lack, and it is available to any store willing to sharpen its identity, narrow its claims, and make its catalog say one thing clearly rather than many things vaguely. The 0.8 percent are not the biggest stores in the study. They are the clearest.

Pulling the findings together produces a clear picture of how AI recommendation differs from the search it is replacing, and the difference defines the work ahead for every store.

Google ranks pages and returns a list, and the shopper does the sorting. AI names an answer, and the sorting is already done. To be the answer rather than one of ten options, a catalog has to be legible to a machine. That means product titles that state plainly what the thing is and who it is for. Descriptions written as answers to real questions rather than as feature dumps or brand poetry. Attributes, use cases, and specifications stated explicitly and consistently, in the language shoppers actually use when they ask. And an overall identity clear enough that an assistant can describe the store in one sentence and know exactly which questions to name it for.

The 60 percent of stores that are invisible are not invisible because they are small, or unknown, or bad at what they do. Anine Bing is none of those things. They are invisible because their catalogs were written for humans and for search engines, and the assistant is left to improvise from whatever it can extract, which is usually not enough to name them with confidence. The fix is not a bigger budget or a stronger brand. It is making the catalog say, clearly and specifically, what the store is the answer to.

Every merchant reading this can start with three questions about their own store.

Most stores have never checked the answer to the first question. Until they do, they are managing a channel they cannot see, in a market that has already moved.