[00:00:05] Melissa Traverse: Hello, and thanks for tuning in to the Nonm-base podcast. I am Melissa Traverse, director of community at Bevnet CPG Media, and this is our How do I Build this series where we share practical advice to help CPG brands build stronger businesses. Consumers are increasingly asking AI what to buy. But when they ask Chat GPT or Claude or whatever AI platform they're using for the best coffee, snack, protein bar, or deodorant, only a handful of brands may show up in the answer. So how does AI decide which brands to recommend and how you make sure that yours isn't invisible? Deana Burke is the co-founder of Boys Club, where she's been studying how AI discovers and recommends products, running thousands of shopping queries across different AI platforms, and even creating a fake consumer brand to see if she could get ChatGPT to recommend it. Today, we are gonna get into what Deana calls the AI shelf, how to figure out where your brand. sits on it, what you can do to improve your chances of being recommended and ultimately what needs to happen for AI discovery to turn into an actual purchase. We're also going to look ahead to a world where AI may not just recommend what we buy, but maybe even make the buying decision for us. Deana, I am just so thrilled to have you on here. Thanks so much for joining.
[00:01:27] Deana Burke: Thanks so much for for having me. It's an honor.
[00:01:29] Melissa Traverse: So I heard you on an episode of Marketplace, and um we'll get into the deodorant brand that you created and tested a little bit, but I thought what you were doing was so fascinating, and I had to reach out. So again, I'm just so glad that this came together. Before we get into the nitty gritty of um all of the AI topics I mentioned, tell us a little bit about yourself and about Boys Club, um, you know, what is it and, and who is it for?
[00:01:56] Deana Burke: I'm a, a career marketer, career tech marketer, um, and yeah, have, have spent a career working on, um, sort of frontier technology and doing the, the marketing and storytelling around that. In 2020, started Boys Club, which is now a media company which covers technology. We originally started it to create an environment and events for women at first to learn about what was happening in emerging fields, and now we've expanded our scope. We cover tech, and most recently, I have been doing a research around AI and LLMs and product recommendations, really inspired by my background as a marketer and really just trying to understand this as a new surface area for marketing and for purchase and emerging consumer behavior and trying to figure out what it all means. And so it started from that curiosity, but now it's turned into a bit more.
[00:02:50] Melissa Traverse: Let's start off with the AI shelf. You've been mapping that out. Can you explain to us what the AI shelf is and why someone who's running a food or a beverage brand should pay attention to it right now?
[00:03:03] Deana Burke: When someone asks Chachabite or Claude or Gemini, what sunscreen should I buy, for example, the model answers with 3 to 5 brand names in a list. And that shortlist is what I'm calling the shelf, the AI shelf. It is Delivered in a quite a authoritarian tone or quite a confident tone by the LLMs, it changes between the different assistants in the different chatbots, but that is essentially the shortlist of brands that an LLM reaches for when it's creating a recommendation is the AI shelf. It is quite opaque in terms of we don't know exactly what goes into the weights or the what goes into the decision making for the AI to make that recommendation. All we can really do is track the output, which is what I'm doing, um, with my ongoing research project. But yeah, that's essentially the shelf. You could just think of it similar to a grocery store shelf where, you know, eye-level placement is a result of lots of different things, negotiation with the retailer and the brand, uh customer preference. There's a bunch of different things that sort of swirl around to determine what that grocery store shelf. Looks like, and it's the same thing with the AI shelf, however, I'd say quite a bit more opaque, but yeah, that's what I'm looking to map, and that's, yeah, the work that I'm doing currently.
[00:04:28] Melissa Traverse: And could you tell us what LLM stands for? You mentioned it before and um and I meant to ask you, but could you explain that?
[00:04:35] Deana Burke: LLM starts stands for a large language model, and you could basically just think of it as The way that you're interacting with your chatbot, uh, of choice. So whether it's Cloud or it's Chatub BT or it's Google Gemini, um, or Metas has one Muse, uh, it's basically like the infrastructure behind how those chatbots work and what, how they recommend the things that they do. Um, if that makes sense. It sure
[00:05:03] Melissa Traverse: does. Thanks. So I know that you've run, you know, maybe even thousands of shopping queries across different AI models. What have you learned about why certain brands get recommended over others, and then, you know, some barely exist at all. What are the signals that seem to matter most?
[00:05:24] Deana Burke: The first and the biggest thing is that the head terms. are very much frozen. They are very hard to move. So an example of that is, um, you know, best natural deodorant. There's gonna be 5 or 6 that are brands that are surfaced, and what I'm seeing at least in, you know, the 6 or 7 months that I've been running these, these tests, is that those head terms and those brands that are surfaced around the head terms, they don't move, they don't move over time. Now, they do move considerably when there's a model change. So when, for example, Chachi BT releases a new model, a new model upgrade, there can be considerable movement. Between models, but over time, things are, are feeling pretty fixed at the top. Um, now, that might sound discouraging, uh, but it's not, because I think where there is real opportunity is in the long tail, and we see a lot of churn in long tail and specific queries. The, the headline of what I've been seeing so far is head terms are fixed. If you got there and you're an incumbent, congratulations, that's incredible for you. Um, but if not, it's, there's still a game to play, but the game just looks a little bit different, and it's, it's about these sort of long tail specific queries, and that is, you know, you spoke to my experiment with Moroan, which was the um made up natural deodorant brand uh that I did over the summer, and that the long tail tail queries is where that brand showed up. So those are the, those are the big um The big findings, um, there's a lot of consensus across models, so, um, there's, you know, there's little nominal changes between them, but for the most part they, they are agreeing, so it's, you know, the work is to, um, it is to not game for any specific model, not to not game for Chachabuti necessarily or Claude necessarily. It's, you gotta go broad, um. And yeah, the, the wording of the prompts and the question is really where, where the magic is, so we can get into that now, but if there's anywhere else you'd like me to expand on, I'm happy to.
[00:07:40] Melissa Traverse: Would you explain head terms?
[00:07:42] Deana Burke: That kind of like goes back to SEO days where um a head term is gonna be like Yeah, best, um, running shoes. They're very broad terms and queries that are really gonna surface the, the big players and the, the incumbents in that category. Um, yeah, best, uh, cheapest is another one, it, it's that kind of that type of query, which for some shoppers is how they prompt and is how some people ask Google, like what's the best. Whatever, but I think the argument with LLMs is that they'll be increasingly personalized as we use them more, as consumers trust them more with their information, better, you know, for better or worse, I'm not making a, a claim about whether that's a good or a bad thing, but just speaking to sort of like trends and uh. Yeah, as, as these LLMs are increasingly personalized, um, those head terms may, may matter less.
[00:08:47] Melissa Traverse: And when you say long tail queries, are you talking about that more expansive query where you're saying like a natural deodorant, I can buy it at Whole Foods, it comes in this scent, that kind of thing?
[00:08:58] Deana Burke: Exactly, yeah, so uh it might actually a good way to illustrate this might be for me to talk about Moroan, the natural deodorant, please, OK. I created a fake deodorant brand. It's called Moroan. The pronunciation of it is debated, but it's fake, so you can really pronounce it however you want. It's called Moroan, and uh I created that brand, um, made it up, created a web page for it, uh, created a couple pages within that site of product Q&A, and I created a sub stack post for it. That was in the brand's own voice, and I Set it aside and then I went on vacation, and basically came back from vacation and had been running a script that was querying all the different LLM querying all the different chatbots to see if they would pick it up and 3 weeks in, Chacha BT was browsing on started to pick it up and started to recommend it above other very legitimate natural deodorant brands, and so that was the Moroan experiment across, I think there was like 4 or 5 sort of long tail queries that I was testing it for that it got picked up for, and again it was like ranking above. Um, very real brands, and so, yeah, that was, that's the Maro experiment, and it went a little bit viral and was fun to see, but it was really an effort to see if I could move to the shelf, if the shelf was movable, and where is it movable, um, and sort of speaking to the long tail of it all, that's where it was movable.
[00:10:34] Melissa Traverse: So you said you, you put a query out. Could, could you explain what that means as well? How do you, I mean, if I had to guess, it's like hanging your shingle out so that AI agents can discover you. Is that what it is? And how do you do that?
[00:10:48] Deana Burke: Basically what I did was I, I just built this website. And transparency, I had Claude built the website. I didn't do it all the the website design and build process took me all of 20 minutes because I was really just optimizing for like this to be as quick of a process as possible. It was a fake brand. I was never gonna do anything with it. I just wanted to stand up a site and see what would happen. So if you go to the site now, you see like there's like the graphic that comes up, that's like the product hero shot, and it's ridiculous, like it's illegible completely, and it's just AI slop essentially. So I had uh Claude build me this website. It was again a a main landing page, and then it had product Q&As that was built out, um. Uh, we had invented the product. It was a, a, a, a natural deodorant that was made with colloidal oats, and like we had a whole set of specs that we, we built together, me and Claude, and, uh, yeah, and, and then I, uh, basically what I did was I, I made sure the site was indexed uh on Google and Bing. So that's just a process of, uh, you know, there's a process where you go to those sites and you submit pages and subpages and an index. So I made sure those sites were indexed before I left for vacation, and then I left for vacation. And just let it do its thing. I didn't do, I didn't do anything after that point. I, I sort of, it was a set it and forget it type of situation, and, but what I was doing was I had, uh, basically, uh, a script that was running automatically every week that was testing to see whether the chatbots were picking it up with the set of queries that I had, like, I have a sensitive skin, I'm looking for a natural deodorant with oatmeal, what do you recommend? So it was, those were the queries that were running automatically, but you know, that was, that had no impact on whether or not they were picked up. That was just me in the experiment phase, seeing if what the results were.
[00:12:44] Melissa Traverse: There are so many folks in CPG who come from the tech world who are probably rolling their eyes every time I ask you to explain something, so please do. So some folks in our. I'm sure, you know, get it immediately. But um, so how did you, in the span of just a few weeks, find yourself being discovered by AI agents when it seems like it's fairly difficult to get picked up considering the size of the shelf and the brands that are already there?
[00:13:14] Deana Burke: Yes, definitely. I would say that me and Claude together, we designed the product, uh, In what we identified to be like white space basically in the category. So we, uh, I did some searching around like what does natural deodorant look like. First of all, to be clear, Native owns head terms completely and like didn't, it didn't shift at all. Best natural deodorant. There was no world where Moroan was going to rank there, nor did I expect it to, uh, but Basically, yeah, it just got really sort of specific with the specs and what condition those specks could speak to, and we were like just wrote that out explicitly in the Q&A. So for example, like, um, yeah, I have sensitive skin, I'm looking for a natural deodorant that X. That's like exactly how that that text is what we put on what I put on the Q&A page. I'm sorry, I keep saying we, that's a ridiculous thing to do since it was my job hop, but like it felt like I was collaborating. But yes,
[00:14:23] Melissa Traverse: that actually feel that actually makes perfect sense.
[00:14:28] Deana Burke:: Um, it's mostly because like I don't want to take full credit for it because it was a collaborative experiment that was happening, but yes, that I put on the Q&A and then when Chacha BT did pick up the brand and recommend the brand, it was paraphrasing back that exact language. So My advice to um to brand owners is to, yeah, building out those Q&A pages. Speaking specifically to how people might Speak around your category or your brand, that was what did it for Moowan, uh. It was like exactly almost word for word repeat of what was on those Q&A pages. So that's, that's how it ranked.
[00:15:13] Melissa Traverse: And when you say Q&A pages, are you talking about the FAQ on the brand website? Yeah, yeah, OK,
[00:15:18] Speaker 2: OK. A nuance and a detail there is, I think there's a lot of received wisdom, especially nowadays around SEO and, and making your product pages be very data rich, and I think that that's important. And I think it's important to have, like, all of your specs in plain language, and all of that is able to be indexed. Both for search, but also now increasingly for AI as well. And so there's both the product and and data layer, data enrichment around your product specifically, but there's also this sort of second thing, which is, yeah, FAQ Q&A. Plain text question, I have this condition, I'm looking for this specific thing, and then plain text answer that uses your brand and products and specs within it, that to me, at least from what I'm observing, seems to be the thing that has the most impact.
[00:16:22] Melissa Traverse: I mean that that right there to me seems like a great nugget of information. That's something that any CPG brand could, you know, press pause on the podcast and do right now and change about their brands. Do you have, do you have any other advice for CPG brands when it comes to discoverability in the way that you're wording your Q&A or anything else on your website or anywhere else?
[00:16:49] Deana Burke: I think for food and beverage brands specifically, there's a lot of different categories and sort of cuts at the data that you could do. So, there is a dietary constraints, so not free, no seed oils, low acid. There's also occasion that you can layer on top of that, so post-workout, pregnancy safe, kids' lunchbox, then there's also the format, shelf stable, single serve, no refrigeration. And then there's also uh maybe this like more nuanced thing which is, OK, every, um, what if every X does Y to me? So like, every, what if every sparkling water tastes metallic to me? What if every, every protein bar I taste tastes chalky to me, and then having a response to that as well, which feels crazy and, you know, can be sort of buried at the bottom of your FAQ page if it's way too specific, but Having very specific answers to very specific questions, I think there's potentially an unlock in that for, for especially food and beverage brands.
[00:17:51] Melissa Traverse: Great advice. So you found that AI treats different categories a little bit differently. If I run a CPG brand, how do I figure out how AI understands my category, you know, which questions customers are asking. Which brands own those answers, and, you know, where I fit in.
[00:18:13] Deana Burke: Well, I would encourage everyone who is uh listening to run a baseline of what their category looks like right now, uh, and how Each how the category that they are in is being answered. So there has been this thing that started to develop recently that I've been observing in the models. Where for more of the commodity. Uh, types of products, the models are answering with specs. So, I'm, I'm, I'm looking for a recommendation for beer. What would you recommend? It starts to list, um, Uh, well, for there's IPAs and there's loggers, and there's it, it starts to go into categories and specs instead of naming brands immediately. Uh, so I think it's useful to know if that, if you're in, like, for example, tinned fish. I did a bunch of testing around tinned fish. And basically it wouldn't name a brand ever, because it was just basically like, oh, do you like herring or do you want tuna? Like that's not, it's not, it's not giving me brands. So it's, it's really useful to, first of all, just get a baseline of how it's answering, um, which could then inform taking those insights, and then informing the Q&A or FAQ page that you're building out on your website, basically looking to see like, how, how is it, how is it answering and how can you design to how it's already answering at the baseline. Um, but yeah, I think probably for, for a lot, um, Probably for for a lot of the folks who are listening who have brands who are more in more commodity categories, that's what it would probably look like.
[00:19:55] Melissa Traverse: I read that article on tinned fish that you put up on Boys Club, and I think you said, you know, maybe you expected it to suggest Fish Wife, for example, because they sort of like. But so if I'm fishwife, and I'm not there, which is still shocking to me, then would you just go ahead and kind of repeat those steps that you just mentioned, like you would fill out your FAQ sheet on your um website, that kind of thing?
[00:20:21] Deana Burke: Totally, because I will say the next question that when the when the models are replying with basically specs or they're they're replying with uh something that's more generic around the category. There, the next question is, you typically a prompt for personalization. So, Uh, with, with running shoes, with, with some of these other things, they're, the, the models, at least so far, are more willing to put a product up 1st, 1st shot. Basically you ask and and it'll say Brooks, but with coffee, beer, things like that, fit tinfish, they're wanting to, the models are wanting to get more information out of you before they recommend a brand. So it's still totally worth doing the work, it's just that There's again that personalization layer, which honestly is an opportunity for smaller brands, cause then you're starting to get into occasion and long tail and the things I spoke to you earlier.
[00:21:19] Melissa Traverse: Are there paid options for discoverability on AI yet?
[00:21:25] Deana Burke: There's a few different ways into the paid conversation. There are ads on Chatcha PT. I have not tested them myself, uh, so I can't speak to how effective they are. That's one of the things I'm planning for this next upcoming quarter, cause super curious about it, and also super curious to see if you are buying ads on their platform. Do you then higher elsewhere, um, they would say no, but I would be curious. I think if you're a smaller brand, a challenger brand, uh, startup, I probably wouldn't invest there. Um, but, uh, yeah, that's happening, and, and perhaps in a couple of years that will be something that is more impactful. Uh, and then, yeah, there's, there, there are, um, Service providers and, uh, you know, sort of SA companies, tech companies that offer Sort of a a behind the scenes gaming to get your brand ranked. Um, I think the Efficacy of those range as all agencies and products range in efficacy, so I can't speak or endorse anyone in particular. I will say they're doing a version of what we're speaking to here. Although I will say some of them are also doing the, the gaming on Reddit and the various places where they're trying to sort of plant seeds. I, I will say I did not do that at all with Moroan, uh, I didn't do any planting of comments or anything, so, um, but I've seen that. be used as evidence for it working elsewhere, so that might be something that I just don't like to, I just don't like the spam element of it. It feels bad, so personally it's not something that I would recommend.
[00:23:04] Melissa Traverse: You mentioned earlier in the conversation that you shouldn't necessarily target one AI platform like ChachiBT or um Cloud or Gemini or whatever. How is, is there a way that we should be thinking about these AI platforms differently from a CPG perspective?
[00:23:23] Deana Burke: My earlier advice still stands. I don't think you should try to be designing for anyone in particular. If you're doing the work to design for AI discoverability, then There's, you're going to be doing the work hopefully to rank across all of them, any of them. I think you need to index your site on Bing, which works with Chat GBT and Google, which obviously is a a gem uh which will be a Gemini Google Gemini relationship there, um. There's also Muse, which just launched, uh, which is Meta's, uh, um, chatbot, and I will say that they are Uh, seem to be much more focused on the commerce, uh, element of their user experience and did announce a partnership with Shopify. I've seen really promising stuff coming out of Shopify in terms of how they're able to optimize for agentic discovery and even potentially like through to payment, which is something that we've not really seen on the other platforms yet. So I think Muse is really interesting, um, in terms of like designing your web page and your FAQ like, I don't think anything would necessarily change there. I guess the only consideration is if Shopify is doing these partnerships, certainly with Muse and perhaps others, might be a thing to consider, um. I have no horse in the Shopify race, but I have, I have seen some, some interesting things come out of, uh, there was like a, a, a Q2, um, earnings report that they spoke about, yeah, just there being a huge increase in, again, that long tail of like specific Queries and then it opening up discoverability for like lots of different niche brands that previously weren't very discoverable, especially on like the first page of Google, which is where we were doing that type of work previously as consumers. So, uh, yeah, I think really interesting stuff there, um. But yeah, I, I think if you're doing the work, you're, you're doing the work for all of them.
[00:25:28] Melissa Traverse: I think the long tail query opportunity sounds especially promising for those brands who are scaling, because they sort of need that differentiated value proposition in order to knock a challenger brand off an actual grocery shelf, so that really does seem pretty promising. Yes, definitely. Um, so you mentioned that you didn't, you know, plant any fake information on Reddit or anywhere else, which I totally respect. Um, I would be really sad if I fell in love with the deodorant and then realized I couldn't buy it. Um, what are some of the other information locations that you've found are helpful to seed with information if you're selling a real product? Like, I know when I search something, Reddit. Almost always comes up, I think, you know, on ChatGPT, um, you know, is it reviews, um, retailer sites, media coverage, like podcasts, where are you finding the AI, um, platforms are pulling from?
[00:26:24] Deana Burke: There's definitely general media is in the mix. So anything that's like syndicated or has like affiliate heavy outlets like a Forbes or some sort of mainstream news outlet. Whether or not we as humans read them anymore, uh, it is a, a high credibility source for these AI models, um, and there's also like category review sites seem to be, uh, something that at least in the citations come up, although I will say citations are all post hoc, so it's really hard to determine, meaning A post hoc citation is uh the the LLM generates a response, and then it, if you ask it to tell you where it got that response from, it is then answering that question separately to say like this looks like a likely place that this answer could have come from. So the, the idea of post hoc citation, honestly, when I first learned it, blew my mind cause I was like, wait, that's not exactly where the LLM where the model got its answer from. It's not, it's just saying that this is There's evidence for this answer on this page, and then sometimes when you go to that page, there's not even evidence for the answer there. So I would say the source's conversation gets really murky, really fast, and um the impact of that is I would say that any of these category review sites. Um, they're not, or any mainstream media sites, they're not guaranteed a placement there is not guaranteed to pull it up into the AI recommendation, um, even if you're seeing that as evidence for other answers on the AI. This is kind of convoluted, but happy to dig in anywhere that might be useful.
[00:28:14] Melissa Traverse: I have a question that's actually a little bit out of left field, but I'm, I'm just curious about this. Like sometimes I'll ask ChatGPT or Claude to answer some question and to only answer with facts that it's, you know, like from articles or facts that they've found online and to provide sources. Is that, how does that fit in with what you just said about it may or may not have come from that source? Am I being lied to?
[00:28:44] Deana Burke: It it's a little bit of theater, I will say, um, and there's also sort of zooming out, there's two ways to think about how these models retrieve information and Recommendations. The first way is um A way to think about it is like browsing off. It's in the models themselves, it's in the weights themselves is is how it's sort of referred to in with AI. That meaning like There's training data Past historical training data that was used to inform a specific model. And that training data has a cut-off date, and When you're asking ChatGPT. And it doesn't have access to the internet, that's what it's accessing. It's basically what's in its training data already. That has already been published, that has had a training cut-off date, and that's what's in the model itself. You're also able to Access the internet now with the AI models and the chatbots, that's what we call with browsing on, and that's what you're speaking to, where you're putting in a prompt and you're asking the chatbot to provide sources. It Uh, generated the response to your question, and then is going and finding sources to back up basically what it said, and that's what we call post hoc citation. So, um, Yeah, the, the sources, uh, again, it's, it's kind of a murky land. Um, there's also now a whole cottage industry of people who are building, using AI to build basically clones of category review sites, top 10 running shoes, and it's a site that they own that they are putting their brand as number 1, and There's a lot of them, and there will be an increasing amount of them, uh, especially since basically the cost to create them is next to zero by using AI. So, uh, yeah, I would say sources get weird, um. Sources get weird.
[00:30:59] Melissa Traverse: And, and when you say training data, is that training data updated every time there is a new update to an AI platform? Yeah, OK, got it.
[00:31:08] Deana Burke: Yeah, so that's like, I, I don't, can't say exactly, but speculate that's why this, it can shift so much with the model update is because basically there was a different corpus of data that it was drawing on.
[00:31:21] Melissa Traverse: Fascinating. Um, your deodorant experiment, it also showed that AI can take brand created copy and return it to the reader in a way that sounds independently vetted, which is basically, you know, what we've just talked about. What does that tell us about how AI decides what's credible? And then, you know, how brands can use that understanding to enhance their credibility.
[00:31:48] Deana Burke: Yeah, I think, uh, well, I'm gonna speak to what it looks like today. I do suspect that Similar to how um the the You know, 15 years ago with Google and search. Everyone started gaming it, and when it was when it was more nascent, everyone sort of figured out the game, figured out how things were ranked, and then started building content farms essentially to game to get to the first result on Google or the first page of Google, and then over time, Google To tweaked their algorithm and tweak their product experience to make it so that that gaming was less effective, and uh they figured out how to make the experience as a consumer as a user better, and, but that took some time for them to figure out, and it took people using it and people feeling like they don't want to go to a content farm website and uh so I do suspect that will be similar to how these Labs and chatbots start to refine their product over the next couple years. So yeah, my brand, I had no, it was a zero authority website, and again I had the website and then I had a sub stack article that was a 00 authority sub stacker. It was the brand itself publishing an article about itself. So there was nothing to say that any of those things should have been prioritized in the algorithm for how the LLM was going to be surfacing it as a recommendation, but it did. Um, but again, I do think that that will change, um. But perhaps there's an opportunity now. I, I don't know. I, I do think that like, as a consumer, There's a, there's a sort of the other side of the conversation, which is As a consumer, it's quite alarming to be um, you know, given a recommendation by a chatbot for a product that doesn't exist. Primarily because of the The tone and the confidence in which these chatbots talk, you know, type or talk or how they, the tone of the recommendation is like, I know. You, you need colloidal oatmeal, and I'm gonna tell you exactly the product to use to get to the condition that you're looking to, to fix. So, uh, and also, which is a little bit more of a subtle thing, but the recommendations are all, you know, you're, you're in the chachabuti experience, you're in the Muse experience, it's all in the same font, and That might seem like a small thing, but for you and I, we have a literacy now where we, we go to Google, we search whatever we're gonna search, and we know now that, you know, that we're gonna get blue links and a little description, and maybe one's gonna be an ad, and that might look a little slightly different color, or we're we're able to see the, the, the, the slug of the, the URL that says, oh, this is the brand's own website, maybe it's gonna talk nicely about itself. I can skip to the next one and go to Reddit or something that You know, what is more, whatever it is, you, we have now learned how to navigate that as a surface area, and that is not true at all whatsoever of chatbots, and so, yeah, just from a consumer perspective, that's something that we're both going to have to be learning, and I do think that the, the brands and the, the, the tech products themselves are gonna be refining over time to improve.
[00:35:31] Melissa Traverse: Well, certainly until the AI platforms kind of catch on the way that Google did, it's game on. um, so getting recommended, uh, you know, if, if you have a CPG brand, having your product be recommended is one thing, but actually having it be purchased is another. So once AI surfaces my brand, let's say, um, typically when I search for products online, they'll give you either a link to Amazon or a link to, you know, some like a maybe the website itself. How do you see that changing and how important are things like retail distribution, availability, price, review, you know, that kind of thing to actually being discoverable in a way that makes the searcher purchase.
[00:36:17] Deana Burke: I think that the uh The purchasing through a chatbot is something that everyone is looking at and consumers are not doing yet. The technology is very emerging there and very nascent, and I do think that it's coming, but we don't yet have the technical infrastructure. is kind of OK, but the rest of them, people have just the infrastructure isn't there to support like a wallet that is then going and and making the purchase itself. Do I think it's coming? Absolutely. Um, do I think that there is a, um, A a a sort of trust threshold that needs to be uh achieved before consumers are willing to do that, um, definitely, yes, and willing to do that autonomously, like willing to do that without like explicit oversight even more so. So I think we're a few years out, but, um, yeah, the, the Through to purchasing is really interesting to to consider. I will say, I have been running, um, on the question about retailers, I have been running an experiment specifically around holiday uh shopping and um different gift categories in holiday shopping. I started a couple of weeks ago, and really interesting thing is coming up on that research, which is that I had suspected. That Amazon was gonna be the top recommended retailer for whatever, anything, and it's not, it's really, it's buried. Target's number 1, Amazon is number 2, but it's like Target was like 36% of answers, and then Amazon was, I think, 7% of answers, and then Walmart was less than 1% of answers. So, uh, the retailer question I think is really interesting, um. And then also within that research, which is new research that I'm working through now, uh, there was a lot of um brand direct links, uh, so, um, and I, I think this is honestly speaking to the Shopify stuff again, um, small specialty brands were getting like going straight to their Brightland.com or whatever it is. Uh, so, a D2C first food brand, like having site optimized, I think is as important as it is to have those. Retailers locked for at least for the Asians.
[00:38:42] Melissa Traverse: Especially now that when you type a query into Google, it doesn't take you to, let's say the brand's website, like if I ask, you know, what's the coolest tinned fish brand there is out there right now, Google may say that it's Fish Wife, but it doesn't necessarily take you to Fish Wife's. Website anymore, so, you know, thinking ahead to the future, if it, if that is what happens, where like if you use Muse and you ask a question about a product, and the website comes up as the place to buy it from, that's huge,
[00:39:15] Deana Burke: huge, huge, and huge for brands where like your margin, your your whatever, cutting, cutting Amazon into the Uh, transaction probably sucks in a way that if people are going direct to your site, it's way better. So, I think that that is really exciting, um, to see, and, uh, yeah, it's almost like the, the retailers and the retail distribution acts as authority for the ranking for the AI itself. So the the AI is going out and saying, OK, where is this hopefully a real product, and it's seeing, looking at all the stores where it's perhaps um Listed, and then should that reach a threshold that it's happy with, um, is surfacing your brand and ideally your website so that people can buy directly.
[00:40:03] Melissa Traverse: Well, that sounds very promising. What else do you see on the horizon? What else are you keeping your eyes open for when it comes to CPG and AI?
[00:40:12] Deana Burke: What I was speaking to you earlier around uh the impetus for this research was really just coming from a marketing background and Thinking that, uh, yeah, in the I have a thesis that in the medium term, we will see the buyer change species from human to bot, and when bots are making purchase decisions, they're making those decisions using a very different set of criteria than humans would. And when I first started this work, it was like, as a brand marketer. Does a 32nd Super Bowl TV spot, does a big flashy brand marketing campaign matter when a bot is shopping on specs, or it's shopping on reviews, or it's shopping on things that certainly aren't like the human emotive feeling that we as humans shop on and make purchasing decisions on, you know, and I'm passing a grocery store shelf, and I see Something that's in a color way that speaks to me and that I find attractive, maybe there's an end cap, like that's there's this entirely different set of criteria that a bot would use. I don't think that this will necessarily affect categories where it's where humans are enjoying the shopping experience that much, so like, for a luxury bag, I think this is, you know, those, those big flashy brand campaigns will probably have more longevity. Uh, but if I'm P&G and I, I'm selling Tide deodorant, and I am hoping that markup in the difference between the generic to the Tide brand, that might collapse when bots are, are buying and when they're making purchase decisions, because the bot will be Has an almost endless appetite for dealing with friction and for shopping on specs and for shopping at the level of detail that you and I would never have the patience or willingness to do, and the bots will and can, and so, yeah, if I, if I'm, I, I think that the shortest term potential victim is, is household goods, so that's an area that feels um At least if, yeah, again, if you're P&G I would be really paying very close attention. Um, probably less so on, on fun or luxury goods, but yeah, I'm, I, I'm thinking a lot about that, and, and honestly like how that pertains to my field of marketing and, and how we think about, like, is there a whole separate internet that's for the bots. And for the humans, and I think yes, eventually, I don't know how quickly we get to that future, but um I think that that's what the marketing discipline will look like largely over time.
[00:43:01] Melissa Traverse: That's a lot to think about, you know, that last comment, it almost sounds like, I don't know if you watch Stranger Things, but like the Upside Down world, there's like the consumer, you know, internet and the bot internet, so it is a lot to think about. Um, is it possible that, you know, if there are 3 things that you could recommend a growing CPG brand to do to make their brand more discoverable on AI, what would those 3 things be?
[00:43:28] Deana Burke: Yeah. So first, definitely run a baseline test right now. So go and um as much as you can get, I run a script and and use the API to query all these LLMs. If you have the technical know-how to be able to do that, uh, you'll get cleaner data back, um, or if you're able to tap someone who can. Uh, or also you could take this transcript and give it to Claude and say, hey, can you build me a script that can, uh, query and try and get to, to clean data around some of these baselines. And basically, yeah, running, um, running to see what your, what shape your category takes currently, and who's winning, what are the different questions, um, and, and how is it answering different based on the different questions. So that that would be the first thing for sure, um, and then the second thing is I I sound like a Brooklyn record, but like really building out those. Those Q&A pages, those FAQ pages, uh, yes, enriching your product information as much as you can in plain text language, but also bringing in those, um, yeah, like the conditions, the occasions, all of that into FAQ. Let it be as long as possible and then be updating that frequently. So I do think this wasn't necessarily relevant for Moroan because it was a one-time thing, but I do think that recency and freshness plays into how the algorithms work around recommendations. So, um, making sure that you're adding to those FAQs, you're tweaking those FAQs, you're changing them frequently. There is a way to see how much traffic, if, if you have access to your own site analytics, there's a way to see how much traffic is being driven by AI currently, so there'll be um A way to see like your source traffic on your site, and it will say like source open AI or whatever, and it's it's really interesting. I think it would be really interesting but if I had a brand right now, I would be looking at how that as a source affects conversion. Versus People site visitors that are being referred from other sources, and again, like getting a baseline of that and measuring that over time. To see if it's even effective, like, you know, that it may be that your referred traffic from AI is converting more often or less often. It's just, you know, the information is power with with those types of things, so that's, that's another thing that I would look at for sure.
[00:46:03] Melissa Traverse: Amazing stuff, you know, I would say I hope everyone out there is taking notes, but you don't have to just take the transcript and feed it right back into cloude. That was incredible information for everybody out there who wants to learn more about Boys Club, and I'm sure that they certainly do. Where should they go?
[00:46:20] Deana Burke: You can find me. Uh, my handle is wet lauded on Instagram and very active there, and uh LinkedIn, my full name, and then Boys Club, uh boysClub. VIP is where you can find us and find more information about what we're up to there.
[00:46:36] Melissa Traverse: I'm gonna follow it right now. Dina Burke, co-founder of Boys Club, thank you so much for joining the Non-Base podcast. That was incredible for everybody else in the audience. Thank you for tuning in to the Non-Base podcast, and we will see you next time.
