The rise of the AI shelf

0:09Deana Burke:Hello and welcome to the Modern Retail Podcast, our show that covers the ways the retail industry is changing and modernizing. I'm your host, senior reporter Gabby Barco. What happens when you ask ChatGPT, Claude or Gemini for a product recommendation? more and more consumers are turning to AI search to decide which natural deodorant to apply or what high protein snack to try. You could almost imagine this personalized virtual store with all these items sitting on the shelf. Well, the AI shelf. The emergence of this new shelf has brought chaos to product discovery. As recently outlined by tech newsletter writer Deanna Burke from Boys Club.

0:56Deana Burke:She decided to test AI bots by creating a completely fictional natural deodorant brand to see if she could get it recommended on the AI shelf. And the experiment actually worked. As product discovery evolves from the physical shelf to the digital shelf and now to the AI shelf, there are a few question marks. Like, how do these algorithms decide which products to surface? And if you're an emerging brand, how do you ensure your products take space on that AI shelf? Joining us to unpack all of this is Jessica Wright, Senior VP of Product at Data Platform Spins Foundry. Jessica, welcome to the show.

1:41Deana Burke:Thank you for having me. Yeah, so I'm really excited to chat today because I know you have been tracking this relatively new phenomena pretty closely in the last year or so. And so I'd love to start out by maybe the bigger question and sort of how it fits into your day-to-day data collecting and analyzing is, you know, what is the concept of the AI shelves and why are brands kind of freaking out about being surfaced on it? Yeah. For most of modern commerce, there was, you know, brands competed on these two shelves and there was a playbook there around how to execute and how to win. You had your physical shelf where, you know, it was all about distribution, placement, visibility in the store, which is still really important.

2:36Then you have your digital shelf where brands were really focused on ranking and search, product pages that were beautiful for humans, looking at things like ratings, reviews, really online merchandising. And we are seeing this emergence of the third shelf or the AI shelf. And what the AI shelf really is, is this subset set of products that an AI assistant can understand and recommend to a user when a consumer is asking a question about a product or a specific shopping event on platforms like ChatGPT or Perplexity or even in Google with Gemini. So what's interesting is that a lot of this shift is due to the fact that AI can now help shoppers fully do their research.

3:32It will compare, it will narrow like choices for you and really do the filtering on behalf of that consumer who's running that initial search before a human ever even makes it to like a retailer or a brand website. So while the consumer is ultimately making the final decision around whether or not, you know, that short list of recommendations that this AI assistant surfaced is the right product for them, these AI assistants are ultimately shaping what options or what products ultimately enter consideration and get returned to that consumer. So for brands, they want to understand, are they legible to these AI systems so that they're accurately understood to be part of that short list of recommendations.

4:27And I also think it introduces a really interesting landscape for larger brands or even emerging brands to compete right alongside national brands in being visible on this third shelf, because the evidence that is used to determine whether or not a product should be recommended is very different than the strategy that was employed for the physical shelf and the digital shelf. Yeah.

5:00Deana Burke:And we'll get into more concrete examples in a minute, but this is actually a good time to maybe talk a little bit about Spins Foundry. And actually, historically, you guys have tracked shelf data, right? On shelf data in the physical stores. And then, of course, e-com and now this. And so, yeah, I'd love to just hear a little bit about what goes into the data collection and maybe how it differs or maybe how it's similar to the other ways that you're doing it on the other channels. Yeah, so Spins has, you know, over 30 years of experience tracking what people are purchasing, where they're purchasing those items, why they're purchasing those items.

5:44And this vast amount of market intelligence around sales performance, like why a product performs well in the market, we use as demand signals to understand what are the moments in which a product should be discoverable when someone is searching for that item. So if you think of the sales data as demand signal, we understand that there's demand for a certain type of product because it has a very specific attribute or it has a very specific price point. So we can see that shoppers are buying more of that item. We use that to inform the specific moments that brands should really take into consideration and make sure that they're discoverable around on this new third shelf.

6:35if people are buying more products that have, let's use snack bars as an example, but if people are buying more snack bars that have a specific ingredient or are gluten-free, and I have a brand that is gluten-free, I also want to make sure that I'm discoverable across these AI platforms where people are starting to do their discovery because we see three and four shoppers are using AI to do that discovery. And AI is influencing a ton of the decision making before someone even goes into the store and purchases that item. Yeah.

7:12Deana Burke:And with the focus on CPG that Spence has, I'm kind of curious, what are some of the popular top categories you're tracking on this third shelf right now? I imagine it's some of the usual suspects, but yeah, curious, what are the hot, buzzy ones. So, I mean, we track all of them. But what is really interesting is when we look at categories that have a ton of innovation and differentiation, because there are certain categories in the market where, you know, price is really the determining factor around whether or not someone's going to purchase that item. They're kind of, you know, commoditized categories raisin thinking, toilet paper, and things of that nature.

8:00And so these AI assistants will index on price and look at price and surface the item, you know, that is sort of the best deal for the shopper who is, or the consumer that is running a search on ChatGPT. But when it comes to more of the categories where there's a ton of differentiation or consumer preferences, so like vitamins and supplements or even certain snacks, snack bars that have very nuanced characteristics or meet very specific needs states, whether that is a specific health outcome that someone is trying to achieve or a sustainability goal. Like that is actually where how these AI assistants understand and interpret products becomes really interesting for us to analyze because a lot of what is needed there is the AI assistants need the evidence to determine whether or not it can align your product to the specific intent and consumer preference that that consumer has when they're asking a shopping question in one of these AI platforms.

9:10Deana Burke:Right. And I imagine a lot of that is determined by, you know, what we've been hearing is that it's not what the brand or the product says about themselves. It's pretty much being scanned from everything from Reddit to YouTube reviews to creators. So is your sourcing coming from that versus it's very easy for a brand to just ask JATGPT about themselves and have it tell them what they want versus you want to hear, you know, whether the consumer is actually reaching them? Yes, I think that's the monitoring aspect that's really important around how we track discoverability. So when anyone runs a search against any of these AI platforms, products will get recommended.

9:59And there's typically a URL, like a website that comes along with that recommendation. We refer to that as a citation. And that can be from your brand.com website. So a surface where you can actually influence the content and information that these AI assistants are actually interacting with. It could be a retailer's e-com platform where you can often syndicate, distribute information to those various surfaces and influence and make sure basically your brand narrative is expressed across those surfaces. And then, yeah, there's all of these other third-party platforms, blog posts or platforms like Reddit, etc., that might also be sources that these AI agents are using to understand your products.

10:58And so as a brand, having visibility into what is sort of the landscape of information that these AI assistants are accessing to understand your products actually helps a brand make informed decisions around what their marketing strategy should be, what their positioning should be, and where they might have inaccurate information, stale information. or gaps that are limiting their ability to show up for the right moments, right questions when someone's asking an AI assistant about their products.

11:43Deana Burke:Okay, so I guess let's say someone were to ask Chad GPT or Jeveni broadly, what is the best coffee brand? What is on the AI shelf, I guess, in that coffee section? And would it change whether I say, of course, you know, from XYZ region or from a sustainable brand or single origin, et cetera, et cetera. Given that coffee, I feel like is such a great example of a commoditized product where brands are trying to reinvent, you know, that category. And so, yeah, I'm kind of curious what you see in these really buzzy categories right now. Yeah. So using that specific example, if someone were to, if I were to put that, you know, what is the best coffee product or coffee brand in ChatGPT, and then you were going to do that, it is very likely we might get the same answers.

12:42They might also be different because it depends if I'm logged in. And so my chat GPT already knows that I actually lean towards more sustainable products. So there's some preference built in. But a broad category search like that is going to look across all websites. It could be the Reddits of the world, typically like best of. you're looking for public consensus. However, brands are trying to differentiate themselves, like you mentioned, and use certifications or different label claims to stand out and align with preferences that their customer base has. So let's use an example. I'm a brand that is a coffee brand and I am, I have, you know, I've invested in marketing copy on my brand website that articulates that we source our beans from a specific local farm, local farm in a very specific region of the world.

13:55And like that narrative could be something that connects with a human who is trying to identify a product that might align with their particular preferences around sustainability. But an AI agent doesn't necessarily, can't necessarily interpret what that actually means. So we've worked with brands to take their marketing narrative and their brand ethos and structure their product information in a way where a LLM, an AI assistant essentially, can directly use that information as evidence to align a product to a shopper's search looking for more sustainable coffee beans. So how we do that is, it is likely that this brand, my brand, let's say my brand has Fairtrade, It's fair trade certified.

14:57It is organic certified. We use very specific beans. We take all of that information about your product. And the first step is making sure this information is structured in a way where we are reflecting that this product is certified fair trade by said certifying organization. So there's trustworthy, authoritative information that this LLM can understand and associate this fair trade certification with this intent to purchase a sustainable coffee brand. We'll also make sure that the organic certification is structured in the product data that these LLMs are crawling and can be leveraged to connect this search around a more sustainable coffee to evidence of sustainability.

15:57You have an organic certification. If a consumer also puts into that search something very specific about the type of coffee bean, if you don't have reference to that coffee bean and roast level or other attributes specifically related to coffee structured for an agent, like a machine basically to read, you won't be eligible to be surfaced in the short list of recommendations that a brand like you know that a that an ai assistant would return to a shopper so the initial thing to do really is run a couple of those searches before you even before you you make any changes to your structured product data run those searches over time see which brands are showing up are you showing up Are you not showing up?

16:54Just to get a baseline understanding of how you rank. Are you showing up? Are you not? Are you ranking in the top five recommendations or not as a brand? Then you do the work to make sure the evidence needed for an AI assistant to accurately interpret and understand that your product is a good fit for a specific type of search is in your structured product data. So on your brand website, this could be what they call it. It's like your schema.org, like standard SEO type data structures. On a retailer's website, this could be making sure these claims are in your marketing copy or in your product titles.

17:40You want to make sure you have this consistent attribution on your brand website, on your retailer, PDP pages, so that there isn't any, there's a consistent kind of narrative that's structured in a way that these machines can interpret around

17:55Deana Burke:your product. So even the turn of phrase being the same helps with that? Helps with that, yes. Yeah, I think where the hiccups lie, I imagine, is like the subjective prompts. Like if I were to put in that same prompt, but I say the best tasting coffee beans, et cetera, et cetera, for pour over or for espresso or for cold brew, all the methods, I imagine that's going to change, right? And if you're that same brand, I guess, how do you balance all of those terms alongside, you know, and what the retailer and what social media is saying? Anyway, I just mean, like, I imagine this is where the brands are trying to figure it out.

18:40Deana Burke:Because best tasting is subjective, right, by design. Yes. Best tasting is definitely subjective. But I think as a brand, it is really important to also understand why your shoppers purchase your product and why is it that they purchase it. So what is unique about it that makes it the right choice for shoppers? And you think about making sure you're discoverable for those moments. Some of these broad category-wide type questions, you want to make sure that you have context in your structured product data to be eligible to surface. But I would say it's definitely harder to win probably some of these broader searches versus those searches that are uniquely related to a consumer's very specific preferences, which I think is where there's such unique opportunity for a lot of these innovative emerging brands who are bringing to market sort of differentiated versions of legacy products.

19:50Deana Burke:Yeah, I like that you call it opportunity and not a challenge, which is what maybe a lot of marketers are thinking. I'd love to hear, is there a specific brand or specific product you can maybe take us behind the scenes on how it's tracking right now? Yeah, I won't name the specific brand, but one of the brands that we were working with is a snack brand. And they are positioned around, you know, it's a kid-friendly snack product that is allergen-friendly. And they have a ton of attributes. So because it's allergen friendly, they're nut free, it's, you know, school safe, things of that nature. And one of the things, you know, when we first started working with them was looking at, are they showing up across searches related to those attributes?

20:45And it was really exciting. And what we saw is like, yes, they were showing up, but they were not showing up consistently, nor were they necessarily ranked in the highest sort of position in those responses. So we worked with them to implement structured data on several of their product pages on their brand.com website. And after we implemented these structured product data changes, what we saw was a 7x increase in referral traffic from AI sources. So on some of these pages where the AI assistants weren't even like indexing them or like it was like they were invisible to the AI assistants. Now they were starting to see traffic.

21:38So that was really exciting. When we looked across searches and their appearance rate along those searches, we saw like a 12.5 increase, 12.5 increase in appearance rate just within 60 days of making these product data updates where they were now surfacing more frequently and at a higher rank around searches related to like nut-free snacks for kids.

22:09Deana Burke:Do you think in this case, it helps it that it's what we see maybe as a niche product, but not free, allergen-free, school-friendly? These are all things that I imagine a parent is searching very specifically. And so to be a brand that can check off all these boxes, yeah, it's just a matter of getting in front of the customer. Whereas like, yeah, I think maybe some of the more broad prompts can cause a little bit more of an issue because there are so many brands competing for those eyeballs. Right. So I think you want to make sure that you are part of the consideration set at first. And if you're invisible entirely, you're never going to even be part of the consideration set.

22:50So having optimized content that is structured to improve your visibility to these AI assistants, even as a baseline, is really important to even be part of that set of eligible products that could ultimately surface. Mm-hmm. Yeah. And then I guess maybe we can talk about how this crosses paths with the IRL and the digital shelves that we've known for a while now.

23:23Deana Burke:So, yeah, are you seeing, I guess, the AI shelf affecting the traditional, like, let's say anything at Walmart, Target, et cetera? And are the products that are more likely to show up in the AI search getting more space in physical stores? I mean, this might be a little abstract right now. Maybe, you know, it's not something because I know retailers move slower. So maybe it's not something brands are presenting yet. But I imagine that is going to merge more and more, right? Yeah. When we speak to brands, they're telling us that retailers are coming to them and they're expecting certain levels of traffic coming in from these AI assistants to brands' pages on that retailer's website.

24:13And so there is definitely a conversation that's being had between brands and retailers. around catalog readiness, product data readiness to ensure that those products can be found wherever consumers are shopping. And since three and four consumers are now using these AI assistants for discovery, you want to make sure that you are discoverable there.

24:40Deana Burke:Yeah, well, it doesn't surprise me that the retailers want that traffic too. But I guess, yeah, are the products that are more likely to show up on these AI shelves getting more space? Like are these retailers, you know, seeing the data and giving them maybe like an end cap or just more visibility in real life? That's where maybe that's still something in progress and maybe it's harder to track for you. But I imagine that's maybe where things are going. Yeah, I think that's probably where things are going. And I think brands also need to, I think we have to think about how people are using these LLMs today.

25:20And I think it would be sort of naive for us to say that, you know, the category managers at RetailerX are not using AI to do further discovery around products when they're kind of thinking about which brands they want to bring on to their shelves, you know, in their next category reset. And so making sure you're discoverable, not only for the purpose of, you know, shoppers, but kind of thinking through, okay, well, what does this mean for my business and how might someone be evaluating our brand against competitors by doing their own research? You just want to make sure that you're showing up.

25:57And then also being able to bring the data to the retailer to say, hey, here's how much traffic my brand is getting from, you know, AI sources. And we're also seeing that when that traffic is coming in from these AI sources, they convert at a higher rate, which is the research and reporting that we're also seeing as well. So I think it will become a new signal to get on shelf at retailers. But I think we're still early to see that direct correlation or relationship between your presence on AI systems.

26:41Deana Burke:Yeah, it used to be where they wanted to see like your D2C or your Amazon data. And who knows, maybe this will be next. But again, discovery is not necessarily conversion. I think that's something I keep having to remind myself. And then for this last portion, I'd love to just hear your thoughts on some of this like wild west of it all, which is, you know, this like the fake deodorant getting surfaced. To me, that's pretty concerning. You know, there's not a lot of vetting yet on, you know, regulations around this. So if you are a brand, I guess not only are you competing with your actual competitors, but you also have, we know the web is littered with a lot of, you know, dropshippers and fake brands, literally.

27:29Deana Burke:So, yeah, I guess like how do you approach it from your end? because you're actually coming through all of the data, right? Yeah, I mean, I think so. You know, we look at it as an opportunity for, especially for emerging brands, right? I think AI, this AI shelf is sort of a democratized shelf where the smallest brand, if you do the investment to structure your product data, be legible to these AI systems, Like you can surface right alongside the largest national brands that are competing in the market alongside you. And so when they're real brands, that's a huge opportunity for those emerging brands.

28:17We have seen that there have been, I guess, bad actors who are trying to game the system. And I think it will be necessary for the companies who are really building out the infrastructure around agentic commerce, like full end-to-end discovery to transaction, to have guardrails in place to ensure that consumers ultimately are not negatively impacted by these bad actors who are trying to sort of game the system. Because ultimately, that would lead to a, I think, lack of trust in how these, trust in using these AI systems. And I don't think that that's where the market wants to necessarily move when it comes to this being even a new channel for conversion for brands and retailers?

29:22Deana Burke:Yeah, no, I'm definitely curious to see where the retailers and the brands take this, especially as things like reviews and recommendations and like affiliate, you know, all the marketing playbooks are kind of now shifting to accommodate all of this. So yeah, Yeah, I know it's still in motion and moving. So yeah, it'll be great to see some of the charts that you put together in the coming year. But thank you so much for joining us. This was, yeah, I definitely learned a lot. Thank you for having me. This was fun.

29:59Deana Burke:Thank you for listening to this episode of the Modern Retail Podcast, a show by Digiday Media. If you haven't already, please subscribe and head to Apple Podcasts to leave us a review and a rating. Find more of our coverage at modernretail.co and follow us on socials like LinkedIn and Instagram at Modern Retail. You can also follow me at Gabriella Barco, that's B-A-R-K-H-O, on all socials. See you next week.

30:41Thank you.