Is Share of Prompt the New Share of Voice?

Speaker A: I would say like the majority of the space right now is just showing you the problem right there. There's like performance dashboards and showing you how you're showing up, but never actually telling you how to change it. And if they are telling you how to change it, they're either giving you a ton of optimization recommendations, never really tying it back, or there's this concept just of AI generated slop. And so basically they, they push out a thousand different variations of articles or websites, et cetera. And for me, it's creating an AI slop ecosystem. And so we are not doing that. We've been in the industry long enough. Like, if you do that for long enough, you'll actually get delisted from search. Right. Because it's, you're spamming the entire Internet and that, that is not a space you want to be in. Let's. That's like Internet jail and you're not getting out of that and has massive implications for your business if that's the case.

Speaker B: Hi, I'm Greg Stewart, CEO of the Market Media Alliance.

Speaker C: I'm Rex Briggs, an author, marketing researcher and entrepreneur. And this is Decoding AI for Marketing.

Speaker B: Hey, Rex, listen, we've been digging in deep on the whole AOGO gem. I mean, you know, or you know, Michael Pettit like came out with that whole framing of the different sort of dynamics of this whole new space for marketers. And I think what's interesting, and I, you know, you saw some of this at the board meetings last week. It feels to me like 2 months ago the CMOS just lit up and went, oh my God, like this is an issue I better pay attention to. And so I thought it was going to be a consumer board conversation amongst them as they try to sort through what to do here and the concerns and issues they had. Remember that?

Speaker C: I do, and I take it a little bit like a canary in the coal mine. It's not the only issue that you need to be paying attention to with AI. There's quite a lot actually you need to be paying attention to. But if you can't get this right, oh my God, you have so much bigger problems. Yeah.

Speaker B: Anybody who was in the search space knows that like got pattern recognition, how important this is going to be. Is that what you mean?

Speaker C: Exactly, exactly. Your ability to be surfaced when people are looking for you is understood for 20 years now since Google really became the dominant search engine. So I think it's an easy bridge from where they've come to understand this is going to be important. And I think One of the biggest challenges I see is that the marketers want to not only know how they're showing up, but how they can influence the outcome. That's their job, to get better results for, for their companies. And there was a new study from Seer Interactive that revealed that 83% of AI citations originate from pages that do not rank in the traditional Google top 10. And that basically means you've got this decoupling from whatever you're doing for your SEO, from whatever might be happening in your share of voice you're going to get in your geo. And so I think that that's probably creating some anxiety for people because they know they have to do something different, but they're not sure exactly what to do. We do know that there's some signs that cross platform corroboration is really important for these AI engines. But again, I think, you know, really Greg, what we need is a lot more research to know what sticks and what doesn't and why.

Speaker B: Totally, totally, totally. I mean that's what was so interesting. I mean, listen, I was there with Tim Armstrong when Google came on the, you know, the horizon. I was running the IB at the time. You were with me doing the multi touch attribution work that we were doing. And it was interesting to me just how much baseline knowledge had to be put in place. Like the one that I remember the most is that um, we did a whole series of studies to understand the brand. Effect of search wasn't even on anybody's mind at the time. I'm not seeing a lot of that research and I think we still got a lot of work to do here for this category to fully develop and fully be understood and taken advantage of.

Speaker C: Well, and speaking of Google, we have a Google alum with us. I remember when that brand research, Larry, uh, Page was, it was important enough for him to meet with you and me and give us a, uh, Google hat and thank you as well as the check to pay for the research.

Speaker B: Mostly what I got from Larry and Sergey was telling me how stupid advertising was. That's what I remember from those early days. I guess they've changed their opinion now from their yachts and wherever. But uh, listen, with us today is Justin Inman. He's the CEO and founder of Embryos. They do a lot of work with Stagwell. In fact, I think they're the basis of sort of the Stagwell engine, a big part around this. We'll let him tell you more about that, but we're very excited to have him with us today. So, Justin, welcome.

Speaker A: Thank you, thank you for having me. I'm super excited to be here.

Speaker C: Yeah, it's great to have you here, Justin. So tell us a little bit about the idea of share of prompt and why it's an important metric and how marketers should be thinking about this space to get us warmed up.

Speaker A: So Embros is the operating system for AI, uh, visibility and brand orchestration. So we help agencies and marketers understand how they're showing up across these major LLMs or AI Ah, surfaces. And then we have predictive capabilities. So we have what we call predictive fix packs or optimization strategies that work across paid, owned and earned media. We track those changes in our end and we actually show the lift associated when those changes are actually executed. So we're the first closed loop platform in the market and for us, the core metric that I think defines it all is share a prompt and it's really our version of share of voice for AI surfaces. So, um, we didn't think that the typical share of voice that or share of church that you see across other traditional channels fit perfectly within this space because it's much broader. We look at AI visibility across your entire digital footprint and like you mentioned, Rex, it's everything, right? It's your PR team, it's your social team, it's your influencers that you work with and how all of those are interconnecting and, and really building the profile of who you are and how you're showing up across these major LLMs.

Speaker B: Hey, Justin, the way you kind of said that though, you made it sound like, hey, if I make a change in, I don't know, Reddit or someplace else that you can measure that impact immediately. Is that what you just said?

Speaker A: Not that fast because the models all operate differently. But we think we have a high confidence level. If you make this specific change, maybe to this Reddit post, we believe that it will have X amount of lift to share a prompt. It will take about two to three weeks to L and there's a cost associated with that. So we want to make sure that we're giving optimization strategies across the entire kind of larger teams before they actually even make those changes. We want to be transparent in saying our confidence level and then also what we think the lift will be if you actually make that change.

Speaker C: Yeah, it might be helpful to talk a little bit more deeply about how the LLMs are working. As I understand it, there's two speeds. There's the fast speed, which is the retrieval augmented generation, or rag, where you Ask a question and you'll see the LLM searching the Internet and searching websites and you'll see a bunch of websites going up and down. It's looking for information, usually does a, uh, breadth first search where it looks pretty broadly, see where it can find the information, it finds some hits and then it goes depth first and goes deeper into them and pulls the information back. That seems like stuff that you could influence pretty immediately relatively quickly because every single one of those searches is uh, in real time and it's happening almost immediately. And then there's the slower part, which is the foundational model and the inherent knowledge that the model has in it. And that one, you have to wait till they retrain the model. And so that could potentially take longer. I mean there's fine tuning and a lot of other post training complexity. But is that roughly the way that marketers should think about the two areas where they could make an influence in the immediate term and in the longer term, or is there a better way of thinking about that?

Speaker A: No, I think that's spot on. I think it's a little bit more nuanced depending on the model and the different variations of the model that live underneath it, how they actually pull. So how ChatGPT pulls versus Claude versus Perplexity versus Grok is very different. And so, you know, perplexity is pretty instantaneous, like going deeper into pulling the latest and greatest from whatever you just published. It might take a couple days for ChatGPT or Claude or Gemini to actually have those newly cited kind of sources pop up. And so we give kind of rough ranges. But you're pretty much spot on Rex, there. So that's kind of the split of how those, at a macro level, how those work.

Speaker B: So once they train the model, are they still going back and re scraping the Internet in the same way that a, uh, search, uh, engine or other would be doing? Are they doing that on that regular basis, these ChatGPT and Claude?

Speaker A: Yeah. So Rex, you probably have a better kind of deeper, more scientific view on this. There's periods of time where the model retrains. Right. But then also there's that instantaneous piece that Rex just mentioned. So it's uh, a two kind of layered approach that happens at the same time. But there's periods of time when those models update and kind of retrain or relearn maybe.

Speaker B: I'm thinking about the past. I thought they sort of trained the model. And then like I remember when I first used ChatGPT, it said, you know, you're not going to find any information here in the last 12 months. It was some very funny thing. It was amazing. As long as I didn't ask. But that's all. That's all.

Speaker A: So it really is scraping immediately.

Speaker C: Yeah, so. So that changed about 18 months ago with ChatGPT and then Gemini and then Claude and Perplexity, probably in the forefront of that trend. And what they began to do is they began to create a search mechanism, a search tool. So initially you had to click the button for web search. Now it just does it automatically as part of the thinking model. But that's the change. So that's what I was referring to as the RAG or the retrieval augmented generation. So the thinking models now go through and say, can I answer this question immediately or do I need to go out and do some search? So I have a feed every day about what's going on with the Strait of Hormuz and it gives me Treasuries updates and Brent crude prices and so forth. So I can start my day knowing how much inflation is going to kick in.

Speaker A: One of a kind. I love it.

Speaker B: What a thing to wake up to. There you go.

Speaker C: Yeah, exactly. But you know, it affects a lot of our customers. Right. A lot of people we work with, a lot of the marketers are influenced by that trend. And so that's a great example of one where it's able to give you the most current information as of 7am every morning because of that retrieval. Now, what Justin said was interesting is that some of the sites are caching the information, so it's kind of current for things that it doesn't think is going to move as fast. And that probably saves them some money from having to do the scraping of the page and the processing of all that data in real time. So that's probably where that couple day delay that Justin was talking about might kick in, depending on how topical or how the element has set. Tracking your category.

Speaker A: Yep, totally.

Speaker C: Hey, but when they're building.

Speaker B: Um, we're going to come back to you in a minute. Justin, what's the price now to build a new model? I don't know. Billions, right? Tens of billions? Hundreds of billions. Rex, what are we at? Do you have any idea?

Speaker C: Well, yeah, if you're building it the way that ChatGPT is. But deep Seek sort of threw us a curveball that showed that you could build a lot of this cheaper. But yeah, it is pretty expensive infrastructure for sure.

Speaker B: So what I'm just trying to reconcile my mind and we'll come back here. Uh, but I Think it plays into some of this. How current is the information and how am I then capturing that when an answer comes out? So when they talk about spending billions in order to build a model, to train a new model, what are they doing when they're building that new model and then what's happening to. Then they just build an automatic sort of updating function within the thing, I guess, is what you're saying.

Speaker C: Yeah. So most of the training has shifted, as Caleb was explaining in our decoding AI for Marketing training session.

Speaker B: That was a little plug for some MMA AI training there.

Speaker C: MMA training plug. Well, I don't want to take credit for the work that he did, but

Speaker B: I know who's the smarter one here. It's your 21 year old. So I'm aware of that.

Speaker C: There you go. Co author of my book. Yes, My son. Okay. So what he was pointing out is that two years ago most of the money was spent in training the model itself. Now it's spent, uh, in what's called post training. And so post training is it tries to predict what kind of answer does Greg Stewart want as a marketer compared to what answer does my doctor or my lawyer or my accountant or whatever. And each of us in our Personas want different kinds of output. And so a lot of the money's been spent in that specialization. And so to bring it back to Justin now, I think the key implication and question is as a marketer, you were talking about entertainment being a category you've been doing a lot of work on. And I can see how that's super important because that's much more likely to lead to a retrieval, augmented generation type of AI, uh, thought pattern where it tries to figure out what information's out there, what's current, and it's trying to corroborate that. It's giving you a good answer by making sure that the type of, you know, this is better or this is the movie to see this weekend shows up in multiple sources. Right. I mean, that's kind of a big point that your system can help a marketer deal with is it's not just what's on your website, which absolutely you have to fix, but it's also how you show up in the ecosystem. Can you talk to us more about how to think about that, how to get that right?

Speaker A: Yeah, totally. So entertainment is very interesting. LLMs, this genre, films, music, gaming about 20% of the time. So it's a massive issue. So if you are, for example, we worked with a, uh, a smaller studio that picked up A film from a festival. And the LLM thought it was a violent thriller. And the, the comp for it was John Wick. And so I don't know if you guys have ever seen John Wick, but someone dies like literally every time you blink. No one actually dies in this movie. And for this movie in particular, it was actually not a violent thriller at all. It was more of a dark comedy. And so the real con for it was actually more of a Marty Supreme. So we worked with that studio to tighten uh, up that narrative for that film specifically because one for box office opening weekend, they want people of going in and say, what should I see this weekend? And it fits to their own personal like preferences, right, that it's not a violent thriller that kind of alienates uh, a large part of the population. And then also six months from now, right, when someone's watching a, you know, trying to figure out what their playlist is for Netflix and says, I like Marty supreme, give me five other movies that I like, they want to be able to be recommended during that period of time too. So it's for long term gains. So we worked with the studio to basically tighten up the boilerplate for pr. When you talk to Hollywood, a reporter, Variety and Deadline create three specific YouTube videos that address the hallucination about when this title is actually getting released. Because there was a hallucination about when it was actually getting released, which is crazy. And that happens all the time in entertainment. Too crazy. Why did this actor take this role? Or what are the comps for this film? And so the LLMs then scrape. YouTube can actually have a more educated guess about what this title is, right? Because it's pulling for more authoritative cited sources. And then it sticks. It's much stickier. We were able to regenerate that film in four weeks across paid, owned and earned media.

Speaker B: Hey Justin, how often are you able to go back and actually find the source of the misinformation? I mean, AI hallucinates. So I'm wondering like, is there a source or am I? Geez, no, I just gotta kind of flood the system with more of the facts so it picks up that thematic rather than make up its own answer.

Speaker A: So the flooding of the system is something that I think we should discuss because I think that's something that's happening pretty uh, rampantly right now within the space. And I have a very strong opinion on that. But typically what happens in the LLMs is they scrape the entire Internet, right? And if they are getting information from different sources that isn't Consistent, then it makes up its own decision about what that is. Right. And that's where the hallucination piece comes into play.

Speaker B: Which, by the way, hallucination is a feature. It's not a mistake. Exactly. It's like AI was built to do that. Yep, yep.

Speaker C: Yeah.

Speaker A: So, so basically what we are doing with our clients and agencies, et cetera, is basically saying we have to tighten up that narrative, specifically across these, like, bigger authoritative sources, whether that's Wikipedia, just changing the phrasing, or your boilerplate when you talk to these PR firms. Or Reddit, uh, is a big source too. Right. Making sure the framing is correct within Reddit channels. And so that is essentially how we're looking at it when we do kind of a deeper dive to understand, like, what are the biggest sources. Typically, it's just because there is not a tight narrative that happens. And so then they just kind of make it up what they think it is. The LLMs think it is. But for us, there might be some cases where we could like, pinpoint specific kind of sites that are kind of driving that narrative. And other cases, it's just more general where we'll just like, okay, we need to tighten up where your social handles and your LinkedIn and your site all need to say the specific things, because that will tighten up the narrative. It will pull from those sources and then get recommended more and the framing that you want it to be.

Speaker C: Another interesting observation is that the number of words that you use in describing it can be an issue, because if the AI is doing the rag pattern, it's looking for a chunk of probably about 40 to 60 words that it can grab and use in its summary in the paragraph. And the longer you go, the more it has to also do the step of summarization, and it might lose something in translation there. When you go through some of the different use cases, does the chunk size and the way that marketers should think about how to get the information out there, does that factor into the way

Speaker A: that you design 100%? Yes. So we make sure we work with biotech as well that have very deeply researched topics. Biotech is a very interesting one because the left is anti Big Pharma and anti Insurance, the right is Maha. And then you also have this massive unregulated health and wellness industry. So science is kind of taking a hit across multiple angles. And so when someone goes to LLMs and maybe asks a scientific question, it's pulling from all these sources that maybe are more politicized than like, what the

Speaker B: actual truth is that's kind of scary,

Speaker A: but, yeah, it's very scary. And so, like, biotech and pharmaceuticals, these companies are like, well, how do we, like, actually put what's true in there versus, like, getting sourced from health and wellness blogs? And so for us, when we're working with them, we're saying, like, hey, you need to, like, make this shorter and more digestible for LLMs to be able to read and format it correctly, for the LLMs to pick up with those top prompts that people are asking specifically about those scientific questions that you want to be able to influence. So, yes, Rex, we look at it from the structure, we look at it from the word count and making sure it's readable and understandable across all the major LLMs.

Speaker C: I want you to go a little deeper on that one because I think what a lot of people might not appreciate is that when the AI goes out and grabs that content, it does something called vectorization, which again, we covered in the training series.

Speaker A: Uh, I knew you take this training series.

Speaker C: Yeah. Well, I'm just thinking that some of the people will certainly know this before, but if you don't know these terms, you will want to know these terms because it's part of our life now. In the vectorization, you get this embedding where it's not the keyword anymore as it was in SEO, it's the idea that the keyword represents. And so if you're looking at, like in real estate, you might look at, uh, the idea of walkability or school quality, or these ideas that people care about when they're picking where they want to live. Maybe in pharmaceutical, you have some other examples. So I think that that's what you're getting at when you said we also look at the structure. So I want you to unpack a little bit more about how you think about the process of what are those underlying features that might be pulled out in the embeddings that you need to make sure that you're talking to so that when somebody's looking for something, you're in enough proximity to that concept to where you're going to be cited.

Speaker A: Yeah, so I think there's a couple things there. I think I was referencing, like, schema too, just to make sure you have the right schema on your site. That's the kind of baseline piece to make sure it's in the right schema. If you don't know what schema is, your webmaster and your search team will definitely know what that means. The other piece is just like topical authority. Right. So making sure that if you are talking about a specific topic, you are coming at it from authoritative of framing. So like, if you have research or if you have a unique point of view, that is adding value, that is something that the LLMs will favor when they actually cite it.

Speaker B: Hey Justin, how do you actually know that you've sort of accomplished like what becomes proof of performance and all this?

Speaker A: Yeah. So, you know, I think you've, you've mentioned this a couple times on previous podcasts as well as like, we are in the early days here just in terms of measurement for the entire channel. Right. And so it's very similar to early days search and social and mobile. For us, at our core, we wanted to make sure we were driving measurement for the actions that our clients are actually taking and proving that the actions that they were taking are actually driving visible lift. Because I think a lot of the, the industry is kind of flooding their, their clients with optimization recommendations with no clear understanding if this is actually going to drive any type of meaningful lift for them.

Speaker B: And so how does Ambrose know that better than somebody else, by the way?

Speaker A: Yeah, so we, at our core, we're not an AR rapper. We built a brand knowledge graph and it's very similar to Google search graph. In the brand knowledge graph, we can map over brands, entities, creative assets, politicians, names, influencers, et cetera. Once we map those over to our system without going too far into like the special sauce, but we have a city map of how the Internet looks at that specific entity, we then track it over time. And when I say a city map, we can see what are the biggest influencing sources online that drive visibility and have a higher correlation to that specific brand. We track that over time, which, uh, gives us a baseline which unlocks that predictive capability for us both at the macro and micro level. So we know, we run simulations saying if you make this change to this publisher or this content, et cetera, we know that we believe or we have a high confidence interval that like this will actually be able to drive X amount of lift for you or drive this type of QPI that we're looking at. It was important for us to like, not just flood our customers with optimization noise without actually saying, before you even do this, this is what we're being transparent here. We think it costs this much, we think the confidence level for this is this high and we think you'll have this much lift. We then track it on our end. Once the change is actually implemented, we'll show the actualized lift that is driving because we can see the sources coming through for the citations and then we relearn. Right. So we get smarter over time. Every single change that happens, we understand kind of what the impact is by that specific client, but also at the vertical level as well.

Speaker C: I love that you're doing that. It's great that you're thinking about the prediction of what we think happens and then what actually happens. One of the challenges of this and search earlier is that it's something that people proactively do. So unlike advertising, where you can create a experiment with a control group and exposed group and you're deciding who to push, what message you're receiving, somebody's intention to find a piece of information, is there a way in which you think that the industry needs to evolve to be able to do the control and treatment type of experiments? It seems like in order to do that we would have to get the cooperation of the large language model, the anthropic. Well, Anthropic may not play this game, but ChatGPT probably will. Gemini probably will. Perplexity probably will. Same thing that we've been talking about on this show a bit with, uh, paid advertising and how that pay to play might factor in here. You know, Microsoft with Copilot has probably done the most advanced experimental work here where they have a certain percentage of people turned on, where they're seeing what happens if they include ads, what happens if they don't.

Speaker A: Yep.

Speaker C: How does that evolve? What should happen next?

Speaker A: Yeah, I mean that's the uh, kind of million dollar question there. But I would say we will need some sort of cooperation from the LLMs directly to do that, like test first control piece. Unless there's ways to think about it with like shutting off a model and kind of looking at it across other ones. But then it kind of opens up another can of worms of saying, okay, well this model is different and operates differently. And like there's just, I think it's, you'll get scrutinized as you know, Rex, probably. But once again I think we're just like, we're in the first inning of this and I think for us, like it was important for us to put, put a stake in the ground and say, okay, we're, we're at least trying it, right. We're getting closer to this understanding that like it's not perfect, but at least we're getting somewhere. I would say like the majority of the space right now is just showing you the problem right there. There's like performance dashboards and showing you how you're Showing up, but never actually telling you how to change it. And if they are telling you how to change it, they're either giving you a ton of optimization recommendations, never really tying it back like I said, or there's this concept, and I mentioned this before, but just of AI generated slop. And so basically they, they push out a thousand different variations of, of articles or websites, etc. And for me, it's creating an AI slop ecosystem. And so we are not doing that. And uh, I think you guys both, we've been in the industry long enough. Like if you do that for long enough, you'll actually get delisted from search. Right? Because it's. You're spamming the entire Internet and that, that is not a space you want to be in. Let's. That's like Internet jail and you're not getting out of that and has massive implications to your business if that's the case. So something I think for all your listeners to know if that's something they're thinking about entertaining, like, don't do it because it will have negative impact for you across different areas of your business.

Speaker B: Geez, Rex, I'm shocked that he's suggesting there's shenanigans in Martech, aren't you?

Speaker C: I mean, I'm glad you brought it up, Justin. I hope that people take that seriously because I do think that it destroys the whole ecosystem. Totally, totally in the long term for the consumer experience. And to your point, at Some point the LLMs are going to do the same thing that Google did, which is heavily penalize people that try to game the system. System. Yeah. Justin is creating a good dialogue, uh, for you, Greg, which is how do we move the industry forward from the first inning to the second inning? Right?

Speaker B: Yeah, yeah, yeah, Justin. So listen, you know, you're obviously closer to Rex and I are. We pay attention. We've talked to other people on the podcast and you know, like I said, my board's ask is lean in. And in fact I actually, I can announce here we are going to do. MMA has done a thing, uh, over the years. We call it the great debates and we pick a topic. So we did great debates around identifiers and that was a big topic when Google was going to cut off cookies at one point, then ultimately didn't happen,

Speaker A: but multiple times that was about to happen.

Speaker B: 1.

Speaker A: But yeah.

Speaker B: Oh my God, the multiple. Yeah. Jesus. I'd never heard anybody, uh, cry wolf so many times as that one was. But that was funny. Uh, whatever. I think the question that I would have is like what do you think we need to know what's the research maybe the MMA could go do independently to help markers understand this space or get into it effectively. Do you have a sense of that? I don't know if you've given that much thought yet.

Speaker A: I said on the IAB AI board we're writing the kind of rules and guidelines for AI visibility. Um, so I sit with.

Speaker B: I used to run the iab.

Speaker A: I know I mentioned that. Yes.

Speaker B: I love that organization.

Speaker C: First director of research for the iab. Sure.

Speaker B: There we go.

Speaker A: We're all IAB boys. Okay.

Speaker B: Yeah, yeah, yeah. Every, everything starts with the ib.

Speaker A: So like you know, I think everyone is realizing there's an issue right? Like that that's this is the uh, the wild wild west. And if we don't put some regulations and at least some guidelines out there, like we're all going to be living in this AI slop world and it breaks a lot of different parts of our businesses. And so there's a lot of issues with that. I think from a research standpoint it would be very interesting to see like how long those tactics that are happening, which is basically like the, the AI generated push to publish concept which is going across a bunch of syndicated publishers, how long those actually sticking into the LLMs for recommendations. And my bet is probably not even a week. Right. And so you're just basically constantly having to publish these things over and over and over and just filling the ecosystem with more AI slop. The other piece that we're really thinking about too, which no one really talks about is just the publisher side. You know the publishers, well, some are

Speaker B: talking about they're suing the LLMs but. Yeah, go ahead.

Speaker A: Yeah, yeah, yeah, for sure, for sure. Yes, that, that is definitely top of mind. Yes. Vera publisher. I'm sorry. There is an opportunity there for them to play that's not like getting basically scraped from the LLMs and then also AI overviews just crushing their organic traffic pretty drastically. Where the brands actually want to get cited more. Right. And as these LLMs get more advanced and not looking at those long tail blogs that are AI slop driven that these higher kind of authoritative cited sources will become more valuable for brands to work with to make sure that they get cited into these LLMs. And so I think that there's something there where like from the buy side we could help the publishers with like an incremental revenue stream where it's like more of a safe way where the publishers are not just giving away all of their content to the LLMs for free.

Speaker B: Will you propose that the LLMs fund the publishers for access to some of that content? Is that what you're.

Speaker A: Yeah, I think so.

Speaker B: And also respect the intellectual property?

Speaker A: Um, yeah, yeah, totally. I think there needs to be a little bit of reckoning on that. But I also think there's also another kind of work stream that will be interesting to see how this all plays out. But is the brands, they want to be cited through these LLMs with the authoritative cited source of Penske Media, et cetera. Can they actually fund the publishers in some capacity too to make sure that their brand gets published more often within these LLMs?

Speaker B: Didn't Google and Facebook fund some of the publishers? Didn't they create sort of revenue for the publishers? I mean, I guess obviously they were.

Speaker A: I think they're working on it.

Speaker C: Yeah, they did something earlier, in an earlier iteration, Greg, that you're talking about. I think Australia or some other country had certain laws and. And again, it was sort of a similar thing with Facebook. They were reposting a lot of articles and there was a complaint in certain countries that it was taking revenue. So I think you're right. This is sort of a different version of M. The same, um, movie. It's a sequel to it, I suppose. And Justin, you're so right. We have to figure out as an industry how we come together and create a healthy ecosystem. And I really appreciate that you are playing in this area around AI visibility in a way that's responsible and trying to create something that's good for the brand. And I just think you should be commended for that. So thank you. Good for the brand, good for the consumer, good for the ecosystem.

Speaker A: That's why I got into it, because I just, I saw there was so much of the space like, uh, you know, it's funny, like some of these companies are popping up. They're like, you look at the background of the founder and they're like, we're in crypto. Like 6 months ago, you're like, you didn't have any marketing expertise. Now you're in Geo aeo. It should be a red flag, right, at the iab.

Speaker B: I mean, I can say this now. We were very nervous about email companies based in Florida. That was a thing that we decided not to let him be members. That was a immediate M disqualifier or whatever. That's been a long time.

Speaker A: Good to know, Good to know.

Speaker B: Antitrust now.

Speaker C: But.

Speaker A: Yeah, yeah, yeah.

Speaker B: Sorry, you were saying?

Speaker A: No, no, I mean, just for us, like we, we, we want to help our, our advertisers and marketers think about long term visibility versus like short term hacks. Right. And so it might take a little bit longer, it might be more expensive in some cases to do these optimization strategies for us, but it's really more, I don't want to say ethical way, but like it's uh, a way for us to really be thinking about long term visibility that will actually stick so they don't have to constantly be republishing these AI slop articles.

Speaker C: Yeah, and there's a really legitimate point which is that marketers should be taking some of the signal as feedback on how they can improve. Because look, if you are getting a recommendation that isn't your brand and you can see what the features going back to the schema and what is the underlying features that the AI is picking up on, maybe you do have a weak point there that you need to work on with your core product and then really trying to create the advocates with your customers who want to talk about your brand. Because a lot of the AI is feeding on organic information. And yes, you can influence it, yes, you can be part of the dialogue. That's what I think. Justin, your company is showing with your prediction of where you could make a difference. But there's also the bigger issue which is how do we actually make our company better? How do we make our product better? How do we make our experience better? That shouldn't be lost on people. How do you bring that back to marketers? Or is that a little bit outside of your scope when you're telling them how they're showing up?

Speaker A: It's interesting. There's one piece of it which is basically how you're showing up, how you're getting recommended. And also by the way, we can get you recommended, but if you're not getting recommended correctly, that doesn't matter. So we need to make sure we clean up that piece of it. It. You know, there's another kind of like area that we're, we're working towards too is just like are uh, the LLMs talking about your brand in a compliant way? Because that is a massive issue as well. Think about like insurance, pharmaceuticals, et cetera. Like they, they want the, like the output to be actually, you know, more compliant. The other piece that I think has been really interesting for us to see as we go down this path is like the LLMs are just like, they've got such insane data for marketers and brands and studios to use for more upstream behavior. So like really almost like strategy for product Launches. Like I've uh, been asked a few times by studios like can we use this data for like green lighting movies? Right. Because absolutely. It is a whole other piece of the puzzle. And so we're, we're working towards like figuring out what that looks like for more upstream behavior as well. But it's kind of like the full gamut. Right. It's from the strategy piece to the execution piece to the governance piece which is across the board. And this whole space is just going to be rapidly evolving very, very quickly.

Speaker B: Hey Justin. Melissa. As we kind of close up here and stuff, I don't know, any, any news you want to share with the listeners? You don't have any, you know, you're not announcing new funding or anything, right? There's no uh, I don't know, anything come up that the listeners should pay

Speaker A: attention to or we'll be announcing new funding pretty soon. So super excited about that. We'll be able to give kind of a, a few behind the scenes updates on client front and we've got some great logos behind us, some great design partners that we've been working with with Stagwell's been in a fantastic partner to us. Uh, we work across all subsidy agencies for AI, ah, visibility for their search plus platform. And we didn't really talk about it but like we've been doing predictive capabilities at the macro level so we've been predicting box office revenue. So we're not looking at any search, social energy, traditional tracking. We're looking at share prompt and our AI model and actually getting to a really like high level of accuracy for that. We just hit 98% for the latest Super Mario movie. We're getting pretty high accuracy for smaller like uh, unknown IPs which is pretty amazing. We are not a box office prediction company but we're using this predictive capabilities across all of our priority verticals. And the point there is not a plug for us but just for all of the listeners, for all marketers to say like hey, everyone needs to wake up. Like it's important. Yes, it's just to get recommended but there is like a predictive aspect to this of like saying how you're showing up actually does equate to real world business outcomes.

Speaker C: That is really exciting. Yeah. Mainly because I think that when you show that even though the activity of AI compared to search is still small, it's growing very quickly when you can see it predictive of what the overall world event is, that's a big deal.

Speaker A: It's huge.

Speaker C: Greg and I in Our first book, I think we were predicting the box office results for Constantine is when the movie Hitch and others were out. And uh, we were using survey results to do it at and then seeing could you move it with Internet advertising. So if you're now able to do that with AI search engagement and have a good prediction, we definitely want to hear more about that. And then if you can then shape it and then bend the curve, I guess I'd love to figure out how we do some good experiments to really validate how much you can then change a trajectory. That's, I think, really exciting.

Speaker A: Every weekend we're doing a box office prediction and once the numbers come in on Monday, we our model retrains. Right. So we get smarter every single weekend. Uh, and like I said, we're not trying to be a box office prediction company, but it's just an interesting use case, say everyone, like we want to put a stake in the ground. How you're showing up actually does equate to real world business outcomes.

Speaker C: Yeah. And then you do automotive sales or you do, you know, all these other travel, uh, events, et cetera. So I think that that's a big, uh. I started in entertainment as well because it's just such a very fast cycle time. So I think it's cool that you're starting there as well. So I love that.

Speaker A: Cool.

Speaker C: Well, thanks for coming to the show, Justin.

Speaker A: I really appreciated the conversation.

Speaker C: Uh, okay, that's it for this episode of Decoding AI for Marketing. I'm Rex Briggs.

Speaker B: And I'm Greg Stewart.

Speaker C: Be sure to catch all our episodes plus subscribe, follow rate and review in your favorite podcast app or check us out on YouTube. You can get transcripts and more information on our website. Decoding aiformarketing. Uh dot com.

Speaker B: And if you're so inclined to learn more about Marketing, Media alliance and some of the work that we're doing to sort of advance marketers understand about AI, then please reach out to either me, greg@mmaglobal.com or just mmaglobal. Com. Thanks everyone.