Amazon has more product data than almost anyone on earth. Yet, in July, it barely showed up in ChatGPT's shopping results.
Searchable analyzed 318,688 ChatGPT shopping responses and found Target led the pack at 25.5%, Walmart took 20.3%, and eBay claimed 13.8%. Amazon appeared a handful of times, reportedly because it keeps much of its catalog away from OpenAI. Clearly, catalog size and ad budget had no say in who or what brands showed up.
Instead, the defining factor was accessible, structured data.
Last decade’s fight was winning search. Now, the fight is winning an AI’s recommendation.
This is where answer engine optimization (AEO) comes in. Here's why incomplete product data is the fastest way to get left out of the answers.
Picture a shopper standing in front of a shelf, holding a product with a half-filled-out label. Maybe the material isn't listed. Maybe the sizing is vague. Most people shrug, make a reasonable guess, or flag down a store employee to ask.
An AI agent doesn't do that.
When ChatGPT, Gemini, Claude, or Perplexity can't find a clear, confident answer in your product data, it doesn't guess on your behalf, and it doesn't ask a human to fill in the gap. It just moves on to the next product that has complete data.
There’s already proof of what this looks like in practice with the Amazon example above. A company with one of the largest product catalogs in the world barely appeared in ChatGPT results at all.
This isn’t a brand-recognition problem. Amazon definitely doesn’t have one of those. Instead, it’s a data-access problem. If an AI system can’t get into your product information, or can’t parse it once it’s there, the size and reputation of your business barely matter at all.
It also cuts against the assumption that AI shopping favors whoever spends the most. The same data showed eBay and Poshmark, a resale marketplace, both cracking the top 15 retailers.
This is hard to explain with ad budgets or brand prestige (because the “retailers” here are just regular, everyday consumers), but easy to explain with clean, structured listings.
When the data is legible, a secondhand handbag can compete with a national department store. Isn’t that wild?
Right now, this might feel like a small corner of ecommerce to worry about. Only about 22% of shoppers currently use AI tools like ChatGPT or Gemini to research products, according to Salsify's “2026 Consumer Research” report.
But that's exactly why this matters now: The field is still small enough that a handful of brands with well-structured data are already claiming most of the visibility, and that consolidation only gets harder to break into as more shoppers adopt these tools.
It's a tension experts Peter Crosby and Spencer Millerberg have been unpacking throughout Salsify's AEO and GEO webinar series. Namely, the brands showing up in AI recommendations today aren't winning because they're the biggest or best-known. They're winning because their data gives AI systems something solid to work with.
So what actually trips an AI agent up? It's rarely one big, dramatic failure. It's usually a handful of small, familiar data problems that ecommerce teams have lived with for years.
A few of the most common culprits:
Missing or inconsistent attributes. No listed material. A size range that's vague or missing units. No clear compatibility information for accessories or replacement parts. A human shopper might assume that one size fits most or check a sizing chart elsewhere. An AI agent has no such workaround. If the attribute isn't there, it isn't there.
Conflicting data for the same SKU across retailers. The same product might list different dimensions on the brand's own site, a slightly different weight on a retailer's page, and a third version entirely on a marketplace listing. To a shopper, this probably doesn’t even register. But to an AI system trying to generate one confident answer, it looks like unreliable information.
Marketing language standing in for real answers. Descriptions like "premium comfort you'll love all day" sound fine to a human skimming a page. They tell an AI system nothing. Agents are trying to answer specific questions (e.g., what is this made of, will it work with what I already own, how does it compare to the other option), and flowery copy simply doesn't contain the information needed to answer them.
No clean schema or structured markup. Even when the right information exists somewhere on a page, if it isn't marked up in a way machines can parse, an AI system may not be able to reliably extract it at all.
That last point touches on something a bit deeper: AI systems aren't just looking for more content. They're looking for a different kind of content: the contextual detail that explains how a product fits into someone's life, not just what it is.
"OpenAI's public data feed, for example, has fields like 'Intended Purpose / Role' and 'Event Context / Use Case,' which is not information that is typically stored and used in traditional ecommerce,” says Rob Gonzalez, Salsify’s co-founder and chief innovation officer, in an interview for TechBullion. “This information is unstructured text and is about describing a product's contextual relevance in the world … [It] gives us strong hints as to what information matters, and it's the contextual data: when is the product used, what it is used with, what is it used for, what is it used instead of, etc."
In other words, a spec sheet alone isn't enough anymore. AI systems want to know when, why, and alongside what a product gets used.
Nowhere is this more visible than in categories where shoppers might have more complicated questions. Take beauty, for example.
According to the “Beauty AI Visibility Index 2026” from 5W AI Communications, five brands lead the pack when AI tools are asked for skincare and beauty recommendations:
The Ordinary (7.0%)
CeraVe (6.0%)
Sephora (5.5%)
La Roche-Posay (5.0%)
Charlotte Tilbury (4.5%)
None of that comes down to which product actually works best. It comes down to what an AI system can find. Every time an AI tool recommends a brand, it's really just reflecting how much solid information it has about that brand, including product specs, reviews, expert write-ups, retailer listings, and so on.
Beauty shows just how high the stakes are, because these shoppers rarely ask simple questions. They want to know what works for their skin type, what order to apply products in, whether two ingredients are safe to mix, or how one product stacks up against another.
If an AI can’t find that info on your product pages, it’ll recommend something else.
This is the big difference between AI search and traditional SEO. A weak SEO page might land you on page two, which is annoying, but recoverable. Incomplete product data doesn't get you page two of an AI answer because there is no page two.
The other thing to note is that none of this is paid.
Unlike a Google search results page, where the top slots are often ads, ChatGPT's shopping carousel isn't sponsored. Nobody is buying their way into it. So when Target, Walmart, and eBay show up again and again while Amazon barely does, it’s got nothing to do with the budget behind it.
Research carried out by DetailPage showed that ChatGPT tends to use a layered approach to recommending products.
“The items that have the best SEO base, then it gets customized to the individual, based on who they are and what they’re looking for,” says Spencer Millerbeg, CEO and founding partner at DetailPage, in a webinar with Salsify.
These two results from ChatGPT were returned at the same time, from the same IP address, but by two different people:
Compared to:
SEO is still important, but there’s more to it now. Recommendations are based on a variety of factors, with a heavy emphasis on data structure and context.
The data aspect here isn’t just about visibility either. Yes, AI can mention your brand, but if shoppers don’t trust what it says, they’re not going to part with their cash.
Salsify's own consumer research backs this up: Shoppers are most likely to trust (and act on) an AI product recommendation when it comes with detailed descriptions and specs. In other words, good data doesn't just get a product into the conversation. It's also what convinces the shopper to actually buy once it's there.
Getting AI-ready comes down to three things: knowing what your data actually says today, fixing it so machines can read it, and keeping an eye on how AI is using it going forward.
Start by looking at your product content the way an AI agent would.
That means mapping your content to the questions shoppers ask, rather than the keywords they'd type into Google. "Best running shoes for flat feet" and "will this work with my existing setup" are the kinds of prompts AI agents field constantly.
It also means checking for contradictions. Does the same product list a different weight on your website than it does on a retailer's page? Does one channel say "machine washable" while another doesn’t say anything about care instructions?
Once you know where the gaps are, you can start closing them.
That starts with the basics: standardized naming conventions, complete attributes (size, material, compatibility, and everything else that used to feel optional), and comparisons that are clearly labeled rather than buried beneath paragraphs of marketing copy.
But it also means going a step further than a typical spec sheet. As Gonzalez put it, AI systems are increasingly looking for contextual fields most product pages weren’t built to capture. This includes data like intended use, occasion, or what a product is meant to be used with or instead of, aka, the kind of detail that tells an AI agent how a product fits into someone's life.
Finally, none of this matters if your data isn't structured in a way machines can parse. Clean schema and structured markup let an AI system easily scan and extract your product information.
Readiness isn't a one-time project. Once your data is in shape, the work then becomes keeping tabs on how AI systems are using it.
That means tracking how often your products get cited in AI answers, and just as importantly, how accurately those answers describe your products, pricing, and policies. AI systems can misrepresent even the best data, and the only way to catch that is to check regularly.
Include content refreshing as part of this process. Millerberg recommends doing content updates every two to three months to keep up with algorithm changes, tackle keyword seasonality, and make sure content fits in with freshness preferences.
AI agents aren't excluding products to be difficult. They're optimizing for confidence, and confidence only comes from data that's complete, structured, and easy to access.
Target and Walmart outranking Amazon, and indie beauty brands outranking legacy giants, both point to the same truth: size and ad spend don't buy you a spot in the results anymore.