Not just Search, Google's AI shift reshapes online shopping experience too

At I/O 2026, Google outlined how its new open standards and a Universal Cart will enable AI agent-led shopping, where AI systems manage discovery, pricing and checkout across platforms

Google’s push into “agentic” AI at I/O 2026 wasn’t limited to search or assistants. One of the more structural announcements came from its Shopping division, where the company outlined how it plans to rebuild the online shopping experience using a set of open standards and AI-led systems.

 

On stage, Google described this as the foundation of “agentic commerce,” a model where AI systems can track, compare, and even complete purchases on your behalf across different websites and retailers.

 

Taken together, this points to Google attempting to standardise the online shopping experience across platforms.

The starting point

During the keynote, Google said that more than a billion shopping-related interactions happen across Google every day, supported by its Shopping Graph — a dataset that now includes over 60 billion product listings that are continuously updated.

 

 

This matters because agentic systems depend heavily on structured, real-time data. Without that layer, automation across discovery, pricing and checkout becomes unreliable. 

The three pillars

Google said that the AI agent-led automated shopping experience in Search is built around three components: two open standards — the Universal Commerce Protocol (UCP) and Agentic Payments Protocol (AP2) — and a new Universal Cart that sits on top of them.

 

Universal Commerce Protocol (UCP)

 

The first building block is the Universal Commerce Protocol. Google described it as doing for commerce what HTTP did for the web — a common language that allows different systems to work together. In practical terms, this is aimed at solving a long-standing problem in e-commerce: fragmentation.

 

Today, the shopping journey is split across platforms — discovery on search or social apps, transactions on retailer websites, payments through separate systems, and tracking through logistics providers. Each step operates in isolation.

UCP is meant to connect these layers. It essentially allows AI agents to move across the entire shopping journey, from product discovery to checkout and even shipment tracking, without breaking context.

 

Google is also expanding UCP beyond retail into areas like hotel bookings and food delivery, which indicates that the company sees this as a horizontal layer for commerce, not just shopping.

 

Google said that companies such as Amazon, Meta, Microsoft, Salesforce, and even Indian players like Flipkart are on board for adopting the standard.

 

Agentic Payments Protocol (AP2)

 

If UCP connects systems, AP2 is meant to handle one of the most sensitive parts of the transaction: payments.

 

The Agentic Payments Protocol (AP2) is designed to allow AI systems to complete purchases on behalf of users, but within defined constraints. During the keynote, the company explicitly addressed the concern that an AI system could make unintended purchases. AP2 introduces guardrails where users can define:

• What products or brands can be purchased
• How much can be spent
• Under what conditions a purchase should be executed

Once these conditions are met, the system can complete the transaction automatically.

 

At the same time, AP2 creates a verifiable link between the user, the merchant and the payment processor. This includes a persistent transaction record, which ensures that both the buyer and seller are working off the same data in case of returns or disputes.

 

Universal Cart

 

The most visible part of this system is the Universal Cart. Unlike a traditional shopping cart tied to a single website, this one works across services — including Search, Gemini, YouTube and Gmail.

 

You can add products to it from different contexts, and instead of remaining static, the cart continues to operate in the background. For example, if you add products to your cart, it will dynamically track price changes and notify any drops, alert you when products are back in stock, and more.

Google is also integrating Gemini’s reasoning layer for more contextual suggestions. For example, if you are building a PC and add incompatible components into your cart, the system can flag the issue and suggest corrections before you make a purchase.

 

The cart is also integrated with Google Wallet, allowing it to surface payment-related benefits such as offers or rewards across different cards.

 

And when it comes to checkout, the cart offers two paths:

• Completing the purchase directly within Google using Google Pay
• Transferring the cart to a retailer’s site

In both cases, the merchant remains the seller of record, which is an important detail in maintaining the existing retail ecosystem.

Why Google is pushing open standards

For agentic commerce to work at scale, it cannot be confined to a single platform. It requires:

• Retailers to expose product and pricing data in compatible formats
• Payment systems to integrate with AI-led transactions
• Platforms to allow cross-service interoperability

Closed systems would limit this to isolated ecosystems, similar to how current marketplaces operate. However, by pushing open standards, Google is attempting to create a shared layer that sits above individual platforms — similar to how the web itself operates.

 

At the same time, this also allows Google to anchor itself at the centre of the shopping flow, even when the transaction happens elsewhere.

How UCP and AP2 differ from MCP

Google’s push with UCP and AP2 also comes at a time when the broader AI ecosystem is experimenting with open protocols, most notably Anthropic’s Model Context Protocol (MCP). At a high level, both MCP and Google’s UCP/AP2 aim to standardise how AI systems interact with external services. But they operate at very different layers.

 

MCP is designed as a model-centric general-purpose interface that allows AI models to connect with tools, APIs and data sources. It focuses on enabling agents to access context — such as files, databases, or applications.

 

In contrast, UCP is far more domain-specific. It is built specifically for commerce, and standardises how different systems in the shopping ecosystem — retailers, platforms, payment providers and logistics services — exchange data across the entire transaction flow.

 

The same distinction applies to AP2.

 

While MCP can enable an AI system to trigger actions, it does not define how sensitive operations like payments should be executed. AP2, on the other hand, is purpose-built for transactions. It introduces guardrails, user-defined constraints and verifiable records to ensure that purchases made by AI systems remain controlled and auditable.

Limitations and open questions

Despite the ambition, there are several limitations that still need to be addressed before Google standardises the online shopping experience:

• Adoption: For UCP and AP2 to work effectively, a large number of retailers, platforms and payment providers need to adopt them. Early partners are significant, but broader ecosystem alignment will take time.
• Control: While AP2 introduces guardrails, the idea of AI systems making purchases — even conditionally — raises questions around user trust, unintended actions, and dispute resolution.
• Platform dynamics: Retailers may be cautious about giving too much control to an intermediary layer that can influence pricing visibility, recommendations and checkout flows.

A shift from “searching” to “delegating”

Google’s agentic shopping experience is no longer just a smarter shopping interface, but a system where parts of the buying process can be delegated. Instead of actively checking prices or comparing products, users can offload parts of that work to an AI system that continues operating in the background.

 

With this, the role of the user shifts from actively executing each step to setting preferences and constraints. Whether that shift happens at scale will depend less on the technology, and more on whether users, retailers and payment systems are willing to trust and adopt this new layer.