Agents can now pay for your product. Who tells them to pick it?
Shopify, OpenAI, and Google have all shipped ways for AI agents to complete a purchase. None of them decide which product the shopper should get. That step is still open, and it is where beauty and fashion brands win or lose.
Over the past year, the big platforms have worked hard on one problem: letting an AI agent buy something. Shopify, OpenAI, and Google each have a protocol or a program for it. The shopper says yes, the agent checks out, the order lands in your admin.
That is real progress. It is also the easy part.
Payment is a solved shape of problem
Checkout is deterministic. There is a cart, a price, an address, a payment method. Either the transaction goes through or it doesn’t. Engineers know how to build that, and the big platforms have the leverage to get merchants to adopt it.
Selection is not deterministic. Before any cart exists, someone has to decide which of 400 foundations, or 900 dresses, belongs in it for this particular person. No checkout protocol touches that decision. It sits upstream, and it is where the sale is actually won.
If you only read the announcements, you could think agentic commerce is mostly a plumbing story. Plumbing matters. But plumbing is the same for every brand that adopts it, so it cannot be what separates you from the brand next to you.
What the agent is really doing before it buys
Take a request like this: “I’m pale with cool undertones and my skin gets oily by 3pm. I need a foundation for a long day at a trade show. Nothing that looks cakey under bright lights.”
Before checkout, the agent has to do a few things in order:
- Work out what the shopper means by cool undertone, oily by afternoon, and cakey.
- Find products whose actual properties line up with those, not just products labeled “oil control.”
- Rank them for this person, and be able to say why.
- Only then hand one to the checkout flow.
The checkout protocol starts at step four. Steps one through three are the matching step, and they run on whatever the agent can learn about your product and about the shopper.
Where the matching step breaks
Most catalogs were written for people who browse. A shade named “Honey Beige” tells a human with a swatch in hand something. It tells an agent almost nothing about undertone. A dress described as “effortless and flowy” says nothing about how it sits on a petite frame or a long torso.
So the agent has two weak inputs. A product described in marketing language, and a shopper described in her own words. Matching between those is guesswork unless something in the middle translates both into the same terms.
That translation layer is what we build at Veristyle. We read the product itself, using computer vision to pull out 200 plus traits across style, fit, and color. We also model the person: skin tone and undertone, body shape, style, occasion, intent, geography. Then we match one to the other. The point is not to help your product get read. It is to help it get picked.
Indexed gets you in the room
If your feed is clean and your structured data is right, you will be in the candidate set. Good. So will everyone else who did the same work.
From there, the agent needs a reason. In fashion and beauty, the reasons are almost always about the person: this undertone, this fit, this event, this climate. A brand that can supply that reasoning in a form the agent can use gets chosen. A brand that cannot gets listed and skipped.
That is the difference we keep coming back to. Indexed means a model can read you. Chosen means a model can match you to a human being and defend the choice.
What to do about it this quarter
You do not need to rebuild your stack. Start by looking at your own products the way an agent would.
- Pick ten bestsellers. For each one, write down who it is for in a single sentence. If you cannot, the agent cannot either.
- Ask ChatGPT or Perplexity a question that describes a person in detail and barely mentions your product. See who comes back and read the reasons given.
- Check which attributes in your feed are about the person (undertone, body shape, occasion) versus the product (material, size, price). Most feeds are almost entirely the second kind.
The gaps you find are your roadmap. Checkout will be handled for you by the platforms. Selection will not.
Find out where you stand
A free AI visibility audit shows how AI assistants currently describe and recommend your products, and where the person half of the question goes unanswered. If you would rather talk it through, book a demo and we will walk through your catalog together.