Agentic Commerce: How AI Is Reshaping Discovery And Trust

Great, I can start. I’m Ellen Dunne, I’m a Product Director at Shopify. I’ve been at Shopify for almost a decade. I have been working a lot in conversational commerce—years ago, this was really focused on merchants actually talking to their buyers in real life and trying to sell through those conversations. I’m really excited about this era that we have entered and this future that has arrived with AI bringing conversational commerce and fully automated conversational commerce to life. So, excited to be here to talk about this and how we can ready everyone for this future.

Hi everyone, my name is Mujtaba Khambatti. I’m sort of new to Shopify, but I’ve been in the industry for well over two decades and I’ve spent the last decade doing work on AI. I’ve seen all the failures and all of the successes lately with AI—the extreme interest with agentic capabilities that AI offers and how consumers are now rushing to do more things with AI, particularly with shopping and discovery. We’ll get into some of that, but it’s been quite fun diving into shopping with AI as the focal point over the last several months here at Shopify. I’m excited to talk to you all about some of the stuff we’re doing and seeing as we work on this together.

I’ll follow up. I’m VP of Strategy at Yotpo. Strategy at Yotpo basically has two sides. I deal with product strategy—like Tomer said, research and thinking about how AI is going to change commerce and what that means for us and where we should invest. The other side is actually strategic partnerships. One of the areas I work on is our Shopify partnership. So, this topic is close to my heart in this forum. Very excited to be here today.

Discovery in the Age of AI

To be honest, when we were preparing for this show, we did a prep talk and it was amazing. We outlined four different topics that we think are interesting from our experience with brands.

1. Discovery in the age of AI agents.

2. What is a PDP (Product Detail Page) in that kind of world?

3. Authenticity, user-generated content, and trust in a world of AI.

4. Practical strategies that brands can take today to drive their strategy in an agentic world.

There are a lot of statistics out there on how AI is impacting discovery, product search, and shopping. We did our own survey and saw 48% of consumers are now searching on LLMs (Large Language Models). I know Adobe had their own survey that the Shopify team shared showing that 39% of US consumers are already using generative AI in discovery. So, definitely, it’s massive. Even if conservatively we say in three years 10, 15, or 20% of commerce is going to move to agentic, this is probably the biggest shift. It’s like saying retail is now opening on the moon and there are a billion people on the moon. This is something that everyone needs to be prepared for.

It’s something that we picked up a few quarters ago—probably a year and a half ago—when we started to dabble into that. I’ll simply mention my own newsletter about it, Commerce GPT, where the entire purpose is to share the hours that we spent with startups, our own research, and our own data science team to share what we see working in discoverability in LLMs. So, check out Commerce GPT.

Leaving that aside, let’s talk a little bit about discovery in the age of AI and what’s called agentic commerce, where shoppers are not going to go to the storefront; they’re actually going to do the research and the shopping inside the LLM. We saw Shopify announce some really exciting stuff, from the MCP capabilities to the checkout inside ChatGPT that was announced a while ago. I’m curious for the builders here that are actually helping everyone from an ecosystem standpoint: How do you view discovery in the age of AI generally, before we go to tactics?

Maybe I’ll kick off and Ellen, you can follow in on some thoughts. What we’re seeing, of course, is an early wave. We have data from BFCM (Black Friday Cyber Monday) that happened in the past where users were converting more when they went to stores that had agents on them. We were also seeing that a large number of purchases were being made—and this is the Adobe quote that Tomer was mentioning—through some AI interactions. It’s quite common in our day-to-day that we have incorporated AI into our lives, and so have consumers.

So when it comes time for particularly considered purchases, there are a lot of actions that users are doing inside, say, ChatGPT or Perplexity to weigh options and get some research. As a consumer, you’re not having to read all these different reports and websites to compare; you get a little bit of intelligence back. We’re seeing this start and more users are doing it for more actions.

The thing, though, is if you’re a merchant, you have to ask yourself: The conversation may be happening outside of your storefront, so is it important for you to show up in that conversation in an accurate way? When users are asking “What are the best shoes to buy?”, is your brand showing up amongst what is considered by that AI as best? If the user is asking “I want to buy a luxury watch,” is your brand considered a luxury watch? Because that is an ambiguous term that something like an AI is disambiguating. But if you follow through on that discovery part, users will then make decisions based on a lot of different aspects like return policies, shipping to your destination, and very specific information. And we know that AI needs to rely on accurate information to be able to answer those questions. So, I want to pause here to motivate the fact that this is happening. We’re seeing this happen. We have evidence even from the last BFCM. But I want Ellen to be able to jump in because there’s a bit more towards making your brand participate accurately in this discovery journey.

I think there are two things that are interesting. One is just to put a finer point on what Mujtaba just shared. A really easy way to think about it is:

• Old World: Someone’s typing into Google “best running shoes.”

• New World: They’re saying “I’m training for a marathon, I need good shoes for flat feet and running on pavement.”

Does your product information deliver on those types of details? If a buyer is asking a similar question around “best brands for marathon-style running shoes,” do you have brand information or structured information that would indicate that your shop specializes in that kind of thing? That discovery and understanding of a brand is really easy to do when you land on a storefront because these sellers have spent a lot of investment making their storefront bring their brand alive. But when these conversations are happening on other surfaces, how do you translate that brand story into a ChatGPT or Perplexity? That’s what I think is really interesting, and those are some of the tools we’re trying to build to help merchants show up in that way. Even when it’s not just product information—which I think we’ll get into a lot in this conversation—but if it’s your brand and your brand story, who are you, who founded the brand, and what you really care about. We’re also at Shopify helping merchants to put that into a place that is accessible for LLMs to pull out and be able to tell that story on your behalf.

It’s almost like the store needs to get richer or more structured. Both, right? And more nuanced. Also, something you both are talking about is the difference between old search—where you get a lot of results, click in, and do your research—versus here, where so much more of the research is being done for you. It is taking you so much further into this research journey that being specific, answering the use cases, the needs, and how you differentiate yourself is so critical. It’s one of the places where Shopify is so unique to be able to power so many different DTC (Direct-to-Consumer) brands with so many specific products. It really is critical to add that kind of information and explain to the agents what your brand and your product is all about to be able to be surfaced.

Yes, and I would say the other thing that we see is that buyers who are coming out of these conversations—where they’ve actually been able to get all this information—when they eventually do land on your storefront to make the purchase, they have way higher intent and they’re way more likely to convert. They’ve already done all that research upfront inside of an AI agent. So they land on your store and it’s like, boom, they get what they need and they make the purchase. That’s the great news. Storefront traffic that comes from these agents is really high intent and really likely to convert.

Storefronts & Product Detail Pages (PDPs)

So, I have a question. One of the things we did in our customer advisory board was show our top merchants a few capabilities of how we can take their reviews and review attributes into the LLMs. One of the first things that came up is control. In the storefront, they have 100% control, and in LLMs, they have zero control. What do you recommend brands practically think about regarding control inside LLMs?

That’s an excellent point. When we see storefronts from the pre-AI era, let’s say 2-3 years ago, they were being optimized for human eyes—mobile screens, large desktop screens—making sure that it’s visual and the information is summarized because people are busy. But now you’ve got AI, and it’s got this voracious appetite to read and ingest, and it can reason over large volumes of information. So, you almost have to have the “human eye” version and the “AI detail” version so that the AI, if it’s Perplexity for instance, is getting the right information and not missing things.

We’ve been thinking through how merchant storefronts can evolve. Sometimes you can put stuff on your storefront and AI will reason over it because it has the web and it can search. But sometimes, you don’t want to clutter up your storefront. So, we’ve created an app that we announced in May at Editions called the KnowledgeBase App. It gives merchants the ability to view what we already know about your store. You install the app and there are auto-extracted FAQs that show up. You can say, “Yep, that sounds good, I have nothing more to add,” or you can look at that and say, “I do have more to add. I think it’s missing a brand story,” or “I would like to say something about how I handle tariffs,” but maybe that was sensitive and I didn’t want to talk about that on my storefront. So, you can now use the KnowledgeBase App to create this extensive version if you want to, that AI will have access to in order to answer questions about your store. Remember, AI is just going to try its very best to answer users’ questions, and the more information you provide it, the more it can then be factual and accurate.

As a former designer and having been in this world for a while, everyone remembers “above the fold” and “below the fold.” Visually, you have to have all the important information above the fold. LLMs in this new world of agentic commerce give us the opportunity to put so much “below the fold” that you would never have put back in the old days. All of that additional content, all those rich details, can now live in a way that LLMs can pull them out, summarize, search through them, and find the pieces of information that are relevant to the exact user doing the search. But, you don’t have to have a product page necessarily cluttered with all that information. Your “above the fold” is still looking great for those who are actually landing and interacting with your storefront. I like that analogy because it goes way back to designing.

So basically, you’re saying PDPs and storefronts in general are no longer just for humans; they’re also for agents. Therefore, when designing, you need to think about your storefront for both. “Above the fold/below the fold” is a great example. Are there any more ideas that brands need to think about when designing their storefront for agents?

I’ll offer one more. Remember when users come to storefronts, we sort of define the path they see. We give them maybe a popup for a 10% discount, some choice spots (enter this door for men’s, this door for women’s), and we leave it to the user to navigate. We try our best to make the visual appeal of our products show up very well. Users aren’t often now operating in the same way as they were 2-3 years ago. We know this from how you see users change their search habits. The big question is: What’s happening to Google? Is ChatGPT taking some of the queries away? The honest truth is behaviors have changed. Users are wanting more conversational ways to look things up. So, you’ve got to embrace some aspects of your storefront to be more conversational. You can have your visual elements as you had before, but having the ability to handle natural language questions where users can search or ask detailed questions is key. If I’m on a PDP and I now want to know what your return policy is—on our product roadmap, we’re going to announce a few things very soon—but I do think storefronts will evolve in conversational ways too.

Yeah, and rethinking your search bar on your storefront even. If you ask a long-winded question, is the user going to get results that make sense? Let’s hope so, because that is the behavior that we’re seeing.

And it’s such a power, right? Because now you’re able to actually get from people the questions that they want answered. It’s not just clicks and monitoring where they’re going and the funnel to conversion; they’re actually asking you and you’re learning from that data. It’s gold. Our roadmap involves taking that and turning it into a loop of collecting the right data from the users to surface and sort the right data back to customers and agents on the PDP. It is such a different kind of mentality to how e-commerce was happening on the DTC website until now. The guesswork is getting done.

You’re going to have a great flywheel of what are people asking about and how are they asking it. Being able to see that in real time and then adjust the descriptions on your product page—whether they care about where it’s manufactured, sustainability, or materials—those are all things you can dynamically update to make your products have a better chance of being discovered.

I have to say, one of the things that gets me excited is it’s a reset in user behavior. Now I can compete with Nike—I can compete with the big ones. Many small brands can have an early mover advantage versus the big ones. Trying to beat Nike in SEO is a war no one is winning. On the other hand, with the generative engine optimization (GEO), this is why I try to instill urgency. If you move quickly, you can get a head start.

If a buyer is not searching for a specific brand name, it does level the playing field quite a bit. In the world of SEO, you’re not going to get on the first page of a Nike search result, but you might get on the first page of “marathon shoes for flat feet.” It is exciting and I agree with you on the urgency.

When we look at this with queries, it plays out exactly like that. The search box versus the chat box: when you look at the queries people put in the chat box, they tend to be longer. As Ellen was saying earlier, “running shoes for flat feet for a marathon”—people will be a bit more descriptive in a chat box and more keyword-focused in a search box. So what that means is that you get the ability to show up because you can meet the intent more richly. A user looking for vegan leather versus a user looking for a specific type of thing that makes them feel special—this is a great opportunity for brands to shine without being drowned in the SEO and ad game that favored bigger spenders.

I was just going to say that any of us that use ChatGPT or Perplexity see these results. I get linked to Reddit pages or blog posts that I would have never arrived on in classic SEO. Brands have such an opportunity to identify this new evolving mechanism. A big part of discovery is reputation management—making sure that you surface not only on your PDP, but on these new channels like Reddit or YouTube expert content. It’s really changing the game.

And I also think ads will arrive there someday. Until the ads arrive—because when ads arrive, prices go up—brands have to take advantage. This is the biggest shift we’ve seen. Even if 10-15% of e-commerce moves to agentic commerce, that’s massive. First movers will have a lot to gain.

Authenticity and Trust

Great segue. We have a saying inside Yotpo that no user reads review number 1,027. But you know who does read it? The LLM. The agent. Therefore, quality and authenticity are becoming much more important. We invested heavily to extract attributes very accurately from the reviews. If a specific product had two or three users say it’s “great for psoriasis,” no one would have ever seen that, but now the agent knows about it. Context really matters. I can ask Ellen, “Hey, this is a great necklace, where did you buy it?” and you can share the story. The agent doesn’t know that. A lot of the PDP is about context, and we are big believers that from the reviews you can extract metadata that gives much more context.

So what’s your thought on authenticity and trust in a world of agentic commerce? Because with AI, everything can be fake. I don’t know if we are actually talking or if it’s your avatar talking to me.

I want to be real. We are all excited about AI, but it is new. As it is new, we’re discovering guardrails and learning things. It does create some anxiety in terms of being authentic and making sure you’re well-represented. Not everything has been solved. Hallucination has and will continue to be a thing that we have to tackle.

Can you explain for everyone what you mean by that?

So if users are asking an AI what your return policy is, the AI is biased towards trying to answer your question. If you provide an answer, it will use that. But if you don’t provide an answer, it may try to invent an answer based on the fact that perhaps 30-day return windows are common. Such promises then become made-up promises by the AI and the user might believe them. AI tackles this by providing sources and links. But what merchants can do is make sure there are adequate FAQs—perhaps using the KnowledgeBase App or putting them directly on the storefront—because you are preparing your storefront to be read by AI. You can take the horse to water but can’t make it drink; the tool is somewhat imprecise. AI is going to sometimes hallucinate, and we are still working through guardrails.

This is why we try to help generate as many reviews and have specific questions and prompts. We actually have an AI assistant that helps users write content exactly for the reason you mentioned. A very savvy buyer, especially for a considered purchase, isn’t just going to look at the content on the store. They really care about reviews. “I have dry skin and I want to buy this product”—I don’t care about any other review except the ones from people who say “I have dry skin and here’s what it did for me.”

Then there’s another body of evidence out there that the AI agents are good at summarizing, which is Reddit and other sites like that. It’s that triangulation. You have some control over collecting reviews and having good customer service, because if people have a bad experience, that’s going to show up on Reddit. You need to be aware of where people are talking about your brand. So it’s those three bodies of content:

1. Your own site and everything rich you can include there.

2. Product reviews, because they’re key to purchasing decisions.

3. The other stuff out there where people are habituated to look and trust.

Strategy and Actionable Tips

Let’s talk about brand strategies. We talked about shifting the storefront, the importance of reviews, and conversational elements. Besides buying Reddit stocks, are there other ways to drive strategy?

I want to come back to one aspect. We left the last segment with “AI can hallucinate, be prepared.” A lot of the talk tends to be worrisome, but it does grow the pot. The ability for us to connect more users that have an intent to buy with products they want is greater. We’ve always had shopping lists in our minds—”If I had time, I would go buy the right desk lamp”—but we never find the time. AI unlocks this. With BFCM, numbers are indicative that products are being matched with greater conversion to users that have intent. I think the ability to frame your brand to embrace that allows you to stand out. When people are choosing a particular style, like vegan leather, the ability to describe the shoes or bags you make very specifically elicits the right reviews and shapes the FAQs, allowing you to stand out.

Just doubling down on the brand story thing. You mentioned my necklace—I could tell you the artist who made it and all of that. Probably those details are not on her website, but it is the reason I continue to buy products from this person. This is where Shopify is really wanting to represent DTC. These merchants are entrepreneurs building a product they’re really passionate about. That is a unique thing that consumers care about. If you haven’t thought about bringing that to life on your website or through your customer service, this is a really good time to do that. I just can’t overemphasize the brand story. It’s an interesting way to distinguish yourself in this market.

If I can tell a story: At Editions, we announced KnowledgeBase as well as Storefront MCP. We didn’t talk a lot about Storefront MCP, but it allows you to create a storefront conversational agent. There is a brand in India—weddings in India are very colorful, multi-day events. This brand built a Storefront MCP chat at the bottom of the screen. The merchant told us that a lot of people feel that personal touch is needed; it’s an emotional time with tons of decisions. Having that personal touch through the chat added a human element. People feel like, “I’m talking to ‘Auntie,’ somebody who is helping me plan what to wear,” versus just a store where I’m left on my own.

Very interesting. That’s a great example.

I think we covered so many different topics. To wrap it up, do you have any small tips for an entrepreneur or an executive in a large brand to carry their brand forward in an agentic-first world?

Just to follow up on the storefront agent example: having a storefront agent has become important. Buyers are expecting to continue the conversation when they get to your storefront. I want to make it clear how easy it is to train your own storefront agent to speak the way you want your brand to be spoken about—using your vocabulary and lexicon. It’s very similar to hiring a sales assistant in a retail store; you train them on how to talk about products and sizing. You can do the same with a storefront agent where you control the voice and tone. It’s a fun thing to play around with. While we can’t completely translate that to how your brand gets spoken about in Perplexity, you can still create “ground truth” with the KnowledgeBase that the LLM will use.

I’ll say three things:

1. Consider being on Shopify. We have our merchants show up on Perplexity and we’re making sure our catalog shows up there.

2. Consider a storefront agent. It’s really easy to set up Storefront MCP.

3. Use the KnowledgeBase App to view all the FAQs associated with your store that AI will be able to see, so you can get more prepared.

I could maybe add two to that list:

1. Treat AI/LLM search like SEO in the early days. It’s trial and error. There is no Google Analytics yet. A great way to do that is to have conversations on your website and see what questions people are asking. Also, look at your reviews and what people are talking about.

2. Focus on sharper positioning of the brand. Agents can be so much more scientific and specific. Make sure you touch on all those criteria and specs in your products for different types of users—not just the main PDP. That long tail of information is a major opportunity.

For me, I also wanted to say: Use Shopify, use Yotpo. Definitely stay curious and go read more. This is why I did my newsletter on Commerce GPT. You have to experiment because this is a reset of the playground.

Thank you if you stayed with us. It’s all about helping you move to an AI-first world. Thank you so much to the Shopify team. You’ve been great. Thank you so much everyone. Bye-bye.