AI is quietly rewiring how people discover products — and retailers have a narrow window before the search box stops being the front door.
On April 28, OpenAI did something that should worry every retail marketer who has spent the last decade optimizing for Google. It turned ChatGPT into a shopping assistant — product recommendations, images, reviews, and direct links to merchants, served inside a conversation and pointedly free of ads. It’s a small feature with outsized implications for anyone who sells online. For a view from inside the infrastructure, we took the question to Maxim Kozlov, who leads the development of the e-commerce site builder at Lightspeed Commerce and spends his days thinking about the plumbing beneath online storefronts.
In late April, OpenAI added shopping to ChatGPT. Is this a real shift, or a headline?
It’s a small product update with a large implication, and I’d separate those two things. The feature itself is modest: you ask for a quiet cordless vacuum for a small apartment, it gives you a few options pulled from reviews and specs across the web, and it links you out to the merchant to buy. No ads, no sponsored slots.
But step back and look at what it changes structurally. For twenty-five years, online discovery has meant a search box and a results page — a surface every retailer learned to optimize, with SEO and paid search and structured data. What just happened is that, for a growing group of shoppers, that surface is being replaced by a conversation. And in a conversation, there is no page two. There’s no row of ten links to claw your way up. There’s a recommendation, and there’s everything that didn’t get recommended. That’s a very different world to compete in.
The obvious pushback is that this traffic is tiny. Why should a merchant care today?
Because the thing to watch isn’t the size of the base — it’s the slope of the curve. And I’d rather be honest about both.
The honest part: AI-referred traffic is still a rounding error next to paid search and email. Adobe, which tracks more than a trillion visits to U.S. retail sites, says so plainly. Anyone telling you AI owns shopping today is selling something.
The slope is the part that should keep you up at night. During the 2024 holiday season, Adobe measured generative-AI referral traffic to U.S. retail sites up about 1,300% year over year — and roughly 1,950% on Cyber Monday alone. And it didn’t snap back after the holidays. By February it was still up around 1,200% against the prior summer, doubling roughly every two months. A rounding error that doubles every two months stops being a rounding error very quickly. If you wait for it to be obvious in your dashboards, you’ve already lost the lead time.
What exactly does this threaten? What do retailers stand to lose?
The thing they lose first is the surface they control, and it erodes in three places at once.
The first is paid visibility. OpenAI launched this ad-free. If the discovery layer doesn’t sell placement, the lever brands have pulled for a decade — buy your way to the top — just isn’t there. You earn the recommendation or you’re invisible.
The second is narrative. If an assistant assembles your product from third-party reviews and specs, then the beautiful product page your team art-directed isn’t where the decision gets made anymore. The decision happens in someone else’s interface, in someone else’s voice.
The third is measurability. Merchants can see a spike in “AI referral traffic,” but they have almost no visibility into how the recommendation was formed, why a competitor surfaced instead, or how to influence it. The optimization playbook that exists for search simply doesn’t exist yet for assistants.
You build storefront infrastructure for a living. From where you sit, what’s the real bottleneck here?
The bottleneck is data readiness. And this is the part I think the industry is underrating.
The models are already good enough to recommend products well. What they’re recommending from is your data — your titles, attributes, descriptions, prices, availability, reviews. When an AI decides which vacuum to surface, it’s reading structured metadata and third-party signals. If your catalog is messy, thin, or stale, you don’t get parsed, and you don’t get picked.
So the inversion is this: for twenty years, the brand with the biggest media budget won the top of the funnel. In an AI-mediated funnel, the brand with the cleanest, richest, most machine-legible data wins. Budget doesn’t help you if the agent can’t trust or read your catalog. That’s genuinely new, and it’s good news for smaller merchants who get their data house in order.
So, concretely — what should a merchant do tomorrow morning?
Treat your catalog as if it were an API that machines will read, because increasingly it is.
Practically: complete, structured product data with real attributes, not marketing adjectives. Specifications an assistant can compare on — dimensions, materials, compatibility, the things a shopper actually filters by. Reviews and Q&A content surfaced in a parseable way. And above all, freshness: accurate, real-time pricing and inventory, because the moment agents start transacting, a stale “in stock” becomes a broken promise and a lost customer.
This is exactly the layer a platform should take off the merchant’s plate. The job of a good site builder used to be making a storefront look right to humans. Now it’s also making the underlying data legible to machines — without the merchant having to think about it. That’s a lot of where our attention is going.
Is this only about discovery, or is actual buying coming?
Buying is coming, and you don’t have to guess — watch what the platforms are building, not what they’ve shipped.
In January, OpenAI previewed Operator, an agent that can navigate sites and complete multi-step tasks for you, shopping among them. Rough, gated, but the intent was unmistakable: go from telling you what to buy to doing the buying. And at I/O in May, Google laid out the same arc far more concretely — a new AI Mode for shopping, virtual try-on, and an explicitly teased “agentic checkout” that tracks an item to your target price and completes the purchase on your behalf, adding it to your cart behind the scenes. Google also mentioned its Shopping Graph had passed 50 billion listings, refreshed by the billions every hour. That kind of fresh, structured catalog data is exactly what agent-driven buying runs on.
Two of the largest companies in tech, plus Amazon, are converging on the same endpoint from different directions. That convergence is the signal. Discovery is just the visible front edge of a move from AI-as-recommender to AI-as-buyer.
Where is this in twelve months?
Discovery collapses into checkout before the year is out. Redirect-to-merchant is a transitional design. I’d expect at least one major assistant to let you complete a purchase without leaving the chat by the holiday season.
Two: a payments-and-identity standard emerges — and a standards fight with it. In-chat checkout doesn’t work until an agent can carry your cart, your payment credentials, and your identity to a merchant securely. The payment incumbents and the AI platforms will ship protocols for exactly that, and there will be more than one, because whoever owns the rails owns the most valuable spot in the new funnel.
Three: this holiday season is the first where “AI-influenced” sales hit a double-digit share. Not double-digit purchases made by agents — the plumbing won’t be ready — but double-digit share of sales an AI tool touched somewhere in the journey. I’d watch for something around a fifth of online orders in the post-season reports.
Four: the bottleneck will be data readiness, not model quality. The embarrassing statistic of the year will be the gap between AI-driven traffic and AI-driven conversion — high intent landing on storefronts that aren’t built to receive it.
Five: “generative engine optimization” becomes a real budget line. The SEO industry pivots hard toward influencing how assistants represent products, and “how do we show up in ChatGPT and Google AI Mode” becomes a board-level question, not a curiosity.
Last one. What’s your message to a merchant who thinks this is a gimmick?
They’re making the same bet the people who ignored mobile in 2008 made. The base is small, sure — mobile commerce was small in 2008 too.
The curve is doubling every two months. The front door is moving from a search bar to a sentence typed to an assistant. Everything downstream of that sentence — which products surface, who tells the story, where the sale closes, who keeps the data — is up for grabs in a way it hasn’t been since e-commerce began. The only real question is whether you’ll be legible, and visible, when the shopper walks through that new front door. Getting your data ready is boring, unglamorous work. It’s also the whole game.
