A customer opens ChatGPT and asks for a product that fits a real constraint: skin type, flat feet, a part that has to match something they already own. The model does not shrug and guess. It does not wave down a shop assistant. If your attributes are missing, conflicting or buried in marketing copy, it simply moves on to a listing that looks complete enough to trust.
That is the uncomfortable lesson in Salsifys AEO 101 piece. Searchable looked at hundreds of thousands of ChatGPT shopping responses and found Target, Walmart and eBay showing up far more often. The most interesting conclusion? Amazon is absent from most LLM results. It barely appeared, reportedly because much of its catalog stays closed to OpenAI. Catalog scale and ad budget were not the deciding factors. Accessible, structured product data was.
Last decade we fought for page-one search. This decade we are fighting to be inside the recommendation itself.
Why agents are pickier than shoppers
Humans approximate; agents optimize for confidence. A shopper can live with a half-filled label, open another tab, or ask someone on the floor. An agent will not fill your gaps for you. Missing material, vague sizing without units, no compatibility for accessories: for a shopper that is friction. For an agent it is an exit.
Salsify’s consumer research still puts AI product research at about 22 percent of shoppers. That is exactly why the window matters now. The field is small enough that brands with readable data are already claiming a lot of the visibility, and that position gets harder to break into as more people adopt the tools.
It also cuts against the prestige story. eBay and even Poshmark cracked the top retailers in that ChatGPT shopping analysis. When the data is legible, a secondhand listing can compete with a department store. Reputation does not parse attributes. Access and structure do.
What actually gets you excluded
It is rarely one dramatic failure. It is the familiar mess product teams have lived with for years, only now it has a harder consequence.
Missing or inconsistent attributes. Conflicting dimensions or weights for the same SKU across your site, a retailer and a marketplace. Marketing language that sounds fine to a skimmer (“premium comfort you will love) and tells an agent nothing about materials, fit or alternatives. And even when the facts exist on the page, no clean schema or structured markup so a machine can extract them reliably.
Rob Gonzalez, Salsify’s co-founder, puts it plainly: a list of specs is not enough. Agents also need the practical context most catalogs barely store, such as what the product is for, when people use it, and what it pairs with or replaces. Without that, the system can describe the object and still fail the customer’s real question.
Beauty makes that concrete. Someone asks what works for oily skin, in which order to apply it, or whether two ingredients clash. If your pages do not answer those questions in clear product data, the recommendation goes to a brand that does. The Beauty AI Visibility Index is less a ranking of “best formulas” than a ranking of who made that information easy to find.
What we recommend
We should not treat AEO as a new content sprint that pours more adjectives into PDPs. We should look at our catalog the way an agent would.
We map content to the questions customers actually ask, not only to the keywords they used to type into Google. We hunt contradictions across brand site, retailer and marketplace for the same SKU. We complete the attributes that used to feel optional: material, size with units, compatibility, care, and the contextual fields that explain role and use. We put structure and schema around it so machines do not have to scrape prose. Then we monitor how often we get cited, and how accurately, because a confident wrong answer is almost as bad as silence. Refreshing every couple of months is not vanity; algorithms and seasons move.
And based on Salsify’s research, we conclude that detailed specs are also what makes shoppers (and not just agents) trust and buy once they are in the conversation. Good data is not only how you enter the answer. It is often what closes the sale once you are in it.
Go back to the opening scene. A customer asked a precise question and your incomplete record never made the shortlist. That is the cost of living with “good enough” attributes and pretty copy. Size and spend do not buy a seat anymore. Completeness and access do.
Salsify’s full AEO 101 walkthrough is here: https://www.salsify.com/blog/aeo-101-why-ai-agents-hate-incomplete-product-content
