The Cart That Follows You Home

TL;DR

Google unveiled Universal Cart at I/O 2026 — a single shopping cart that works across Search, Gemini, YouTube, and Gmail, letting consumers add products from different merchants and check out without leaving Google's ecosystem.
Launch partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify. Rolling out in the US summer 2026, expanding to Canada, Australia, and the UK.
The Agent Payments Protocol (AP2) — announced alongside — lets AI agents make payments on users' behalf within defined limits. This is the infrastructure for autonomous shopping.
Brands that are not optimised for agentic discovery will simply not be in the consideration set. The purchase journey as retailers have understood it is being dismantled.
Google's Shopping Graph contains 60 billion product listings. People shop across Google more than a billion times a day. The scale is not theoretical.

What Happened

On Tuesday at Google Marketing Live in Mountain View, the company did something it has been building toward for years. It announced Universal Cart — an intelligent, cross-platform shopping cart that follows consumers across every Google surface — and the Agent Payments Protocol (AP2), which lets AI agents spend money on users' behalf.

The cart is not a website feature. It is infrastructure. A consumer can add makeup from Sephora via the Gemini app, cleaning products from Target through Google Search, and shoes from Nike via a YouTube video — then check out once, through Google, without ever visiting a merchant's site.

The technical backbone is the Universal Commerce Protocol (UCP), an open standard developed with Shopify earlier this year. UCP lets Google's systems communicate directly with merchant backends. Retailers remain the Merchant of Record — they legally own the transaction — but Google owns the customer relationship, the data, and the checkout experience.

AP2 is the quieter, more consequential announcement. It is a protocol designed to let software agents make secure payments within user-defined parameters. Combined with Gemini Spark — Google's new 24/7 AI assistant — it creates the plumbing for a world where consumers set an instruction and an agent executes it. The brand finds out it was selected only when the order lands.

Launch partners: Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants including Fenty and Steve Madden. US rollout begins summer 2026 on Search and Gemini, with YouTube and Gmail to follow. Expansion to Canada, Australia, and the UK is planned.

What It Actually Means

Stop thinking about this as a shopping cart. It is not a shopping cart.

It is the checkout layer for the agentic internet.

For two decades, e-commerce has followed a predictable architecture: search → browse → select → checkout. Each step happened on surfaces the brand controlled, or at least influenced. The brand built the store, designed the experience, told the story, and captured the data.

Universal Cart dismantles that architecture. It replaces it with a model where the agent — not the consumer, and certainly not the brand — makes the first round of decisions.

Here is the sequence that matters:

A consumer sets a parameter. ("I need running shoes under $150 with good arch support.")
An AI agent — Gemini Spark, running on Google's infrastructure — searches, filters, compares, and selects.
The agent adds items to Universal Cart across merchants.
The agent checks out using AP2, within the consumer's pre-set spending limits.
The brand receives the order. The brand was never in the room.

This is not speculation. It is the stated architecture. Google's VP of global ads, Dan Taylor, said the company is "moving from marketing automation to marketing intelligence." Vidhya Srinivasan, VP of ads and commerce, framed it as shopping becoming "more powerful, intelligent and fun." What neither said, but what the architecture makes clear, is that the brand's role in the purchase decision has been structurally reduced.

The Stakeholder Landscape

Google is the unambiguous winner. It already controls roughly 90% of global search. It already has the Shopping Graph — 60 billion product listings, updated in real time. It already has Google Pay, Google Wallet, and the consumer trust that OpenAI lacks. Universal Cart closes the loop. Google now owns discovery, decision, checkout, and payment. It is not a marketplace, it insists. It is a "matchmaker." This is a distinction without a difference when the matchmaker controls every surface the match happens on.

Amazon is the named target. Universal Cart is a direct challenge to Amazon's checkout dominance. But Amazon has something Google does not: the physical fulfilment infrastructure. Google's cart routes transactions to merchants; it does not fulfil them. That gap matters — but it narrows every time a consumer chooses the convenience of Google's unified checkout over clicking out to a merchant site.

OpenAI is the cautionary tale. It abandoned native checkout in ChatGPT earlier this year, reportedly after discovering that payments infrastructure is far more complex than it anticipated. Google, with decades of payments experience and an existing wallet product, did not have that problem. The Bain study showing only 24% of users would be comfortable using AI chats to make a purchase is a problem for OpenAI. It is not a problem for Google, which already holds credit card data for hundreds of millions of users.

Brands and retailers face an existential question: are you in the agent's consideration set, or are you not? Holly Enneking, VP of marketing at Markup AI, put it bluntly: "A brand is either in the agent profile, or it's not in consideration, full stop." The fields that power agentic reasoning — product fit, materials, care instructions, compatibility — are, according to Max Sinclair of Azoma AI, "the most ignored real estate in e-commerce." Brands that have neglected their product metadata are about to discover how expensive that decision was.

Consumers gain convenience and lose visibility. The Universal Cart will find better deals, flag incompatibilities, apply loyalty discounts automatically, and track price changes. It will also give Google more data about who buys what, in which colour, at what price point, and at what time of day than any company has ever held. The trade is convenience for data. Most people will take the trade.

Physical retail is not spared. The Colliers report released last week found that 85.1% of US retail sales still flow through stores, and 71% of retailers are expanding their physical footprints in 2026. But the same report found that consumer adoption of AI agents is projected to jump from 19% to 46% by the end of 2026. The store is not dying. But the decision about what to buy is increasingly happening before the consumer ever walks in.

The Hype Check

Is this being overhyped? Partly. Universal Cart is rolling out in the US this summer. It will not be in every market, on every surface, overnight. Merchant adoption of UCP is still early. AP2 is a protocol — not a product — and the timeline for agentic payments at scale is measured in years, not months.

But the direction is not hype. The infrastructure is real. The launch partners are real. The scale — a billion shopping interactions a day, 60 billion product listings — is real. And the structural logic is inescapable: when the agent controls discovery, the brand that is not discoverable by the agent does not exist.

Cross-Layer Implications

Security and privacy. AP2 creates a new attack surface. An agent authorised to spend money within defined limits is a high-value target. Google has not yet detailed the security architecture, but the protocol's design — user-defined spending limits, transaction boundaries — suggests a recognition of the risk. The question is whether the safeguards are sufficient.

Regulatory. The EU AI Act's obligations around high-risk AI systems may apply to agentic commerce infrastructure. The EU AI Office has not yet commented on UCP or AP2, but the combination of autonomous financial decision-making and consumer data aggregation will attract scrutiny. Google's insistence that it is "not a marketplace" is likely a pre-emptive regulatory posture.

Talent and operations. The Vogue piece quotes Ben Hussey of Katana Cloud: "A 97% accurate inventory figure that was acceptable when it lived inside your Shopify admin becomes a source of abandoned carts and poor Google Merchant ratings when surfaced in a universal checkout flow." This means retailers need real-time inventory synchronisation, accurate product metadata, and AI-optimised product feeds — skills that most retail organisations do not currently possess at scale.

The UK consumer caution story — EY's revised forecast of 0.3% consumer spending growth, down from 0.9%, with inflation potentially exceeding 4% — is the counterpoint. While Google builds the infrastructure for frictionless spending, consumers in key markets are pulling back. The tension between infrastructure capability and consumer capacity will define retail in 2026–2027.

What This Means for You

If you are a retailer or brand: Audit your product metadata now. Fit, materials, care instructions, compatibility, sustainability certifications — these are the fields that AI agents use to make recommendations. If they are incomplete, inaccurate, or absent, your products will not surface. Invest in real-time inventory synchronisation. The gap between your stockroom and your product feed is about to become customer-facing. Consider whether your DTC channel is strong enough to survive being bypassed by the agent. If your brand relationship depends entirely on the transaction moment, you are exposed.

If you are a marketer: The shift from SEO to AIO (AI optimisation) is no longer theoretical. AI Overviews already reduced click-through rates for top-ranking pages by 58% as of May 2026, according to Ahrefs. Universal Cart accelerates this. Your job is no longer to drive traffic to a product page. It is to ensure your product is the one the agent selects. That requires different skills, different data, and a different relationship with the platforms.

If you are a consumer: You are about to be offered extraordinary convenience. The trade is your data. Google will know what you search for, what you watch, what you email about, and what you buy — and it will connect all of those dots. You can opt out by not using Universal Cart and checking out on merchant sites. Most people will not. Be aware of what you are trading.

If you are an investor: The companies that build the infrastructure for agentic commerce — inventory synchronisation, product data management, AI optimisation, agentic payments security — will capture value disproportionate to their current size. The companies that depend on controlling the purchase moment will lose it.

Uncertainty Ledger

AP2 timeline. Google has teased AP2 integration "in the coming months" but has not committed to a launch date. The gap between protocol and product is where execution risk lives.
Merchant adoption beyond launch partners. UCP is an open standard, but adoption requires technical integration. The speed at which mid-market and long-tail merchants adopt UCP will determine whether Universal Cart becomes the default or remains a premium-tier experience.
Regulatory response. The EU, UK, and potentially US regulators have not yet engaged with agentic commerce infrastructure. The regulatory posture — particularly around data aggregation and consumer protection — could reshape the rollout.
Consumer behaviour. The Bain data (24% comfort with AI purchases) suggests a trust gap. Whether Universal Cart closes that gap or is rejected by privacy-conscious consumers is unknown.
Amazon's response. Amazon has not yet responded publicly. Its fulfilment infrastructure is a moat Google cannot easily replicate. But Amazon's own AI shopping features are less advanced than Google's. The competitive dynamic is unresolved.

Bottom Line

Google's Universal Cart is not a feature. It is the checkout layer for an internet where AI agents — not consumers — make purchasing decisions. The infrastructure is real, the launch partners are real, and the scale is unmatched. Brands that are not optimised for agentic discovery will simply not exist in the consideration set. The purchase journey as retailers have understood it for two decades is being dismantled — not by a competitor, but by a protocol. The brands that win will be the ones that understand the agent is now the customer, and the agent does not care about your brand equity. It cares about your data.

Sources

ADWEEK, "Google Challenges Amazon With New Native Checkout, Rolls Out AI Ad 'Explainers'" (May 20, 2026) — Tier 1
TechCrunch, "Google's new Universal Cart wants to follow you across the entire internet" (May 19, 2026) — Tier 1
Vogue, "Google Just Changed How We're Going to Shop" (May 20, 2026) — Tier 2
Forbes, "What Google's Universal Cart Means For Agentic Shopping" (May 20, 2026) — Tier 2
Chain Store Age, "Google launches smart digital cart; upgrades Agentic commerce standard" (May 19, 2026) — Tier 2
Chain Store Age, "Study: In-store online order fulfillment expected to rise" (Colliers report, May 14, 2026) — Tier 2
CNBC, "Radar reaches unicorn status in series B funding round" (May 19, 2026) — Tier 1
InteriorDaily / EY, "Consumer caution deepens as global tensions weigh on UK retail" (May 15, 2026) — Tier 2
Bain & Company, agentic AI commerce consumer trust study — Tier 2
Ahrefs, AI Overviews click-through rate research (May 2026) — Tier 2