"Which one is better?" is not a product question
The first question a shopper asks an AI assistant gets a shortlist. The second one, which of these is better, is where the pick gets made. Nothing in a product feed can answer it, because better is a word about the person. Here is what that means for your catalog.

Most of the attention in AI visibility goes to the first question. “Best black ankle boots under $300.” Did we show up? Were we on the list?
Fair enough. But watch someone actually shop with an assistant and the first answer is rarely where it ends. She gets four options, reads them, and then types the thing that matters: “Which one is better?”
That second question is where the pick gets made. And it has an odd property. Nothing in your catalog can answer it.
“Better” has no value in a product feed
Take two ankle boots. Both black leather, both block heel, both around $280. Line them up the way a comparison table would.
Boot A has a 2.75 inch heel, a leather sole, an almond toe, and a close fit through the ankle. Boot B has a 2.25 inch heel, a rubber sole, a round toe, and more room in the shaft.
Which is better? You can’t say. I can’t either, and I look at product data all day.
Now add the shopper. She walks 20 minutes to work, has a high instep, and wears mostly cropped pants. B wins, and you can explain why in one sentence.
Same in beauty. Two medium coverage foundations, $48 and $52. One sets matte, one stays dewy. Better for whom? Oily skin in Houston in July and dry skin in Denver in January get opposite answers.
“Better” is a word about a person. Product data supplies the two things being compared. The thing they are compared against is her.
What the assistant does with the follow-up
By the second question, the assistant is holding more about the shopper than about any single product. It has what she said in her first message, and often what she told it weeks ago. So it does the sensible thing. It takes each option and checks it against her, one fact at a time.
Where a listing gives it something to check, it writes a reason. “The lower heel and rubber sole will be easier on a 20 minute walk.” Where a listing gives it nothing, it has two moves. It can piece an answer together from reviews and whatever else it finds about you. Or it can fall back on the things it can always compare, which are price and star rating.
That fallback deserves more worry than it gets. When neither product says who it suits, the comparison collapses into price. A brand that spent years on fit or formulation ends up in a tiebreak over a few dollars.
When one listing says who it suits and the other doesn’t, the one with reasons tends to win. It may not be the better boot. It is the one the assistant can defend.
Indexed gets you into the comparison. It doesn’t win it.
This is the indexed vs. chosen gap at its most literal. Indexed means you were one of the four. Chosen happens one message later.
Most GEO reporting stops at the first message. Mention rate, share of answers, how often the brand name appears. Those are worth knowing. They don’t tell you what happens when your product is put head to head with a named alternative for a specific woman, and that is the moment she decides.
It is also the step that disappears from view first. When an agent shops on her behalf, she never types the follow-up. The agent runs the comparison on its own and comes back with one product. You don’t get to see the round you lost.
What goes in the catalog
A comparison needs two things from you: what the product actually is, and who that makes it right for.
At Veristyle we get the first from computer vision on the product itself, more than 200 style, fit, and color traits that a photo holds and a feed leaves out. The second comes from a model of the customer: physical traits, skin tone and undertone, body shape, style, occasion, intent, where she lives. The work is matching the two and writing the result back as fields a model can read.
For the boot, that is heel height tied to how long she can stand in it, shaft width, toe shape against foot width. For the foundation, undertone per shade, finish, and the skin types and climates the formula holds up in.
One thing I’d push back on: this is not about claiming to be better for everyone. A product that says plainly who it is for will lose some comparisons. It should. Those are the orders that come back as returns.
Try the second question on your own products
- Pick a best seller and its closest rival. A competitor’s product, or the nearest thing in your own line.
- Ask an assistant to compare them with no person in the question. Note what it compares on. Usually price, materials, and reviews.
- Ask again with a person in it. Height, skin, climate, where she is wearing it. Read the reasons it gives for each product, and notice which ones are specific and which are filler.
If the reasons for the rival are sharper than the reasons for yours, you have found where the pick is going.
See how you do in a head to head
A free AI visibility audit shows how assistants describe your products today and what they say when the question is about the shopper. To run the comparison on your own catalog with us, book a demo.