[00:00:00] Welcome to Unpacking the Digital Shelf, where industry leaders share insights, strategies, and stories to help brands win in the ever-changing world of commerce. Hey everyone, Peter Crosby here from the Digital Shelf Institute.
[00:00:28] As the squished funnel of the AI shopping conversation continues to proliferate, we are seeing strong data that indicates conversion rates are higher than traditional search-based journeys. And with studies showing that 60% of consumers blame the brand itself, not the site or retailer they got the information on, the stakes for getting answer engine optimization right are rising.
[00:00:52] So Jing Feng, Co-founder and COO at AI marketing platform Bluefish, joined us to share best practices and the cross-organizational orchestration required to win the recommendation. Welcome to the podcast, Jing. We are so delighted to have you on. Thank you for making the time. Absolutely. Hi to both of you. Great to see you again. Thank you for inviting me.
[00:01:14] Of course. You and your team sit in agentic commerce every day, sort of in the flow of it, the stream of it. And you do a lot of time to try to understand, take a lot of time to try to understand the outcomes and the impact it is having for brands, which is a really popular and pressing question these days, especially as people make their 2027, what the heck are we going to do about this plan?
[00:01:41] So there's a lot of talk about all this agentic, how AI is affecting the consumer journey. So, you know, just give us your take on everything. Great. Yeah. You know, all of that stuff and what really matters right now. What should folks be paying attention to? Yeah, definitely. I think you called that out, Peter. A lot is happening right all around us.
[00:02:07] I think to us as marketers, a lot of marketers are experiencing this intense pressure. I think the primary context and given the limited amount of time we have, I think the biggest key thing is just that the consumer journey is collapsing and it's collapsing into effectively a single interface.
[00:02:28] And I'll go into what that means, but essentially, you know, discovery, comparison, and increasingly even the purchase itself is happening inside that AI conversation. And that's before the shopper even reaches your brand site, right, or a digital shelf or, you know, et cetera. And then on the other main category, they refer to it as like taste-led categories. It's basically, you know, consumers who already know their preferred product or brand.
[00:02:56] They're saying that, you know, AI traffic, AI referred traffic is bringing 1.3x, the number of net new customers relative to search. That's a 30% increase, right, on discoverability just based on the channel alone, right? So, and this kind of leads back to what I was saying is that AI is the full funnel. This is a really important takeaway for brands.
[00:03:24] It's as much of a brand exercise as it is a performance marketing exercise, right? And there's a really big mechanical shift that's underpinning all of this, meaning, you know, historically, as marketers, we are used to thinking about these different channels, different touchpoints all along a consumer journey.
[00:03:46] But in AI or, you know, now with AI, I should say all of these individual channels and touchpoints are training the models on how to behave and who to recommend essentially, right? So it means that models, not just retailers or shoppers are now deciding what products, right, appear in what order, at what price, et cetera.
[00:04:11] So I think this is probably the biggest thing is this mindset shift that needs to happen, that AI is not a new tactic or it's not just a new channel. It's a whole new audience, right, that's going to make decisions on behalf of your consumers. It's your growth lever. You have to treat it that way.
[00:04:37] And because there are consequences if you don't, right, given the growth and the brands who are building their orgs and who are orchestrating their marketing strategy around this shift, right? These are the brands that are already pulling ahead, but these are the folks who are going to retain advantage. So is that data? Yeah, yeah, absolutely.
[00:04:58] And it makes me think, you know, everyone's trying to figure out whether this activity can go under the incremental growth column. And so when you talk about those Shopify stats in my sort of somewhat blurry understanding, is that the fact that those conversion rates are going up at the rate that they are and it's new customers,
[00:05:23] assuming that proves out across platforms and over time, that's considered incremental. Would you describe it as that? Like versus just somebody switching the channel or the way in which they bought? What do you say, what do you and your customers talk about when you think about sort of how to justify this investment? A hundred percent.
[00:05:46] I think, and I think that's the real question, right, Peter, which is like, how do you, from a CMO level and a budget perspective, right, how do you allocate funds? I think the reality, and this is part of the challenge, is that it is both incremental and basically reshaping, right, all of the existing tactics and budgets.
[00:06:10] So that means that brands and marketers are being put in this uncomfortable position where they have to invest and they have to justify the incrementality, but they also have to understand that if you don't, it is, you know, other competitors are basically taking a bite, right, out of your existing market share.
[00:06:34] And they will, and in significantly compounding ways, right, because of how fast AI is developing. So I think that's why it's uncomfortable right now, because you have to find that extra budget to invest. You can't necessarily say it's incremental yet. I think everybody is, you know, looking at different data sets to prove this out. But, you know, and there is this mystery of the disappearing traffic, right?
[00:07:02] Because a lot of traffic has disappeared going direct to brand sites, but the referrals aren't necessarily there. But that's because people are learning, right, outside of the ecosystem that you control. So now you really have to think about, well, what does my brand.com even do, right? Is it still education if folks have, you know, already learned the majority of what they want to learn?
[00:07:29] So what do I want this part of the journey to now look like? And how do I optimize for that? And so the reality is that there probably needs to be a investment across the board to reshape how every channel is thinking about AI and is optimizing, right? And thinking about AI as its own channel. But it's not always clean cut, right? Especially if you're coming from a CEO's office, right? Of how should I invest? What's the ROI?
[00:08:00] Because the reality is those numbers don't exist anymore because the chain has been broken, right? And so that's the kind of uncomfortable place that we're in. Reminds me of the early days of e-com when we were like, we know it's important. Please give us resources and budget. We're social. Yeah. Yeah, exactly. And social is still going through its own identity. Yes, manifestation. Yeah, I thought that's a better way of putting it, yes.
[00:08:25] But in the same way that most brands who are investing in TikTok are spending money, are not necessarily making money, but are seeing the effect on Amazon or on other channels. I think that's really the same for AI. It takes money to invest in looking at AEO optimization and creating content and things like that. And you'll see the effect. It just might not be a 1 plus 1 equals 2 type of situation. And that's hard to explain. Yes. But, you know, everybody's seen this chart, right?
[00:08:52] Like the adoption in social, mobile. We thought that moved fast, right? This is moving. I think the chart is like at 3 or 4x the pace. So we're having all of our brand customers, right? Retailers are having to make that decision now in a much more condensed timeline. And that's very uncomfortable. But the impact, I should also say, right?
[00:09:21] The inverse of that is the impact is also really, really condensed, right? And fast. And I think people should not forget about that, right? Because if you're an established brand and, you know, and you have a bunch of challenger brands around you, now's the time for them to swarm, right? And vice versa. So this is really critical to think about. Yeah. Yeah.
[00:09:46] And I like how you talked about how it's a part of your strategy and you have to kind of think about it holistically. There's a lot of conversations around, oh, well, is SEO dead? Or like, how does AEO mix with SEO and what does that look like? And I see them as both distinct, important features of your overall strategy. But how do you talk about that with brands? Like, are they competing? Are they working together? How do you kind of chat through that? A hundred percent.
[00:10:15] It's definitely working together, right? I think most folks understand that. Peter, I'll call you out. If you and I both, you know, put in the exact same prompt for good skincare, right? We're practically the same person, James. Yeah, exactly. We can be pretty sure that we're going to get very different recommendations. I believe that is true. Right? Right.
[00:10:41] So from a data and data integrity standpoint, it means as a brand for the same prompt, you might show up 80% of the time for me, but literally 0% of the time for Peter. Right? So you have to make sure that you're optimizing for intent, that you're optimizing for the way that AI and AEO works.
[00:11:01] Because if you're just transferring that playbook and you're just focused on SEO tactics and just focus on the prompts, then it means that you're essentially getting bad data, right? And then you're going to make poorer decisions and you're going to get, you know, worse ROI, right? So there are really important implications to treat them as distinct things, right? Where having both is a necessary part, right, of a good marketing strategy.
[00:11:30] Yeah, we would call that the context layer of data that you need. That the use cases for whom, at what time, what's happening, what's the weather. Any of that stuff has to be available to the AI to be able to tap into, depending on what it knows about you and the context of you. They have to have the context of the data to then serve that up. And that's a big task. It is.
[00:11:59] Putting that together is a lot of work and needs to be automated, right? A hundred percent. Yeah, and we refer to them as like hidden contexts, right? Because it's below the surface. It's, you know, the interaction that I have at this point with my AI, right? My new BFF is I can ask a simple question and then it's essentially asking me, oh, but what about this? I remember that, you know, you like X, Y, and Z.
[00:12:25] I remember that you live in a state that's high altitude, has like a lot of sun exposure. Therefore, we recommended this, right? So the hidden context really comes out, right? In that conversation, you really want to make sure that your products are optimized for those hidden contexts.
[00:12:44] So, you know, when there is incorrect information about a brand online, a survey from a company called Rhythm, R-I-T-H-U-M, found that 60% of consumers blame the brand itself, not the site or retailer they got the information on. Or the AI, yeah. Or the AI, yeah. And so that's super challenging because, you know, you were talking about it earlier. The brand has just less control, less sort of exact control.
[00:13:14] And also the volume of what's required to win that recommendation is so much higher. And how do you suggest that brands handle that in a sort of lower control environment? Yeah. Well, first I was going to say, isn't that a super enlightening statistic, right? Enlightening, horrible. Yes. Bad, scary. We've got a lot of adjectives. Yeah, totally. Yeah.
[00:13:39] Well, I think the same study literally also said that 90% of those consumers abandoned the cart after blaming the brand. So not a super forgiving, like, environment, right? Which is really, has really important implications. So maybe I'll back up and just say, you know, it's interesting, like, that we talk about visibility in AI a lot, right? I think that's a starting point.
[00:14:08] But once you layer something like this data in, you can immediately see that greater visibility can also equal greater brand risk if the information is incorrect, right? Now, this is really critical for marketers to understand because a lot of, depends on the vertical, right?
[00:14:29] But a lot of marketers don't have this risk framework, right, or safety framework that they really, like, built out yet, right? So this is an important area for marketers to seriously think about, especially when we relay this back to, relate this back to what we talked about earlier, about it being full funnel.
[00:14:52] So if you're not being included, right, or discovered because of incorrect assumptions, right, in the upper funnel, then you're not going to show up, right, in the lower funnel. So, but back to your question around control, because it is a scary citation. So the thing that I would say is that you actually have more control than you think as a brand.
[00:15:20] And this is actually by us looking at inaccuracies at scale, right? And what we've found is that across verticals, across brands, about 30%, it, you know, goes up and down based depending on the vertical, but about 30% of those inaccuracies can actually be traced back to the brand's own pages. Right?
[00:15:48] Meaning a meaningful share is actually within the brand's control. So it is their fault. It's just at the site. Yeah. It's just at the site of film. Not to point fingers. Right. So not everything is just wild hallucinations. And they do exist. They 100% do, right? But there is some control that you can exert, right?
[00:16:10] We also found, you know, like 64% of those AI errors are due to incorrect numbers. So this was something that surfaced quite a lot. So it could be a dosage. It could be a price. It could be a rate. Which, again, has super important implications, right? Right. If the consumer is never leaving that conversation and making a decision based off of everything that's showing up there. Right.
[00:16:38] So, but again, you know, you can ultimately, if you can trace, right, where the incorrect information is being generated from, then you can do something about it. Right. And that is why, from a Bluefish standpoint, we're so focused, not on just if something is correct or incorrect, but on providing that traceability back to our customers.
[00:17:05] Because that's where the action can actually come in, right? So that you can go and figure out which of my own first party pages are, you know, providing the incorrect information, which third party pages are, you know, just, you know, out there wilding out. Right. And of course, there's a percentage that's going to be just hallucinations that, you know, is maybe out of my control today, but hopefully in the future. Right.
[00:17:32] This is something that I can, you know, directly influence with AIs as well. As someone who used to work with an OTC brand, the comment you just made around dosage just made my like antennas pop up. Because like, even if you as an over-the-counter brand don't want to engage with LLMs, you need to see what they're saying.
[00:17:55] Because if someone is, to the stat, like if the consumer's blaming the brand versus the LLM and has incorrect dosing information, I mean, we're getting into territory where you should not take medical advice from an LLM. But again, it is like the number two use case of LLMs, I think, from data. Yeah. Yeah. For sure. You still need to, I guess the point I'm trying to make is you still need to know what LLMs are saying about your brand, even if you are not doing anything about it. Yes.
[00:18:22] And especially around these very sensitive gray areas, right? Around risk and safety. A hundred percent. Yeah. I'm not sure that the risk team at their company would say, even if you don't do anything about it, it would be unacceptable. I mean, once you know, you kind of have to do something about it. You would hope so. Yeah, I know. We live in a very difficult world. So key takeaway for the audience is don't just listen to LLMs when you're, you know, 13 for dosage. That's one of the takes. You heard it here.
[00:18:53] Maybe countless lives, Peter. That's what we do here at the DSI, save lives. But Jing, do you have any examples of any brand experiments around agentic search that have worked, maybe haven't worked? Well, our audience always loves to hear those fun examples. No, examples are important, right? Right. Because I do think that there's a lot of talk around AEO.
[00:19:20] But then very quickly, you know, the question, especially as it relates to budgets, right? And investments is where is the attribution, right? Does it work? So I can walk you through an example of a global beauty brand, for example, that we work with. So this is a beauty brand that sells within retailers, right? On Amazon. They have their own e-com site.
[00:19:48] So really kind of across the board. They have a few brands, right? That are within kind of like a similar division or, you know, product line. But they really wanted to, of course, they started out, same as everybody else, by just wanting to understand discoverability across the board.
[00:20:09] But very quickly, and I'll kind of walk you guys through what happened, they were able to pinpoint some really critical aspects that then influenced how LLMs were presenting their brand, especially on Amazon. And that actually resulted in really significant increases in actual sales, right? And ROI.
[00:20:37] So essentially, you know, they're a Bluefish customer, of course. They started with just kind of basic audit of and benchmarking, right? Of what's happening across the different LLM surfaces in the platform. They could then see that a number of their PDPs, especially on Amazon, were lagging competitors, right? And as a result, they were just not surfaced, right, as highly within Rufus responses.
[00:21:07] And so then using a lot of the diagnosis data, right, and analytics in our platform, they were able to audit those PDPs and then just pinpoint the very specific narrative gaps. Right. That they were not strong enough on relative to competitors.
[00:21:28] And a lot of this was around, you know, it was a variety of like medical slash kind of dermatological claims, essentially, right? And so we worked with them, right, to figure out what's the right way to craft that narrative to really showcase the right representation of their products.
[00:21:52] And I want to be super clear that this is not about gaming or abusing the models. I think this is really important, right? Because hopefully, you know, you guys heard and everybody else also heard at Google Marketing Live recently. They were really clear in saying, look, if you're going to go out and use AI to create a thousand pages on certain topics, we're going to market a spam. And if you don't stop doing it, we're going to punish you for it. You don't want to be doing that.
[00:22:22] What we're really trying to do is understand how LLMs are extracting information. So sort of going to that idea of gaming the system, which became, I don't know whether it's fair to say it became easy to do in SEO land. But certainly a lot of people made a lot of money being the gamers of that system. Do you feel like when you talk about don't throw up thousands of pages of content,
[00:22:50] are you talking about then your, I'd just like to dig deeper on that. Like, should your investment in adding that context layer and adding those things that will get you into more considerations and up your recommendation, are those better off happening on sort of your core site rather than a lot of new stuff spinning up? Or what are your takeaways from that kind of warning from Google Live? Yeah, that's a great question.
[00:23:20] I think it's two things. I think one is just that it should be about quality, not quantity, right? Is part of what, you know, that warning is implying. Then the second piece of it, I think it's also important, which is like, where do you deploy? The AI agents do work differently, right? To humans. It's very practical in many ways.
[00:23:46] And in trying to be helpful, it is trying to understand all of your circumstances. And so what is helpful in messaging to this new audience is going to be different to what is helpful in messaging directly to a human. It's less emotional. It's got to be use case driven is often what we see, right? You have to be specific.
[00:24:11] It's not about, you know, to this, to my AI, right? It's not about how the, you know, the tinted, you know, skin care or tinted sunscreen is going to make it easier to get ready in the morning, right? That's more of an emotional message. It's that it is SPF 50. The tint does not, you know, appear artificial. It doesn't rub off on clothing.
[00:24:41] What we see in terms of successful content is that it's very use case driven, right? And very, very specific. That's where the performance that you're looking for is really going to come out. You still need to market directly to consumers, right? You still need to have that emotional messaging and to have in order to, you know, retain that brand loyalty and to connect with the shoppers.
[00:25:11] But it does have to be both. And it is going to be really different. And so that's the quality that I'm referring to. Once you've determined, you know, what it is that you need to say, right? Which audiences, what particular use cases are really important to these audiences, right? What are the things that I really need to get out there for the AIs to understand? Then it's a matter of authority, right?
[00:25:40] Because at the end of the day, AIs are trying to parse through the entire internet. And just like a human, they're parsing out, you know, maybe the really outspoken people on Reddit. Yeah. You know, and they're parsing out the slop, which is literally the warning, right? Yeah. And they're saying, if you can actually be helpful, if it's authoritative. And many people agree with you because you're also trying to validate.
[00:26:10] Then, you know, we will make this recommendation. And so that validation is super important, is all part of that mix to establish authority. And that's not necessarily on brand.com. And that kind of brings me to my last question, which is, who are the people at the table for all of this work? Because it is different. It is broader than I think.
[00:26:38] You know, I don't think, maybe this isn't entirely true as I start to say it, but PR wasn't heavily involved in SEL. Like, they might dip in and say, are there some terms you want me to go after or something like that? But it feels like this is much more of a sort of all hands on deck kind of thing. And I'm wondering, in terms of your product, who are the users? And then who are the consumers that may not be actual hands on in Bluefish, but are, you know,
[00:27:07] or maybe they are in the way that they're able to get data, but maybe not do a lot of it. I'd love to know sort of who the players are in all this work and how close do they get to your product? Yeah, no, it's a great question. The short answer is, you know, there's no one team that should be the only users, right? I was just at the, you know, MMA and we were talking about this, which is really just orchestration
[00:27:36] across marketing teams now that we're in this new paradigm where every channel affects the AI channel, right? So in terms of users, I would say, you know, it's still, we're still in relatively early innings of this space. So we have, you know, a lot of SEO folks who, you know, originally were handed this mandate.
[00:28:02] So this orchestration is super critical for CMOs to really think about and how to structure their teams and really enable their teams to be cross-functional, to be successful in this space. And what's your, because I keep thinking about the actionability of the data that you or anybody else who's in this, the recommendation.
[00:28:29] So you do an audit, you look for those opportunities, those become recommendations, you know, I imagine. And then the recommendations need to be probably prioritized. I don't know. I mean, I'm wondering. Yeah. Because orchestration is about getting the recommendations into action. And when you talk about that number of teams, how do you, and who probably already have some flow of workflow or task management or something like, how does all of that today work?
[00:28:59] And what is your ultimate vision for how that happens at machine speed potentially? Yeah, absolutely. And I think, I think a few things here. One is just that the, you know, every, we work with Fortune 500. And we've seen a couple of models though starting to emerge. And one is a more centralized model where, you know, you have one team that not only has the knowledge
[00:29:27] and the time or the mandate, but they also have the resources, right, to help execute. So we've seen this emerge as like one kind of model. Another model is more of a distributed model, right? Where each brand has their own resources, their own ability to execute their own ways of working. But there's a centralized kind of governance committee, right? Or something like that.
[00:29:55] That's just looking across everything from a governance standpoint and less execution. What I would say is that whichever path is the closest and most achievable for you as a brand is the path to take. The most important thing is to put one foot in front of the other and make moves here. Because this space is shifting so quickly.
[00:30:17] And orchestration, I think, is going to be a big determination, right, of whether brands win or lose just because the space is moving faster than we've ever experienced it before. And so, and of course, from like a scale standpoint, Peter, you called this out. That's important. And like, that's why from a platform standpoint, we provide those orchestration tools so that you can collaborate together, right?
[00:30:47] And measure together and take action together in one place. But from an org standpoint, you know, messaging to CMOs is plan for that today, right? Like how you're going to have that governance and execution, whatever version of that looks like. That's what I'm also seeing from the org side of things. And I keep, I know I keep using e-commerce as like a comparison, but it's very similar to the days of like the early COE where it's like, hey, this is important.
[00:31:16] Let's put some people on it. Okay, great. Now we know it's important. We got everybody's buy-in. Let's democratize it back into the business. So I think we're going to see a lot of that with AI where like, if you have a chief AI officer or if you have someone who's focused on AI, that's not a role that may necessarily continue for 10, 15 years. It's like, hey, let's establish this is important. Let's put dedicated time to it. Let's educate. Learn the motions. Yeah, exactly.
[00:31:43] And then it becomes part of the business because you shouldn't be thinking about AI as a separate channel. So I encourage people to just look at the correlations between what you've already been through and what's worked and just kind of apply those to this new but faster change that we're seeing. Yeah, 100%. We've been through it before. Like, we're resilient, right? Like, as marketers.
[00:32:08] So it's all about, you know, just starting down that path and taking the lessons, like you said, Lauren, right? That we've already learned. So before we completely let you go, I just want to let our listeners know that bluefishai.com is the place to go to get really rich content and advice around all of these things that we're talking about from technology to orchestration to organization. And it's a great resource. And so I'd recommend folks do that.
[00:32:35] And so again, Jing, thank you so much for bringing all of that to our attention. Thank you, guys. Thank you so much, Jing. Thank you. Thank you.