Why Agentic Commerce Is Won in the Product Feed, Not at Checkout

Agentic commerce is the part of online shopping where AI agents search, compare and sometimes buy on behalf of a shopper. Two open protocols frame it: the Agentic Commerce Protocol, maintained by OpenAI and Stripe, and the Universal Commerce Protocol, launched by Google in January 2026. For brands, the lesson of 2026 is clear: checkout models are still settling, while AI product discovery is already live.

• AI assistants now shortlist products inside ChatGPT, Google AI Mode and Gemini before a shopper visits any site.
• ACP and UCP are open standards that let agents read catalogs, build carts and pass payment credentials securely.
• OpenAI moved away from Instant Checkout in March 2026 and now lets merchants use their own checkout.
• Google’s agentic checkout started with US retailers, with Canada and Australia expected in early 2027.
• Product data quality decides which items agents recommend, long before payment is involved.

Agentic Commerce Moves Product Discovery Inside the AI Assistant

Agentic commerce changes where the shortlist is built. Instead of typing a query, opening ten tabs and comparing product pages, the shopper describes a need to an assistant, which searches catalogs, compares options and returns a handful of products with prices, reviews and links.

Three stages are involved, and they are not equally mature:

• Discovery. The assistant finds and compares products. This is live at scale: since March 24, 2026, ChatGPT shows visual product browsing and side by side comparisons to free, Go, Plus and Pro users.
• Decision support. The assistant answers product questions. Google launched a Business Agent on January 12, 2026 that lets shoppers chat with brands directly on Search, starting with US retailers.
• Transaction. The agent completes the purchase. This is the least settled stage, and each platform handles it differently.

The practical consequence for a brand is that a large part of the comparison now happens on a surface it does not control, using data it may not have prepared for that purpose.

Two Open Protocols Structure Agentic Commerce: ACP and UCP

The Agentic Commerce Protocol and the Universal Commerce Protocol both define how AI agents, businesses and payment providers exchange information. They overlap, but they come from different platforms and cover different scopes.

	Agentic Commerce Protocol (ACP)	Universal Commerce Protocol (UCP)
Maintained by	OpenAI and Stripe	Google, co-developed with Shopify, Etsy, Wayfair, Target and Walmart
Announced	September 29, 2025	January 11, 2026
Status	Open source, Apache 2.0 license	Open source standard, endorsed by more than 20 partners
Main scope in 2026	Product feeds, promotions and secure payment credentials	Discovery, checkout, payment, order management, and since March 2026 carts, catalog access and loyalty linking
Where it runs	ChatGPT shopping	AI Mode in Google Search and the Gemini app
Merchant entry point	Product feed shared through ACP, directly or via partners; automatic for Shopify merchants through Shopify Catalog	Active Merchant Center account with products eligible for checkout

UCP is also designed to work with other agent standards, including the Agent Payments Protocol, Agent2Agent and the Model Context Protocol. For a marketing team, the detail of each protocol matters less than one shared principle: both expect clean, structured and current product data.

The 2026 Checkout Reversal Shows Where the Value of Agentic Commerce Sits

Agentic commerce was first presented as buying without leaving the chat. Six months later, OpenAI changed course, and that change says a lot about where brands should invest.

• September 2025: OpenAI launched Instant Checkout in ChatGPT for US users, starting with Etsy sellers and limited to single item purchases.
• January 2026: Google launched agentic checkout in AI Mode and the Gemini app for eligible US retailers, which remain the seller of record.
• March 24, 2026: OpenAI shifted away from Instant Checkout. Merchants can now use their own checkout, and ACP is positioned as the layer that carries product feeds and promotions for discovery.
• May 20, 2026: Google announced a Universal Cart for checking out with Google Pay across participating retailers, rolling out first in the US.

Two platforms, two opposite choices on checkout, one constant: both rely on product feeds to decide what to show. Checkout is where the models still diverge. Discovery is where competition between brands already happens.

Product Data Is the New Shelf in Agentic Commerce

In agentic commerce, the product feed plays the role the shelf used to play in a store: it decides what gets seen. Platforms say so openly. OpenAI states that its product results are organic, unsponsored and selected on relevance, with ads shown separately, taking into account signals such as availability, price, quality and whether the merchant is the primary seller. Google has added dozens of new Merchant Center attributes for conversational shopping, such as answers to common product questions, compatible accessories and substitutes.

That makes catalog gaps measurable. Take a catalog of 4,000 products where 1,200 lack complete attributes for size, material or compatibility. That is 30 percent of the range, because 1,200 divided by 4,000 equals 0.30, and those products are poorly equipped to match the detailed questions shoppers ask assistants.

A readiness checklist for the product feed:

• Complete attributes: variants, materials, dimensions, compatibility and use cases, written the way shoppers describe their needs.
• Current price and stock: agents favor products they can confirm are available.
• Reviews and ratings attached to products, since assistants use them to compare.
• Conversational attributes in Merchant Center where available, including answers to frequent product questions.
• Consistency between the feed, the product page and marketplace listings, so agents do not have to arbitrate between conflicting facts.

Marketplace teams already work this way, because an incomplete listing is harder for shoppers to find and to compare. It is the same discipline that Primelis Market automates on Amazon, now extended to AI shopping surfaces.

Measure Agentic Commerce Before It Shows Up in Revenue

Agentic commerce first shows up as visibility, and only later as sales. Waiting for revenue to prove the channel means reacting months late. Four indicators can be tracked today:

Indicator	What it tells you	Where to read it
Product presence in AI answers for category prompts	Whether agents shortlist your items	A fixed prompt set tracked weekly
Share of voice on AI surfaces	How you compare with similar brands	Google’s AI performance insights, available since September 2026 in the US, Canada, Australia, India and New Zealand
Referral traffic from assistants	Whether shortlisting turns into visits	Web analytics, segmented by assistant referrers
Conversion rate of assistant referrals	Whether those visitors buy	Web analytics and order data

Agentic Commerce Rewards Brands That Prepare Their Catalog First

Agentic commerce is already reshaping how products are discovered, while the way purchases are completed is still being decided platform by platform. ACP and UCP give the rules of the game, the 2026 checkout reversal shows that discovery is where value is secured, and the product feed is the asset that decides whether a brand is shortlisted. Across 300+ global brands and more than 1 billion dollars of media managed each year, the retailers best placed for AI shopping are the ones that treated feed quality as a performance lever before it became a channel.

A 30 minute review of your product feed against the attributes AI shopping surfaces read will show which part of your catalog agents can recommend today, and which part they cannot.