AI can only recommend what it understands. Most catalogs only deliver half of that.
Brands are racing to make their products understandable to AI. But understanding has two objects: the product and the person it's for. Most catalog work covers the first and stops.
I keep coming back to a line we put on a graphic recently: AI can only recommend what it understands. People nod at it. Then they go back to their product descriptions and fix the grammar.
That’s the gap. The sentence is true, but it hides a question. Understands what, exactly?
Understanding has two objects
When a shopper asks an assistant for something, there are two things the model has to understand to answer well.
The first is the product. What is it, what is it made of, how does it fit, what color is it really. This is where nearly all the effort goes right now: cleaner titles, richer attributes, structured data, pages a crawler can actually read.
The second is the person. Who is asking, what do they look like, what are they doing next Saturday, what have they bounced off before. A shopper who writes “warm undertone, combination skin, outdoor wedding in Austin, under $60” has given the model a full paragraph about herself.
Most catalogs have nothing to say back to that paragraph.
What the half-understood catalog looks like
Take a foundation. The page says “medium coverage, natural finish, 30 shades.” A model can read that perfectly. It now understands the product.
But which of those 30 shades suits a warm undertone? Does the formula hold up in heat? Does it work on combination skin or only on dry? The page doesn’t say, because the page was written to persuade a person scrolling, not to answer a question about a person.
So the model does what models do. It guesses from reviews, from forum threads, from whatever a competitor happened to state plainly. Sometimes it guesses your way. You can’t count on it.
The same thing happens in apparel. A dress description covers fabric and length. It doesn’t say which body shapes the cut flatters, or that it runs a half size small in the shoulders. Those facts exist. They live in a returns spreadsheet and in the head of your best stylist.
Why this matters more as agents take over
Today a human still reads the answer and clicks through. Soon an agent will compare, decide, and check out on the shopper’s behalf. When a person is deciding, a vague product page can be rescued by a good photo and a hunch. An agent has no hunch. It has the data you gave it.
If the data covers the product but not the person, you’re in the candidate set. You are not the answer. We call that being indexed. Being chosen takes the other half.
What to do about it
You don’t need a replatform. You need to add the person half to what you already publish.
- Pick your ten highest volume products.
- For each, write down who it’s for in plain terms: the undertones, body shapes, skin types, occasions, and climates it serves, and who it doesn’t.
- Check whether any of that appears on the page or in your structured data today.
- Mine your returns reasons and reviews for the relational facts (runs small, pulls orange on deep skin) and state them directly.
Most brands find the answers already exist inside the company. They just never made it to a place a model can read.
At Veristyle, this is the work. Computer vision reads 200+ style, fit, and color traits off the product. A model of the shopper covers the physical traits, tone, shape, style, occasion, and intent. The match between them gets written back into your catalog, on the stack you already run.
See which half you’re missing
Pick a product and ask an assistant the way a real shopper would, with her undertone, her event, her budget. See whether your brand shows up, and whether the reason it gives is true. A free AI visibility audit runs that test across your catalog. Or book a demo and we’ll go through yours together.