Universal Commerce Protocol (UCP) Explained: What Every Ecommerce Brand Must Know

Episode Transcript

00:00:00:01 - 00:00:22:21
Mike
Hello and welcome to another episode of Growing E-commerce. I’m your host, Mike Ryan. Unfortunately, Chris can’t join us today, but we’ll be back in the studio for normal production next week and we look forward to seeing you then. Meanwhile, I’m going to revisit a topic that we first covered back in February: UCP, the Universal Commerce Protocol. What is it?

00:00:22:23 - 00:00:48:15
Mike
How does it work? This was actually our single most popular episode of the year so far. It even beat an interview with Marvin about AI and PMax for search, which we accidentally sent to the podcast feed. That shows you how popular this topic is and how much interest there is. We’ll get into a little bit about why that is.

00:00:48:16 - 00:01:21:19
Mike
Let’s just jump into it. I think the situation right now is one of fear, confusion, or impatience. There are a lot of emotions going on. Organizations, broadly speaking, have an AI mandate. Across many levels, they are asking: Are employees using AI enough? Are they using it too much? But also, when it comes to topics like the future of search, web traffic, and the impacts on our customer acquisition channels.

00:01:21:20 - 00:01:46:12
Mike
Everyone saw what happened to publishers when AI overviews rolled out, and people are spooked. No one wants to go dark or become invisible. No one wants to get left behind or miss this. That is the urgency driving interest here. The next thing that comes into place is that the topic is complicated.

00:01:46:12 - 00:02:14:15
Mike
It’s hard to understand. You’ll see wildly varying information about how to implement it or what is really necessary. Because of all this, I want to simplify this topic. It’s worth revisiting; we know a lot more now than we did back in February. I want to spell out: what is UCP really? Why does it matter?

00:02:14:19 - 00:02:47:13
Mike
And why is Google building this and everyone signing on? Then we’ll get into some of the basics of implementation. I want to give you a reasonably detailed answer that can help you understand what this is actually about. Within the scope of a podcast episode, I can’t get into every single implementation detail or scenario. I’m not a developer, but I think that’s valuable here because I’m just out here trying to pay attention.

00:02:47:13 - 00:03:16:12
Mike
In January, I was reading all the documents, and I’ve come back and looked at them again more than once. I think we can talk about this at eye level and come a bit further together. That’s the goal of this episode. To kick that off: what the heck is UCP? This in and of itself is a subject of a lot of confusion.

00:03:16:14 - 00:03:44:08
Mike
To describe what it is, I think first we have to take a step back and look at this topic of agentic shopping. If you’re a long-time listener to the show, you’ll know I don’t particularly like that phrase. I view this more as AI-assisted shopping. It’s a little less buzzworthy and describes more how consumers experience it.

00:03:44:12 - 00:04:16:18
Mike
Consumers don’t know what agentic commerce is. Consumers know that an AI is helping them find products and do stuff. If you just go back a year, the popular idea was that computers would be literally navigating our e-commerce shops, scrolling through, adding things to the basket, and doing the whole checkout.

00:04:16:18 - 00:04:50:12
Mike
There were big articles about this in all kinds of publications, but the feedback was not good because, frankly, this was never how this was going to work. Maybe we’ll get there one day, but it’s dangerous, slow, expensive, and doesn’t scale. You’d hear stories about an AI spending 45 minutes to buy the wrong product with the most expensive shipping option.

00:04:50:12 - 00:04:59:07
Mike
This got everything off on a very bad footing.

00:04:59:09 - 00:05:26:15
Mike
This isn’t really how we should imagine this happening. Again, there’s a lot of public awareness of this now, which increases what we call the trust gap in terms of consumer adoption of AI for shopping. Consumers think, “I would never give my credit card to an AI.” And that’s good; you shouldn’t give your credit card details to an AI.

00:05:26:16 - 00:05:42:13
Mike
I also wouldn’t do that. But that’s not how agentic commerce or AI-assisted shopping is actually working or will work in the future. This is basically a misconception at this point.

00:05:42:15 - 00:06:13:18
Mike
The alternative to this would be the old-school plumbing situation, and this is also not a proper solution to agentic shopping. What you can do here is, on the one hand, imagine an indeterminate large list of all the AI surfaces where consumers are and might use an AI system. This could be things like ChatGPT or Google AI mode.

00:06:13:18 - 00:06:47:20
Mike
Or if we look at the direction Google is taking, that could also be YouTube or Gmail. Google wants to embed a universal card across all of their surfaces so that you can shop identically from basically anywhere on Google. It can also be other surfaces in the future, like AI-powered experiences in a marketplace we don’t even know yet.

00:06:47:21 - 00:07:22:19
Mike
On the other hand, you’d have all the merchant shops and the merchant catalogs behind that, and you would need a connection like an API between every single one of these. This is something that I nicknamed the “glass spider web,” because it would be this unimaginably complex network of connections—totally brittle, fragile, and built to break.

00:07:22:19 - 00:07:52:13
Mike
So this was also not how this was going to work. What we have instead is a pretty elegant solution. It is complicated, but once you wrap your head around it, it’s not bad. It’s this thing co-developed by Google and Shopify called Universal Commerce Protocol.

00:07:52:15 - 00:08:25:07
Mike
This provides a third way. It is more like conventional plumbing, but it’s scalable and it does give AIs the freedom. But it can also power experiences that have nothing to do with AI. We have that whole list of consumer surfaces and platforms, and then we’ll have a list of business platforms, retail shops, and stuff like this.

00:08:25:08 - 00:08:59:05
Mike
The other complex layer is that there are payments, CRM, ERPs, and PIMs in here. Universal means it should be compatible; everyone should be able to use it. Commerce, I hope we all know what that is.

00:08:59:09 - 00:09:30:17
Mike
The maybe mysterious word here is “protocol.” To give another example, a protocol is basically a way for systems to communicate. I work at a company that uses Google Workspace. Some of my friends or customers work at companies that use Microsoft Outlook or Microsoft Teams. We are able to send emails to each other despite that fact.

00:09:30:17 - 00:09:58:00
Mike
It’s because there’s an email protocol that standardizes things and allows these different systems to communicate. We’re talking about a Universal Commerce Protocol, and it consists of these different layers that help bridge between the consumer platforms and the business platforms. The first layer of that is a services layer.

00:09:58:00 - 00:10:30:23
Mike
In this case, we’re talking about shopping, because actually, commerce is a lot bigger than just buying products. There’s also travel, hotels, and booking. There can be a lot more that will be supported here eventually. But the first use case being tackled by UCP is shopping. That should give you a glimpse into how big this topic can be.

00:10:31:04 - 00:11:02:04
Mike
They want to reinvent not just shopping, but all of commerce. They want an infrastructure to handle this across these different kinds of verticals. Within the shopping vertical, it comes down to capabilities. This is the next layer in the communication between these two sides of the market.

00:11:02:08 - 00:11:30:09
Mike
This is about defining what e-commerce actually is. What are the main things that need to be done? What are the biggest Lego blocks of which this whole thing is built? These are things like catalog, carts, checkout, and order management. In order for AI to navigate this, there just needs to be these defined areas of activity.

00:11:30:10 - 00:11:58:03
Mike
Then there’s another level to this which are called “extensions.” These are things like discounting and others. Capabilities are the building blocks or the cornerstones, and extensions are the smaller things. I think they recognize that there’s a risk of scope creep here, and they don’t want to trivialize what a capability is.

00:11:58:03 - 00:12:45:05
Mike
At a high level, you can have something like AI mode and then your web shop built on your e-commerce platform. Your shop will support certain capabilities, the AI will support certain capabilities, and they talk to each other. This is a thing called “negotiation.”

00:12:45:06 - 00:13:02:08
Mike
They look at each other and find out what they have in common; that is how they know they can interact with each other. This is very important.

00:13:02:12 - 00:13:40:14
Mike
The final most technical layer of this whole thing are what’s called “transports.” This can be APIs like a classic REST API. This is an architectural style for building an API, and it’s the way a lot of the internet functions. Then there are these newer ones like MCP (Model Context Protocol), Agent-to-Agent Protocol, and others.

00:13:40:14 - 00:14:10:23
Mike
For these newer, more general-purpose AI protocols that have been built, Google and Shopify don’t care how you want to transport information; it’s going to be supported. This is the point of it being universal. They’ve built this system that helps businesses and AI services talk to each other. That’s the point of all of it.

00:14:11:03 - 00:14:33:03
Mike
There doesn’t need to be a custom API between every single thing, and the AI does not need to be going on your actual website and clicking around. That is the whole point of UCP. I hope I’ve explained it clearly enough.

00:14:33:03 - 00:14:53:19
Mike
The next thing about UCP is very important: the idea of universality. For any kind of standard like this to succeed, it needs to be accepted and adopted. We talked about this on past episodes. OpenAI had a competing standard.

00:14:53:20 - 00:15:20:05
Mike
Theirs came out first; they were co-developing it with Stripe and it was called ACP. It still exists, but it’s not going to make it. Across every layer, UCP has been endorsed and accepted, which is very important to it actually becoming a standard.

00:15:20:06 - 00:15:50:20
Mike
It doesn’t matter which category you’re talking about; the big AI hyperscalers have all accepted it: Amazon, Meta, Microsoft, and Google. Google likes to joke that this is probably the only thing they ever agreed on, which might actually be true. Then marketplaces and retailers: there have been big endorsements from Walmart in the US, Zalando here in Europe, Etsy, Wayfair—the list goes on and on.

00:15:50:20 - 00:16:16:08
Mike
Everyone who’s anyone has endorsed this. Then there are the commerce platforms. There’s probably a little bit more work to do here, but Shopify, Salesforce, and Commerce Tools have shown acceptance as well. This is where it gets a little complicated when it comes to implementation because not everyone is a Shopify shop.

00:16:16:08 - 00:16:46:16
Mike
Even for Shopify, it’s not perfectly plug-and-play. We know plenty of retailers and brands who have bespoke commerce platforms. We’ll have to see how this develops. I also want to mention the payments layer because this is super important: Visa, Mastercard, PayPal, Stripe, Klarna, and every other one you can think of.

00:16:46:18 - 00:17:13:19
Mike
Probably the broadest adoption is among payment processors. They’re all on board. If there are payments to be processed, they will be there. They’re very excited about this. That’s important too, because some of this will push you toward Google Pay, but ultimately every payment processor is supported.

00:17:13:21 - 00:17:42:00
Mike
Let’s talk about implementation. A friend of the podcast, Joey Bednar, expressed some frustration that there are videos out there saying all you need to do is populate this one field in your Google Merchant Center and you’re good to go. Then there are people like me mentioning that the entire UCP dev documentation would probably be a 300-page book.

00:17:42:03 - 00:18:07:01
Mike
Where does the truth lie? It depends. The worst place you can start right now is probably UCP.dev. The reason I say that is because that is designed for every use case that could possibly happen.

00:18:07:02 - 00:18:35:03
Mike
It might not be clear to you what is relevant and what is not. My recommended place to start is Google’s documentation. You might be getting a little confused—wait, Google has UCP documentation?

00:18:35:03 - 00:18:57:12
Mike
This is where you really have to untangle these things. Again, UCP is the total framework; it’s the whole structure I described of all the capabilities and use cases. Google has their own implementation of UCP, which is in Google AI mode and Gemini, and it’s also manifest in organic search.

00:18:57:12 - 00:19:24:13
Mike
Start there because it gives you some boundaries. It is right now the biggest opportunity in terms of UCP because we’re so early here. This will be big in the next few years; it’s going to take time. It will depend on more implementations and use cases coming forward, and on consumers adopting those. This is not going to happen overnight.

00:19:52:15 - 00:20:38:09
Mike
For the people telling you that you’re missing out right now, there’s a little bit of truth to that because if you did the things you need to do to succeed in UCP, it’s all interconnected and valuable to do. It can’t hurt you to start working on this, but it’s also perhaps not as necessary right now as some people would say or as this organizational AI mandate would imply.

00:21:12:02 - 00:21:41:10
Mike
I see things jumping to the top of roadmaps in just about every client call I’m in. You need to not forget about the world that exists today and business as usual. What are the optimizations you should have been doing last year as well? It just needs to be balanced. Let’s get into the implementation.

00:21:41:12 - 00:22:10:06
Mike
I would start with Google because their AI mode is the most advanced. Agentic commerce features in ChatGPT are on pause right now. Perplexity and other services—Amazon is kind of its own thing. Right now, Google is basically the show.

00:22:10:08 - 00:22:37:22
Mike
There are a couple of core steps you need to do on the infrastructure side. You have to publish something called a “discovery profile.” I believe this is one thing that Shopify is pretty good at automating, but it still might not be perfectly plug-and-play. You have to host this profile, which says things like which UCP versions you support.

00:22:37:22 - 00:23:10:23
Mike
It also says which capabilities you have enabled—for example, catalog, cart, or checkout. This allows the AI, when it meets your business, to negotiate and see what you have in common. That’s all done through the discovery profile. There will also be your transport endpoints, like a REST API or MCP.

00:23:10:23 - 00:23:36:08
Mike
The next thing would be to choose and secure those transport bindings. This is how the platforms and shopping agents talk to each other. As I mentioned, it’s all supported. A REST API is probably the best choice for a lot of developers because they’re already familiar with it.

00:23:36:08 - 00:24:02:21
Mike
I suspect REST APIs will have a lot of gravity because they are already so dominant. By the way, this seems to be what Google prefers. If you go into Google’s documentation, they want you to work with REST APIs.

00:24:03:01 - 00:24:31:22
Mike
If you go in Shopify’s documentation, they’re focused on MCP. I love when Mom and Dad are fighting. You guys developed UCP together! But maybe that’s the beauty of it as well because, again, this universal thing means it doesn’t really matter—it’s all supported anyway.

00:24:32:00 - 00:25:02:17
Mike
Then you need to configure your payment handlers. This is specifying the payment methods that you accept, like which cards, wallets, or processors. If you are accepting raw credit card data, you need to comply with a 12-step framework called PCI DSS. You’re probably delegating this to a provider like Stripe or PayPal.

00:25:33:08 - 00:26:02:11
Mike
I don’t want to talk about identity linking because I’m based in Europe, and I think this is going to be a big sticking point based on GDPR and privacy regulation. There’s going to be a lot of interesting news headlines in a while. I’m not commenting on that; I’m not a developer and I’m definitely not a lawyer.

00:26:32:04 - 00:27:04:21
Mike
Then there’s implementing capabilities. The big three for you to worry about would be catalog, checkout, and order. There’s also carts, but Google is building a universal cart. For the catalog: what is this? It’s functions about searching your catalog and getting products which allow the AI agents to discover and fetch your pricing and real-time product availability.

00:27:04:21 - 00:27:31:01
Mike
That is pretty much handled by Google Merchant Center. Google Merchant Center has that stuff in there for you. This is a huge advantage Google has from a market standpoint because they have tens of millions of merchants in Merchant Center already, whereas OpenAI has to start from scratch to build their merchant supply.

00:27:31:03 - 00:28:06:12
Mike
Then there’s checkout. There is some work to do here, and this will depend partly on your e-commerce platform and what they’re going to do for you in the longer term. I think more of this will become plug-and-play as it should. It shouldn’t be rocket science, and in the intermediate term, it won’t be anymore.

00:28:06:14 - 00:28:36:10
Mike
But the core job of checkout is to handle fields like “create checkout,” “update checkout,” and “complete checkout.” For the AI to be able to do this, they need to know what’s going on. This needs to be continually updated during the session with flags like “requires escalation” or “ready to complete.”

00:28:36:10 - 00:29:05:11
Mike
One of the reasons the documentation is so long is that they duplicate this for every transport. But again, with Google, you just have to focus on the REST API documentation. For order, this is basically supporting webhooks so that the order status can be understood—is it shipped, in transit, or delivered?

00:29:35:04 - 00:30:01:07
Mike
My biggest tip is to start with Google. They need you to prepare your Merchant Center account, which is your shipping, returns, and product feed. That’s not rocket science. They want you to set up your business with Google Pay—until antitrust regulators come in, at least—but other options are supported.

00:30:01:08 - 00:30:28:21
Mike
Publish your UCP profile, complete the native checkout integration with the three core REST endpoints for session creation, update, and completion, and sync the order status. If you go check out Google’s documentation, it’s somewhere in the middle. It’s not as simple as activating a single field, but it’s also not as complicated as reading 300 pages.

00:31:21:07 - 00:31:52:05
Mike
Let’s talk about native commerce attributes in Google Merchant Center. This would be where the false claim comes from that all you need to do is populate this one field. You do need to do that, but it’s not the only thing. The native commerce attribute is a per-product attribute in your feed that you populate with a 0 or 1 value.

00:31:52:05 - 00:32:04:00
Mike
This determines if that product is going to be allowed to have native checkout. Not all of your products are eligible for native checkout.

00:32:04:02 - 00:32:32:19
Mike
There might be strategic considerations as well. Google is becoming a marketplace here. It’s really a change in the relationship we’ve had with Google, where they were a traffic acquisition source. We used to have all the first-party data on our websites and own the customer relationship, and now they get all that fun data. This is a potential risk.

00:32:53:11 - 00:33:17:20
Mike
Not all products are eligible. Common sense stuff: custom products, age-restricted products, and subscription products are not supported by UCP yet. UCP is not playing the age verification game, and their checkout is too simple right now for subscriptions.

00:33:45:01 - 00:34:12:02
Mike
There is a “consumer notice” attribute that you may need to populate for safety or legal declarations. Then there is a “merchant item ID.” If you have a different item ID in your checkout than what is in your feed, that’s a mismatch. The merchant item ID is a unified ID you provide as a supplement.

00:34:42:15 - 00:35:06:20
Mike
All of these should be delivered by your supplemental feed. Just a couple of other notes: your products need to be in stock with complete shipping and tax setups, and you’ll need a return attribute or return policy label. If your feed is prepared and native checkout is available in your market, your products can start appearing.

00:35:29:16 - 00:36:07:02
Mike
UCP is also just useful infrastructure. In some ways, it’s bigger than the AI-assisted stuff because right now there are already “buy” buttons being populated in Google organic results. It has nothing at all to do with AI, but it’s infrastructure that can be used to power other experiences.

00:36:07:04 - 00:36:34:08
Mike
You might be an AI skeptic—consumers will never do this—and you could be totally right. But it doesn’t matter because the infrastructure will also support existing experiences. This achieves a very similar result to “Buy on Google,” and “Buy on Google” had nothing to do with AI.

00:37:00:03 - 00:37:30:18
Mike
The next horizon is optimizing discoverability. For some of us in Europe, it’s the first horizon because native checkout isn’t here yet, but we can already start working on our discoverability in AI services. UCP is bigger than Google and bigger than shopping.

00:38:00:06 - 00:38:24:04
Mike
I get funny questions like “When is UCP coming to Europe?” UCP is already here—Zalando is building stuff with it. The question people are actually asking is: when will native checkout come to AI mode in Google? These are different but overlapping things.

00:38:46:16 - 00:39:10:19
Mike
Let’s talk about conversational attributes. There are now six of these available globally, so you can start improving this stuff now. These support new ad formats coming to shopping and the classic Google SERP. Google will also be using these to write new titles for you.

00:39:48:16 - 00:40:12:02
Mike
Even if you don’t believe in the AI stuff, these can help you have better Google Shopping titles. The first six are: Q&A, document link, related products, item group titles, variant options, and popularity rank. Five of the six came from OpenAI’s feed spec.

00:40:12:04 - 00:40:48:20
Mike
What Google is doing right now is just closing any gaps with what OpenAI tried to offer. OpenAI is activating their feed for the product listing ads they’re starting to sell. Google doesn’t want any daylight between OpenAI’s feed spec and theirs. I think OpenAI did a good job writing their specifications.

00:41:16:14 - 00:41:39:03
Mike
Regarding “Question and Answer,” you can add up to 30 FAQs. If you have FAQ data, this is a great asset. “Document link” even gets privilege over FAQs; it’s Google’s preferred source. Imagine you have a product detail page with links to a spec sheet or instruction manual—Google’s AI is just going to read that.

00:42:03:01 - 00:42:33:16
Mike
I think this is a preferred source because it comes straight from the manufacturer, so there’s very little room for error. If you have these, they are a golden standard of verified information. It’s especially great for categories like electronics.

00:43:00:03 - 00:43:31:12
Mike
I love “Related Products” too. You can specify if things are part of a set, are required parts, or are frequently bought together. This helps the AI support your average order value (AOV). It helps with upselling and ensures a good customer experience so if they bought Part A, they also got Part B.

00:43:31:15 - 00:43:57:14
Mike
A shout out to “Popularity Rank” as well. You can provide the percentage rank of your sales, providing your own bestseller data. Google can write copy like “the number one bestseller,” and we know that label drives a lot of volume on Amazon.

00:44:42:21 - 00:45:11:15
Mike
There are attributes that have been around for years but are finding new life: “product highlight” and “product detail.” Google introduced these because they wanted to better rival Amazon’s experience, like those bullet points on a product detail page.

00:45:11:15 - 00:45:43:03
Mike
The product highlight is a simple list of benefits—4 to 6 bullet points where you’re selling your product. The product detail is where you’re telling about your product, which is more spec-oriented. But now they are inviting you to submit a wide variety of product facts, like a flavor or an occasion.

00:46:15:04 - 00:46:35:15
Mike
This is all going to be important. Go look up the episode about the new share of voice metric in Merchant Center to understand how visible you are in AI services like Gemini. There’s also something coming to Merchant Center called UCP analytics.

00:47:04:01 - 00:47:30:14
Mike
They’re going to give you metrics like “number of clicks to buy,” “purchases” (total successful transactions), and “purchase rate,” which is your conversion rate for native checkout. That will be a good one to watch. Average order value: you can see if that goes up when you add related products.

00:47:59:22 - 00:48:38:02
Mike
Cancellations, refunds, and “account linking,” which is basically your customer match rate. Identity linking will power things like people getting their member pricing inside of AI mode and not just on your website. They’ll have metrics for the success rate of those links. This is a first step in the right direction of giving merchants data back.

00:48:38:03 - 00:49:19:18
Mike
Let’s just recap. I hope that UCP and the things around it are a little clearer to you. There’s a lot of urgency and not a lot of understanding, and that’s an uncomfortable situation. My key takeaways: don’t boil the ocean. Start with Google; it’s the biggest opportunity and they’ll tell you exactly what you need to focus on.

00:49:52:04 - 00:50:13:17
Mike
Step two is improving your feed to improve your discoverability in general. It’s just something you should do anyway. I had a fun time talking this through with you. Next week we’ll be back in the studio with Chris. This has been another episode of Growing E-commerce, brought to you by Smarter Ecommerce, also known as smec.

00:50:13:18 - 00:50:31:12
Mike
You can learn more by visiting Smarter Ecommerce. If you enjoyed this episode, give us a review—but only if it’s a good one! Thanks. We’ll see you next time.