Who will hold the shopping cart?

One of the biggest shifts on the internet, which now feels like a lifetime ago at this point, was when aggregators broke websites into content. Readers stopped going to NYTimes.com, and stayed on their feeds. A written article from the Times became a tweet, or a TikTok that restated the reporting with barely a passing nod to the source. Content became tiny islands, and aggregators had the map, boats, and oars to reach every one.

This is happening to ecommerce.

Even my most AI-pilled friends don’t use AI to buy things, myself included. For product research? Of course. Monitoring prices? Definitely. But the purchase was never the hard part. Attention and desire are, and will remain so. Visa recently released its own shopping benchmark, grading 22 AI models on 15 shopping tasks. Checkout wasn’t one of them, and this benchmark came from Visa.

So when Google previewed Universal Cart in May, I was skeptical. But having tested it since launch, I’ve come to understand it’s less about automating any single part of the purchase journey and more about adding new surface area to it.

Here’s what I’m seeing: Google built a cart that sits above retailers, the evidence reinforces that AI is a discovery tool more than a checkout tool, and retailers have more room to maneuver than that setup suggests. What follows is what I’ve found so far, and what I’m still watching.

What Launched

On October 15, Google launched Universal Cart in the U.S. across Search and the Gemini app. It’s most helpful to think of it as a universal price watcher: one saved list across retailers, with price and restock alerts, promo codes, and card perks. You can check out in Google Pay with participating merchants, or send the cart to the merchant’s site. Either way, the merchant stays the seller of record.

Under the hood, the price tracking runs on Google’s Shopping Graph. The Universal Commerce Protocol (UCP) is how Google talks to retailers for checkout and cart transfer. It was announced in January and co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and it gave retailers a concrete answer to a question many were asking: how do we make our products readable to AI agents?

Evidence so far suggests AI is mostly a discovery tool. In March 2026, OpenAI ended Instant Checkout in ChatGPT, which it had launched 7 months prior, saying the first version “did not offer the level of flexibility that we aspire to provide.” Only about a dozen Shopify merchants had reportedly gone live by that point. ChatGPT now sends shoppers to retailers’ sites to finish buying. This reiterates AI traffic works as a discovery channel more than a checkout experience. While this may change as capabilities and tools update, my read remains that checkout inside an AI experience will be a small portion of purchases.

What’s interesting is that making your products discoverable also makes them comparable. If ten merchants sell the same product, the buyer can see it at a glance. Unsurprisingly, this launched just in time for shoppers to load up their carts before Black Friday and Cyber Monday. In a few weeks, we’ll start to see how much this feature affects the ecommerce landscape.

The Cart Retailers Can’t See

Cart abandonment is always a key metric in ecommerce, because adding products to a cart is one of the highest-intent signals a customer can give.

I read through the Universal Cart docs and found that retailers won’t see data on what shoppers save. Google sends the retailer a single request when the shopper transfers the cart or pays, with no telemetry before then. That’s not surprising, since carts on non-owned channels provide fewer analytics than owned channels.

To be fair, a save in Google’s cart is cheaper and earlier than a cart-add on a retailer’s own site, so it is a weaker signal. But Google is pushing discovery upstream of even its own search results page, and it seems willing to accept a cost for it. Alphabet’s stock fell more than 10% after its October 28 report, mostly on profit and cash-flow misses. But Search revenue also grew 12%, short of expectations, and an executive acknowledged the lower monetization in AI Mode as one factor, alongside currency effects.

For consumers, this removes a mental model from online purchasing. Instead of “buy item X from Merchant A or Merchant B,” it’s just “buy item X when it’s cheap enough.” It’s similar to how Google Search trained us to think in keywords, and now LLM-assisted search has brought us back to natural language.

Who Gets Squeezed

Universal Cart is great news if you’re a multi-brand retailer that competes on price. It’s less good news if you’re a specialty retailer whose edge is something less tangible. It’s…moderately good if you have differentiated, esoteric products, especially if your consumers are high-information.

Two examples here:

1. I recently bought a backpack for work, and the online carry community nerds out hard on specs and details. In previous searches, I would spend a lot of time on Reddit, niche community sites, and YouTube reviews. AI tooling let me focus more on specs, features, and trade-offs, and found both the niche and mass-market products in one go.
   
2. I’m almost out of a mass market protein powder I use as a daily ingredient for my breakfast. The price is currently up 25% on Amazon, where I usually buy from, since I started writing this article. So in the spirit of testing the Universal Cart, I used Gemini to look for cheaper alternatives.
   
   Result: prices seem to be up across the board on this particular flavor but Gemini found me a better deal from a retailer I know, but never would’ve checked directly. That store was near the bottom of the first page of Google Search results, too.
   
   Conclusion: another retailer won a replenishment order from Amazon (overcoming a lot of entrenched behavior on my part) simply by having conveniently accessible lower pricing on a non-differentiated product. I expect as Universal Cart gets more adoption it’ll become the de facto first step for several customers.

Retailers Have Options

Retailers need to decide where on the AI search accessibility spectrum to land. At the most conservative end, they can narrow access to outside agents and offer a controlled experience. At the other, they try to be as broadly accessible as possible. On October 21, 2026, TechCrunch reported that Amazon has begun blocking outside AI agents from buying on its site, including Meta’s Muse.

Walmart went the other way, but used its scale to dictate its own terms. Its new Agent Gateway lets approved agents search its inventory and buy for a shopper. Each agent gets its own profile inside the customer’s Walmart account, and Walmart’s rules say agents can’t see payment information, can’t go beyond what the shopper wanted, and get blocked if they don’t comply. Walmart also helped build UCP. There are no usage numbers yet, so I’ll keep watching for emerging patterns.

Promised at Launch, But Not Yet Shipped

Loyalty matters more when a brand is one differentiated line of text next to a price. UCP’s loyalty extension, merged in May, supports tiers, points, and member pricing, and its authors say the goal is that AI checkout doesn’t erase earned benefits. But the live cart shows credit-card rewards, not retailer loyalty. Loyalty and account linking are on UCP’s road map but not yet shipped.

Suggestions follow the same pattern. In May, Google said the cart would flag incompatible parts and suggest alternatives. That example wasn’t mentioned in the October Universal Cart announcement. The plumbing is in the spec: open proposals for substitutes and recommendations. Cross-selling is hard for retailers, and cross-retailer cross-selling would be revolutionary. It takes enormous data or enormous manual effort, and Google has about the biggest pile of data there is, so I’m interested to see what comes of this.

What I’m Watching Out For

Agentic shopping is still evolving, so here are several things I’m curious about:

• What will Black Friday and Cyber Monday indicate about changing behavior? Prices are up across the economy, and shoppers are hunting for good deals. A simple cross-retailer price tracker, in the most familiar interface there is, launching right before the biggest shopping weekend of the year, has a lot of room to be adopted.
• How do retailers compete inside Universal Cart? Constant discounting isn’t a strategy. Maybe it’s loyalty, but the cart doesn’t use retailer loyalty yet. Maybe it’s selling different things on Google than on their own sites, like store brands, exclusives, and bundles that can’t be lined up like-for-like. Maybe it’s brand, since when shoppers compare at the product level the brand has to win before the search.
• Where do retailers land on the spectrum of agentic commerce? Some will narrow access to outside agents, as Amazon has. Others will try to be as broadly accessible as possible, as Walmart is, but with varying levels of agent terms. Most will fall somewhere in between, and some may sit it out entirely.
• Will more retailers gate better prices and perks behind membership? If comparison is the threat, being incomparable is the defense. REI, Costco, and some luxury and health brands already work this way, and a member price can’t be compared or tracked until the shopper signs in. The cost is conversions, since a shopper who can’t see the price may buy elsewhere.
• Will anyone pay to be in the cart, and who picks the suggestions? Google already sells merchant-funded “Sponsored deal” placements in AI Mode, and UCP-integrated brands can use them. A working paper from Frankfurt School and WU Vienna, not yet peer reviewed, found paid ChatGPT traffic converted at 0.37%, versus 1.17% for Google search ads and 2.21% for ChatGPT’s unpaid recommendations before ads launched. 82% of the advertisers in the study saw no purchases. That’s specific to ChatGPT, but it raises the question of whether ads can earn their keep in an AI shopping surface without hurting the recommendations that make it valuable.

For a decade, retailers fought to own the cart. Google just built one that sits above all of them, and other platforms are building on the standard Google led: TikTok’s new Buy Direct checkout, still in testing, requires UCP too.

Whether that’s a convenience, a toll booth, or both is what the next few months of holiday shopping will start to answer.