All articles

Your CRM knows what she bought. Her AI assistant knows who she is.

Brands have spent a decade collecting first-party data, and almost all of it describes transactions. The shopper's AI assistant holds something different: her own description of herself. Here is what that changes about where the model of the customer lives, and which part of it a brand can still own.

The best gift guide might not be on your website. Customers can now tell AI who they're buying for, what they like, and how much they want to spend. Then ask it to choose.

Pull up the record for one of your best customers. You will see orders, average order value, the last email she opened, a loyalty tier, maybe a quiz result from two years ago.

Now think about what that same woman has told ChatGPT. That she is 5’3" and most trousers need hemming. That her skin is olive and foundations go orange on her by lunch. That she runs warm, hates stiff fabric, and is dressing for a job where she is on her feet all day. That her sister is impossible to buy for.

There are two records of this customer. Yours is the thinner one.

What first-party data actually describes

I say this as someone who likes first-party data. It is real and you earned it. But read the fields. They describe what happened in your store: what she bought, when, for how much, and what she sent back.

They rarely say why. A return coded “did not fit” doesn’t tell you it was tight across the shoulders. A repeat purchase of one shade doesn’t tell you she is still looking for a better match and settled. And none of it covers the nine other brands in her closet.

The assistant has the why, because she typed it. In her own words, with no dropdown limiting what she could say. And it carries over. ChatGPT keeps details from earlier conversations unless she turns that off, so the next question starts with everything she already said.

The person half used to be three words

We keep coming back to one idea at Veristyle: every shopping question has a product half and a person half.

For twenty years the person half was tiny. A search bar got “black ankle boots” and nothing about the woman. So the product half did all the work, and product data was built to match. Category, color, material, price.

That has flipped. The person half now arrives long, specific, and remembered. She tells the assistant who she is, or who she is buying for, what they like, and how much she wants to spend. Then she asks it to choose.

Nothing about your catalog changed, and it still got harder to be the answer. A catalog that could handle “black ankle boots” has very little to say to “ankle boots for a narrow heel, a 20 minute walk to work, and a wardrobe that is mostly navy.”

This is the gap between indexed and chosen. Indexed means the model can read your boots. Chosen means it can explain why they suit her. The explanation is built almost entirely out of things she said about herself.

You will not own her profile. You can own the match.

Here is the uncomfortable part. The individual profile sits with the assistant, and the assistant is not going to hand it to you. No integration gets you her chat history, and you shouldn’t want one.

What a brand can own is the other side of the match: a structured understanding of the things shoppers say about themselves, and which real traits of your products answer each one.

“Goes orange on me” is a statement about undertone and oxidation. Which of your shades carry the pigment that avoids it? That is not in the shade name.

“I run warm” is about fabric weight, lining, and breathability. “Hates stiff” is about drape and hand. “Petite” is about rise and inseam, and also about where a jacket’s waist seam falls.

That mapping is what we mean by a model of the customer. It takes what a person says about herself, turns it into structure, and matches it against what the product actually is. We get the product side from computer vision on the items themselves, more than 200 style, fit, and color traits that a photo holds and a feed leaves out. The customer side comes from modeling the shopper: physical traits, skin tone and undertone, body shape, style, occasion, intent, where she lives.

Catalog cleanup is a project that finishes. This doesn’t, because every matched question teaches you another way people describe themselves. That is why it compounds, and why I think it is the moat.

A test you can run this week

  1. Collect ten things your customers say about themselves. Not about the product. Pull them from reviews, return notes, and support chats. “I’m between sizes.” “I have a short torso.” “My skin is sensitive but oily.”
  2. For each one, find the field in your catalog that answers it. A field, not a sentence in the description. Count how many of the ten you can cover.
  3. Ask an assistant the full question. Put the person in it, leave your brand out, and read the reasons it gives for whatever it picks.

Most teams cover fewer than they expect on the first pass. The ones you can’t cover are the roadmap.

See how your catalog answers the person half

A free AI visibility audit shows how assistants currently describe your products and what they say when the question is about the shopper. If you want to go through your own catalog with us, book a demo.

Related reading

See Veristyle on your catalog

Book a demoGet it on Shopify