Your product feed has a field for everything except who it's for
Titles, tags, variants, materials, GTINs. Every feed is thorough about the product and silent about the person. That missing field is where AI agents decide, and where most fashion and beauty brands lose the pick.
Open the product export from your Shopify store and scroll across the columns. Title, handle, description, vendor, type, tags, price, compare-at price, SKU, barcode, weight, image URL, alt text. If you have done some GEO work, add material, color, size, and a few metafields.
Now look for the column that says who this product is for.
It isn’t there. Not on your store, and not on almost anyone else’s.
The feed was built for a different reader
Product feeds grew up serving shopping ads and marketplaces. Those systems needed to answer a narrow question: is this item what the shopper typed? A search for “black ankle boot size 8” is a filter problem. Match the color, the category, the size, show the price.
The trouble is that the shopper has stopped typing filters. She is typing the sentence she would say to a friend, and an AI agent is reading it.
What the agent is actually asked
Here is a question that looks like a lot of the ones we see in the Visibility Lab:
“I’m pear-shaped, 5’4”, and I run warm in summer. I need a dress for an outdoor wedding that doesn’t need a strapless bra. Nothing clingy."
Count the product words. Dress. Maybe wedding. Everything else describes her: her body, her height, how she feels in heat, what she will not tolerate. The product half of the sentence is a category. The person half is the whole decision.
Now the agent turns to your catalog to find a match. It reads your feed. The feed says the dress is “Marlow Midi, floral, rayon blend, available in 6 sizes.”
That is all true, and it answers none of her questions.
Three things that live in no column
Fashion and beauty run on attributes that only exist in relation to a person. Three come up constantly.
Fit is a relationship. A size chart tells you the garment’s measurements. It does not tell you the dress will skim a fuller hip and nip at the waist, which is the thing a pear-shaped shopper is asking about. That is cut, drape, and stretch interacting with a body.
Undertone is a relationship. Your foundation shade called “Sand 3N” has a pigment in it. Whether that pigment reads warm or neutral against a specific person’s skin is not in the shade name. Shade names are marketing. Pigment is data, and almost nobody has captured it.
Occasion is a relationship. “Wedding guest” is not a tag on a dress. It is a combination of venue, season, dress code, and how much the shopper wants to be noticed. The same dress is right for one wedding and wrong for another.
Each of these is a match between a product trait and a person trait. A feed can only hold one side.
What happens when the field is missing
The agent still has to answer. So it improvises.
It leans on whatever language it can find: your description copy, reviews, a fashion editor’s roundup, a Reddit thread. If a thread says a competitor’s dress is great for pear shapes, the agent takes that as the evidence it needed. The brand that wrote the explicit claim on the page wins, or the brand somebody else happened to write about.
This is the difference between being indexed and being chosen. Indexed means the model can read your product. Chosen means the model can match your product to the person asking. Most GEO work moves a brand from invisible to indexed. The last step needs the missing field.
What the field looks like
We are not suggesting you stuff “great for pear shapes” into every description. That is the copywriting fix again, and it won’t scale across a catalog of thousands.
The better version has two parts, and both are structured.
On the product side, you extract traits from the product itself. The photos already contain a lot: neckline, waist definition, fabric drape, length, how structured the shoulder is, pigment depth, finish. Computer vision can read those off the image and write them down in a form a model can use. We pull more than 200 style, fit, and color traits this way, which is more than any merchandiser would hand-enter.
On the person side, you need a model of the shopper that uses the same vocabulary: body shape, skin tone and undertone, style, occasion, intent, geography. The match is then a comparison between two structured descriptions, not a guess from prose.
When both sides speak the same language, “who is this for” stops being an adjective in your marketing copy and becomes something an agent can compute.
Why this matters more as agents get more control
Soon the agent will compare options and complete the purchase while the shopper does something else. The checkout protocols from Shopify, OpenAI, and Google handle the transaction well. They don’t decide which dress gets bought.
That selection happens upstream, on whatever the catalog supplies. A clear answer to “who is this for” gives the agent a reason to pick you. Without one, it is a coin flip weighted by review count.
A five minute test
Pick one product you are proud of. Write down the three most specific shopper sentences you think it is perfect for. Then paste each one into ChatGPT, Perplexity, and Google, and see which products come back and what reasons are given.
If your product shows up, read the reasons. If it doesn’t, read the reasons for the one that did. Either way, you are reading a spec for the field your feed doesn’t have.
If you would like that run across your whole catalog, our free AI visibility audit does exactly this. Or book a demo and we will show you what the layer sees in your products.