Agentic commerce: 15x more orders from AI channels

Episode recorded in Polish (English subtitles available on YouTube); this is our edited English translation. Recorded in spring 2026 (released 6 May 2026), so figures and Shopify features reflect that time.

Why is everyone asking about selling through LLMs?

Tomek: Hi, this is Tomek Runowicz and Maciek Czerniak. We run wecanfly, an agency that helps established brands with complex migrations to Shopify Plus. And this is Flying with Shopify, the channel where we talk to e-commerce experts and founders. Welcome to another episode.

Today we're sharing one cup of coffee, because you have to adapt to the situation, just like you have to adapt in e-commerce, Maciek. And that's roughly what we want to talk about: the thing we hear about from every direction, at every event we go to, whether we're speaking or just attending. In the corridors everyone is asking very similar questions: where is commerce heading, where will sales actually happen, where should we make our products available, and how much will e-commerce change as more sales go through LLMs?

Maciek: I've noticed the same thing. At every event we attend, brands are getting very interested in agentic commerce, meaning selling through LLMs. There are lots of questions about what brands can realistically do right now and how it's all going to work.

Tomek: Let's be more specific about what those questions are and where they come from. The most common concern among merchants is that most of them are seeing organic traffic to their site fall, and that drop doesn't necessarily come with a drop in revenue. That isn't intuitive for them, and it's what usually kicks off the discussion.

Maciek: There are estimates showing that brands doing more than 100 million dollars a year are seeing traffic drops of around 10%, and the smaller the revenue, the bigger the drop. Brands at around 10 million dollars are seeing drops of 40 to 70% in some market segments.

Tomek: With our clients, I'd say it's somewhere between 15 and 30%. But I'd add why that is. The data you mentioned is based on the US market, which leads in most areas, so any shift happens there first. I've seen that report, and up to 70% is a huge number. We're not seeing that here yet, because in Poland the shift is only just starting. I think it shows the direction we're heading in.

Maciek: Speaking of reports, Shopify has shared data with a few interesting statistics. Two caught my attention most. First, the number of AI orders attributed to agentic channels grew 15 times between January 2025 and the end of the first quarter of this year. Second, and really interesting: orders placed through LLMs are worth 30% more on average. These are bigger baskets.

From being found to being recommended

Tomek: That brings us to how strongly changes in shopping behaviour show up in these channels. Until now, a store owner's job was to rank in Google, where customers searched for the product they needed at that moment, and we wanted them to find us as quickly as possible. Now we're moving to a scenario where discovery, the stage where people look for the product they want to buy, doesn't happen in our store. It happens through agents: in Gemini, ChatGPT, Copilot, wherever more and more people use AI as their assistant. So today we don't just want our store to be found; we want it to be recommended. And when a customer lands in the store after that kind of recommendation, their purchase intent is completely different. They're past the consideration stage, they're well informed, they often know exactly which features convinced them, and they arrive ready to buy.

What is agentic commerce?

Maciek: True. So let's talk about agentic commerce today, and about visibility: what to do to be visible and recommended there. First, a definition. We understand agentic commerce as purchases made during a conversation with an AI engine. That's the first scenario. The second is purchases made by agents autonomously, with our consent but without our involvement. That's probably the shortest and simplest definition.

Tomek: Shopify's vision is for brands on Shopify to be "AI native by default", ready for this from day one. Shopify isn't just reacting to market changes; it's shaping them. I mean above all the Universal Commerce Protocol, the foundation for all this agentic traffic, which Shopify developed together with Google. Now other giants of US e-commerce are joining, Amazon, Etsy, I think Walmart, and a few others. Can you give us a snappy definition of UCP?

Universal Commerce Protocol: a shared language for agentic commerce

Maciek: The Universal Commerce Protocol is a communication protocol that lets agents process commercial transactions. On the internet you have HTTP and HTTPS for serving web pages, FTP for uploading files to a server, and MCP for communication between agents and other tools. In the same way, UCP will handle the whole area of agentic commerce.

And importantly, there's no escaping it now. It's a standard developed by the giants, and everyone will have to join.

Tomek: Good luck building your own. Especially since Shopify, as you said, has taken the stance it probably always takes: don't follow trends, set them. Don't wait to see what happens; help create it, because then you have a say in how it looks. And here's another interesting thing Shopify announced, not to its own merchants but to the whole e-commerce industry. Stores on Shopify, as you mentioned, are ready for agentic selling by default, so there's nothing you physically have to do.

Maciek: That's important too. You don't have to add a new sales channel to your store and spend ages configuring it. Formally it will be treated as a new sales channel, but it's switched on by default, and you can switch off each of these channels individually later.

Shopify Catalog and the Agentic plan for stores not on Shopify

Tomek: Right, I put that too loosely. What I meant is that you don't need to add a new app that you then have to configure. You just get it. But Shopify went a step further and told store owners on other platforms: we don't need to, and don't want to, push you to migrate to us. You may not be ready, it may be a big investment, you may have concerns. That's fine. We have a service for you based on Shopify Catalog: list your products in our catalog, and agents will read them just like products from Shopify stores.

Maciek: Then let's explain what Shopify Catalog and the Agentic plan you mentioned actually are. Shopify Catalog is where your data is stored on Shopify's servers, mapped so that it's structured, easy to read and accessible to agents.

Tomek: And it's integrated with your platform in real time, so you don't have to worry that the data in the catalog doesn't match what's currently in your store.

Maciek: That matters too. Let's finish the Universal Commerce Protocol topic. It covers the whole purchase journey. First, it lets you be discovered. Second, it lets the transaction be processed. In ChatGPT's case that runs on Shopify Checkout, which also handles various edge cases: it understands your store's promotions and discounts.

Tomek: I'd go beyond discounts and say your business logic. That's what matters.

Maciek: Exactly. Another important point: if you're on Shopify, you have Shopify Catalog by default. If you add it as a layer to a store that isn't on Shopify, you also feed agents structured data and improve your discoverability, but above all you control what information about your products is sent to agents.

Shopify Knowledge Base: controlling what agents say about your brand

Maciek: And that brings us to the third really important thing, an app Shopify added: Shopify Knowledge Base. It's an app built by Shopify in the admin that lets you control how agents answer questions about your brand.

Tomek: A quick aside, because it's one of the questions that keeps coming up in those corridor conversations. Anyone who searches for product information through agents today wonders what information the agent will pull from their store, and whether it's what they'd want shown for that product. The problem is that the agent doesn't necessarily rely on what it finds in your store. It may rely on information from someone else. So when you sell through this channel, it's really important to control the narrative, meaning what's being said about your brand.

How agents read your store, and why product data matters more than layout

Maciek: That's right.

Tomek: In a moment we'll get to how to build visibility in these places and get ready to be recommended as a store. But first, the technical side of the change. We can see that chats and agents read our store differently. Nice images and a minimalist, inspirational product page layout aren't necessarily the best source for the information they're after. Above all, we need to make sure our product data is very clean, clear, structured and as complete as possible. We fill our stores with genuinely practical information, not generic text written for robots. That's the difference from old-school SEO, when we sometimes stuffed content with keywords just to tick a box. Today the product information has to contain what users actually search for and need, for example to compare products and answer all their questions. So we need to work on how we present product data, so that it's complete, readable and, above all, useful, rather than written for some rule or algorithm.

Maciek: Yes. That's one of the reasons the Universal Commerce Protocol has this data standard. The catalog has required fields, such as title, price, product image, URL or ID, and a set of recommended fields that are worth filling in. And here's a good tip for Shopify store owners: just talk to Sidekick and ask whether your data is structured and complete, and whether anything is missing. A very quick audit. We'll come back to how to tackle this.

SEO vs GEO: from being searchable to being recommended

Tomek: OK, Maciek, we've already started on this. The way we want our store to be, let's say, positioned is changing, because the environment where the store appears is changing: it's now language models. Until now there was SEO; now there's GEO. What is it, how is it different from SEO, and how much do we need to prepare for the change?

Maciek: We've had guests on the podcast who talked about this, but for anyone who's been living under a rock and hasn't heard of the trend: we're talking about visibility in ChatGPT, Gemini, Claude or Copilot. The question we often hear from merchants is: what do I have to do to show up there? Because with SEO the playbook is well known. You build links, create valuable, engaging content, build your domain authority, have a correct code structure with schema, and so on. The goal of SEO as we know it was to let people discover the brand, to be searchable. GEO focuses more on being recommended.

Similar things work here as in SEO, but SEO alone isn't enough to be visible in LLMs, and the playbook looks completely different. We'll break it down in a moment. What matters is that LLMs do rely on SEO to some extent. When you search for a product or any information, you'll notice the model also runs a search, goes through various pages and shows you the list. But once the Universal Commerce Protocol is in place and the data is available through an API, the model will prioritise the data it gets that way, and only then read pages.

The three pillars of visibility in LLMs

Maciek: To set out the rules, meaning what we as merchants should do, we need to talk about the three main aspects of LLM visibility we have to take care of. The first is data: presented properly, in the right structure, delivered through Shopify Catalog and the Agentic plan you mentioned. It's data presented by default as definitions and values, easy for agents to take in and read.

The second is user sentiment about your brand: what users write and say about you on social media, on YouTube and on sources other than your own site. On the one hand there are the reviews you publish on your site, but there are also opinions about you on Facebook, Instagram, forums. Who knows what AI will have access to and what it will be able to read. Influencer collaborations will probably play some part too, and media coverage in traditional publications certainly will.

Take Eight Sleep as an example. They make mattress covers that regulate sleep temperature. Anyone who watches the Huberman podcast probably already has one on their wish list. They did a big research study and published an extensive piece on sleep, which was then cited very widely, in industry media and beyond, and got a lot of attention. They got a lot of citations and, as a result, a lot of mentions. That's another big area.

Tomek: Yes. When it comes to citations and user reviews, we should mention the 70/30 split, which is about citations from external sources, but, and this has to be stressed, credible ones. If you look at what SEO sometimes looked like, building links from sites with very little connection to us, it didn't deliver results back then, and it certainly won't now. Surprisingly, Reddit is highly valued, and all kinds of forums, more in the US than here, because Reddit isn't as popular in Poland.

The context of a review matters as well. A review that just says "great product" isn't enough. What can matter is someone writing: I was in this specific situation, and thanks to this product or service I handled it really well. Or: this backpack worked brilliantly on my trip to the mountains. Statements like that, with context, later answer the questions people ask in chats, for example in Gemini: "I'm going to the mountains for a few days, what gear should I take?" A review where someone describes the context in which they used a product is very valuable. I think that's a good reason to engage customers and ask them for more detailed reviews, not just "it's OK, thanks", because that doesn't add much.

Maciek: Exactly. To build on what you said: the trust pyramid for content looks like this. The least trusted is what you write about yourself, that you're the best and so on. The second level is what AI recognises as paid partnerships, rankings and similar content published on external sites. And at the top is real user sentiment: genuine reviews, structured star ratings, videos, photos, unboxings, whatever AI recognises as valuable, credible information about your brand.

Tomek: So we have three main pillars. The first is user reviews we collect about our products. The second is citations from external sites and content worth citing. The third is well-organised data, which on Shopify we get out of the box.

Maciek: Exactly.

Tomek: And one important point to close this part: technically sound SEO is still a strong foundation. It's not as if the rules of the game are changing completely. What SEO gave you, especially well-built on-site SEO, is still a strong base for taking the next step and being both found and recommended by agents.

Maciek: Exactly.

What buying through an agent looks like in practice

Tomek: If we're ready for our store to be recommended, let's look at how buying through agents works in practice. Today it mostly works across the Atlantic, and customers in Europe will only start experiencing it. Polish clients who sell in the US, and we have some, can use it too.

Maciek: They already can.

Tomek: But in terms of scale, it will probably reach Poland and Europe soon. So what does it come down to, and what role does Shopify play?

Maciek: You've probably followed the whole fuss about agentic shopping in ChatGPT. OpenAI announced it would deliver the whole solution itself, with six months, an unlimited budget and top talent, and it turned out they couldn't do it. It was too much for them. And that's where Shopify rides in on a white horse, with a tool it built years ago: Shopify Checkout Kit, which you can use in a mobile app, a kiosk or anywhere else. They come in with their checkout and with a protocol developed together with Google, OpenAI's competitor, and deliver the whole solution. When a user finds a product and decides to buy it, the chat window shows the same checkout you have in your store. That's great, because Shopify puts back into brands' hands what OpenAI wanted to take away from them a bit. You keep your brand's visibility, all your business logic, all your payment methods, shipping methods, discounts. Everything you had just works. I think that's brilliant.

It also shows Shopify's philosophy. They've always stood behind brands, and they've always positioned themselves against Amazon, which is somewhat antagonistic towards brands: it doesn't promote brands, it promotes products. Same story with all the marketplaces. That's their philosophy.

Customer data, brand and order handling

Tomek: I'd add a few points, Maciek, about concerns store owners may have. Purchases that happen entirely outside their store don't mean they lose ownership of customer data. The store owner remains the owner of the customer's personal data. The fact that the purchase happens outside your store doesn't mean it's not your customer, someone you build a relationship with and keep. The second thing, which we may not have said much about today but which has just come up, is the worry that brand building stops mattering, because everything happens inside a language model stripped of branding.

Maciek: Quite the opposite.

Tomek: Quite the opposite, as you said. At some point the customer sees your brand's checkout and comes into contact with your brand, so building the brand and what sets you apart still matters and will keep mattering. To sum up this part about buying outside your store: it's another move Shopify, in my view, follows very consistently with every upgrade and new feature. It always follows where the customer is and gives you the option to sell where your customers are, whether that's social media, classic search or marketplaces. Now there's a new space where customers are showing up. They're there, it's picking up incredibly fast, we can see our own behaviour changing. Shopify lets you be there too without doing anything beyond what you do every day. You keep running your business and don't worry about technology. That's another fundamental that isn't changing.

Maciek: Yes. Amen. And one more important thing, because people may ask about it. An order placed in ChatGPT is then processed as normal: it comes into your store like any other order and you fulfil it as usual. And again you have the whole customer contact, the customer experience, which can set you apart or be your Achilles' heel, so you have to keep looking after it.

Tomek: Yes.

Maciek: I think the ordering process itself, the checkout, will look a bit different in Gemini and Copilot. Those two versions are in the works, and we'll see what they release. But orders will definitely keep flowing into your admin, and you'll be able to handle them as usual. And one fairly important thing we forgot about the Agentic plan and Shopify Catalog. Brands that aren't on Shopify but connect to it keep all their business logic, so they can have their PIM or OMS connected for order handling. It won't be very intrusive, and it will be relatively cheap, because you only pay when you make a sale through that channel.

Tomek: Yes, only for transactions.

Action plan: where to start

Tomek: So, Maciek, to finish, let's leave our viewers and listeners with a concrete action plan: what to do, in what order and where to start, so they don't miss the moment. We've said it's coming fast, it's already here in some markets, and some of our clients are already seeing it. I think fairly soon it will affect most stores, if not all of them. What three steps would you recommend?

Maciek: I think the easiest thing is to talk to an LLM about your brand and ask it questions from different stages of the customer journey: top, middle and bottom of the funnel. If you sell microphones, for example, ask which microphones it recommends for recording a podcast. Then ask about a specific brand, and finally ask for a comparison: why it thinks this brand is better than that one, and for a recommendation. Then just look at what answers LLMs give about your product and your brand. You can also ask why you're not visible somewhere, or why you are. That's the first thing.

Tomek: So a kind of audit, or rather a brand radar, checking how our brand shows up in those results? Or was your suggestion more about the products, separately from the brand?

Maciek: Both, because the point is to understand what the recommendations are based on, and to check whether your brand already shows up there, and in a good light.

Tomek: That's how I understood it: what the LLM says about the brand and what data it's based on.

Maciek: The point is to establish that. Tools like Ahrefs' Brand Radar help here. I also know Michał Sadowski from Wrocław is building a tool like this, and Krzysiek is building something similar too, though I haven't seen what it looks like inside yet. There are now quite a few tools that help you see what's happening with your brand in terms of LLM visibility. It's definitely a good idea to check which sites and domains the citations of your product come from. That's potentially a great target list: you can work with those domains and sites to prepare content about your brand and products and publish it there. That's the second thing.

The third, probably the quickest to do, is looking after your product page. Remember that LLMs handle text better, and well-structured text in particular. Besides an inspirational description, add product specifications, educational questions about the product, information about awards, certifications and use cases. On the technical side: if you have a size chart, show it as an HTML table rather than a JPEG image. If you hide text in accordions loaded with JavaScript, it's better to have it all exposed directly so bots can access it. You can check this too: there are tools that show how bots see your page and whether all the important information appears. Many e-commerce platforms and CMSs already output product data embedded in the HTML as JSON-LD, which is very useful for bots. Those are the basics I'd do first.

Tomek: I'd add the simplest step we already mentioned: a technical audit of the site, which on Shopify you can even do through Sidekick by asking whether any elements are missing. And what you said about the product page will be a very interesting challenge for many, because on the one hand we want those pages to look good, be clear and readable, and on the other we have to manage and present well all the information LLMs need. Do you think those two can coexist, and that we can do it well?

Maciek: I have no doubt at all.

A chance for smaller brands?

Tomek: One last question that just came to mind. Do you think this change, which is coming and already happening, is a chance for underdogs to break through against the big players?

Maciek: Great question. I think it's a huge opportunity for brands that move first and can get very good visibility in a relatively short time. I don't know if you've had the same experience, and this is anecdotal, but when I look for brands or products, I often get recommendations for brands I'd never heard of.

Tomek: Yes.

Maciek: And what's interesting is that when you searched in Google, you'd look through the first page, maybe the second, and get 10–20 options. Now you get three or four. So the purchase intent of someone arriving from that kind of recommendation is much higher. That explains the better conversion rates, and that's probably how I'd explain the bigger baskets too.

Is agentic commerce for every product?

Maciek: One question comes up quite often in conversations with clients and merchants: is this really for every product? We've probably talked about it more than once. I can totally see a future where a device like Alexa or Google Assistant at home, with your consent, listens to what you're talking about and hears, say, that you're running out of toothpaste. And it orders the one you always buy, from where you always buy it, and it's at your door the next day. There are products that can be bought autonomously, with your consent of course. There are products where you'll actually need AI's help to choose, like those microphones, when you don't know what to compare and you rely on a recommendation. But there are also luxury, emotional products, where you decide with your heart, your emotions, your eyes, and there it just won't work. There I'd see more potential in solutions like Gorgias, an AI chat on your own site that helps you pick a dress or put together a whole look around it. So, to answer: it isn't for everyone, but it definitely is for commodity products.

Tomek: Which is probably 80 or even 90% of the market.

Maciek: I'd use this rule of thumb: if marketplaces are your main sales channel, this is probably your next step.

Wrap-up: why act now

Tomek: To wrap up, we can borrow a thought that matches what Shopify says in its own materials. I like that they don't hide that they think it's still very early and that a lot in this area is still undiscovered, but they feel they need to be there already and help shape it. The same goes for our stores. Even if we're not seeing it yet and have plenty of doubts about whether this channel will bring good sales and whether it's for us, as we showed today with a list of simple tasks, we're not talking about a big effort. And as a brand we give ourselves a chance not to miss the moment when it arrives, and maybe to break through, get to the top and turn that into sales.

Maciek: I can even see it in the inbound traffic we get as an agency. The SEO we did for so many years certainly laid some groundwork for this. For a while now, most of our enquiries have been coming from LLMs, whereas for years, as far as inbound goes, it was Google only. And interestingly, the change happened very fast, I'd say within a few months.

Tomek: A few months, I think so too. It was probably a bit faster for us, because with services there's always a search-and-compare stage: before choosing an agency to contact, people do deep research first. But it shows that the moment this catches on in classic e-commerce can come really fast. All the more reason, I think, for us as store owners to build our own image in the world of agents, rather than letting the agents do it for us. Because they'll read us anyway.

Maciek: You're there either way. And Shopify already gives us very good tools to prepare. It's also about controlling your brand's reputation to some extent. Imagine someone gets information about your brand from LLMs, and what they say about your returns policy, delivery times, payment methods, product availability or promotions turns out to be different from reality. Think how many of our merchants get one-star reviews by ricochet because the courier didn't deliver, even though the store had nothing to do with it. You can take a hit the same way here.

Tomek: That's right. OK, Maciek, thanks a lot for the conversation.

Maciek: Thanks.

Tomek: I think we've covered the topic as far as we planned to today. I have to say, making any content around what Shopify releases is a full-time job.

Maciek: A full-time job, but it's a comedy too. Some time ago we recorded an episode about B2B, released it, and the next day Shopify made it out of date.

Tomek: Not so much out of date; they made B2B available beyond Plus.

Maciek: Yes. That's what I mean: the information we gave wasn't fully current any more, because it applied to more merchants. But Shopify has got us used to that; you have to stay alert. Maybe we'll move to live streams.

Tomek: Listen, Maciek, to TikTok.

Maciek: You got it. Thanks, see you, bye.