Agentic Commerce Protocols: What ACP and UCP Mean for Your Product Feed

Agentic commerce protocols are open technical standards that let an AI agent discover products, compare them, and complete a purchase inside a chat or assistant. Two now matter to advertisers and merchants: OpenAI’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol. Both depend on the same asset, a clean product feed.

What the two protocols are

Both protocols solve the same problem. Without a shared standard, every shopping agent would need a custom integration with every retailer. A protocol gives agents one predictable way to read a catalog and place an order, and gives merchants one integration instead of many.

Stripe’s September 29, 2025 announcement describes ACP as an open standard co-developed by Stripe and OpenAI, with a single integration that lets a business sell through AI agents while keeping control of products, branding, and fulfillment. At launch, US Etsy sellers were available to US ChatGPT users, and Stripe said over a million Shopify merchants were coming soon.

Google’s January 11, 2026 announcement frames UCP as an open standard that works across the entire shopping journey. For a buyer, the implication is that product discovery is moving from a results page you bid on to a conversation where an agent decides which catalog entries answer the question.

What ACP asks of merchants

Product feed: a structured file or API stream that lists every item you sell with its attributes. OpenAI’s commerce documentation states that feeds give ChatGPT the catalog data it needs to index products, and that each record must include a title, description, images, price, and availability.

The same documentation tells merchants to validate required fields on every record against the specification, to send the full feed once a day by file upload, and to push changes through the API. Promotions can only be sent through the API. It also states that feeds are currently limited to approved partners, who apply through a form.

The implication is practical. Access is gated, so most small merchants will reach ChatGPT shopping through a platform such as Shopify or a processor such as Stripe, not by building the integration themselves. Even then, the platform can only pass along the data you maintain.

What UCP asks of merchants

Google’s announcement says eligible US retailers can enable checkout through UCP in AI Mode and Gemini, and that retailers remain the seller of record. Retailers can also activate a branded Business Agent, a chat experience on Search, from Merchant Center.

The data requirement is the more telling detail. Google states that it is adding dozens of new Merchant Center data attributes for conversational discovery, covering answers to common product questions, compatible accessories, and substitutes. These complement existing feeds and were rolling out first to a small group of retailers, with wider availability planned over the following months.

Google also described a Direct Offers pilot in Google Ads, where advertisers present exclusive offers to shoppers ready to buy and Google’s AI decides when to show them. It said it continues to test ads in AI Mode. For paid media teams, the shift matches what we covered in Google AI Max for Search: inputs matter more than manual bidding levers.

ACP and UCP side by side

Dimension	ACP (OpenAI and Stripe)	UCP (Google)
Announced	September 29, 2025	January 11, 2026
Where it works	Instant Checkout in ChatGPT	Checkout in AI Mode and Gemini
Named backers	Stripe and OpenAI; Etsy and Shopify merchants	Shopify, Etsy, Wayfair, Target, Walmart as co-developers; 20+ endorsers
Required product data	Title, description, images, price, availability on every record	Existing Merchant Center feed plus new conversational attributes
Access	Feeds limited to approved partners	Eligible US retailers, rolling out in stages
Seller of record	Merchant keeps control of fulfillment	Retailer remains seller of record

A feed-readiness sequence

The two protocols differ in plumbing, but both reward the same preparation. The order below puts the cheapest, most protocol-neutral work first, so it holds up whichever standard your customers end up using.

1. Audit required fields: confirm every product has a title, description, images, price, and availability, since missing fields are the failure ACP’s validation step targets.
2. Fix price and stock accuracy so the feed matches your site, because an agent that quotes a wrong price creates a refund, not a sale.
3. Rewrite descriptions to answer buyer questions directly, including size, compatibility, and materials.
4. Add substitute and accessory relationships where your platform supports them, which maps to the attributes Google describes.
5. Set a daily refresh, matching OpenAI’s recommended cadence, and route promotions through the API where the platform allows.

Teams already working on owned customer data will find the overlap in our guide to first-party data and AI targeting.

What is still unproven

Both announcements describe capability, not results. Neither OpenAI’s documentation nor Google’s announcement, as reviewed, published conversion data showing that protocol-ready merchants sell more. Access remains partial: ACP feeds are limited to approved partners, and Google’s new attributes began with a small group of retailers.

We also looked for current advertiser terms for paid placements in ChatGPT, and for a figure on the size of its ad pilot circulating in secondary coverage. We could not confirm either from a primary source, so we left them out. Check each platform’s documentation before budgeting, because terms in this area change quickly. For the wider picture of what is and is not autonomous, see agentic advertising in 2026.

Frequently Asked Questions

Do I need to choose between ACP and UCP?

No. They are separate standards run by different companies, and a clean product feed serves both. Most merchants will connect through their commerce platform or payment processor, not directly.

Does a better feed guarantee that an AI agent recommends my products?

No. Neither company has published evidence that feed quality alone decides selection. Complete, accurate data is a prerequisite for being considered, not a ranking guarantee.

Is this relevant to a service business with no product catalog?

Not directly. These protocols cover physical and digital product purchases. Service businesses should focus on how AI search describes them, which is a content and entity question.

Preparing structured data for AI shoppers is the kind of unglamorous work that decides whether an agent can use your catalog at all. At Grow with Hafsa we build AI marketing and performance systems that start from clean inputs, and we are happy to look at your feed with you.