All articles

Same product, three sellers. The agent picks one. Is it you?

Most beauty and fashion brands sell through their own store and at least one retailer. When an AI agent answers a shopper, it does not list all of them. It picks one. Here is what decides which listing wins, and why it is often not the brand's own.

Imagine not being part of this answer. A shopper asks ChatGPT which makeup products are actually worth buying, and the assistant starts building a shortlist.

Here is a situation we see constantly in the Visibility Lab, and almost nobody plans for it.

A shopper asks ChatGPT for jeans that work on a petite frame, budget under $120. The answer names a product. Good news: it is your product. Then she taps through, and the link goes to a department store. Or to Amazon. Not to you.

Your product got chosen. Your store did not.

The agent only has room for one link

Humans compare across tabs. An agent compares in one pass and hands back a shortlist of three or four items, each with a single place to buy. It is not going to list your own site, your two wholesale partners, and a marketplace seller side by side and let the shopper sort it out. It resolves that for her.

So for any product sold in more than one place, there is a second contest hiding inside the first. First the agent decides whether your product fits the person. Then it decides which listing of that product it trusts enough to send her to.

Most GEO work stops at the first contest. The second one is where a lot of brand revenue quietly leaks to retailers.

What decides the second contest

We ran this exact query for a large US retailer recently. Petite jeans, a budget, a brand the retailer carries. Two competitors carried the same product. The assistant surfaced both competitors first and the retailer not at all.

The product was identical in all three places. The difference was the data around it. One competitor had size and colorway variants structured in a form a model could parse: petite lengths as their own values, inseams stated, every color a real variant rather than a line in the description. The retailer had the same facts, but they sat in copy and images where the model could not read them as data.

That is the whole story of the second contest. An agent picks the listing that gives it the most to reason with about the person asking. Petite is not a product fact. It is a statement about who the jeans are for. The listing that said so in a structured way was the one the agent could defend.

Why your own store often loses this

It feels backwards. The brand knows the product best. The brand wrote the copy, shot the images, and has the fit notes from the design team.

But look at where that knowledge lives. On a brand site, who-it’s-for information tends to sit in a lookbook, a fit guide, a size chart image, a quiz, or a paragraph of brand voice. It is written for a human who will spend four minutes on the page.

A retailer, by contrast, forces every product through a feed template. The template is crude, but it is structured. If the retailer’s template has a field for inseam or shade undertone and your site does not, the retailer’s listing is more answerable than yours, for your own product.

The sunk cost of a beautiful product page shows up here in a very concrete way. The persuasion surface you built for humans is invisible to the thing that now decides where the click goes.

This is a person-half problem

Every shopping question has a product half and a person half. The product half is the category, the price, the material. All three listings tied on that. The person half is petite, cool undertone, runs warm, long torso, needs it for a specific occasion. That is where the agent found a difference, and that is where the pick was made.

In fashion and beauty the person half is most of the question, because fit, undertone, and occasion are relationships between a product and a body. No spec sheet carries them. This is why we build a model of the customer alongside the computer vision read of the product, more than 200 style, fit, and color traits, and match the two. It is also why the brands that do this show up as the listing the agent trusts, rather than one of several it could have chosen.

The numbers move when you fix this. In a recent Lab run on 20 products from a luxury accessories brand, the share of relevant ChatGPT answers that recommended the brand went from 18 percent to 54 percent after we added structured occasion, material and care, climate, and fit data. Same products. Same prices. Different answerability.

Three checks worth an afternoon

If you sell through more than one channel, do these before the next planning cycle.

  1. Run the same question against each listing. Ask ChatGPT or Perplexity for your product with a person in the query: a height, a skin tone, a body shape, a use. See which seller the answer links to. If it is not you, you have found the leak.
  2. Compare the structured data, not the copy. Pull your product feed and the retailer’s listing side by side. Count the fields that describe the person. The listing with more of them usually wins.
  3. Put the person half where a model can read it. Fit behavior, undertone, occasion, climate, who it flatters. As fields, not adjectives. Consistently across the catalog, not just on bestsellers.

The platforms will keep making it easier for agents to buy. They will not decide which listing the agent buys from. That part is still yours to win, and right now a lot of brands are losing it on products they make.

See which listing the agent picks for your products

A free AI visibility audit shows how assistants currently describe your products and where they send the shopper when they recommend you. If you would rather walk through your catalog together, book a demo.

Related reading

See Veristyle on your catalog

Book a demoGet it on Shopify