Discovery has shifted from links to conversations, and that’s changing how customers find, evaluate, and buy. In our first podcast, we sit down with Patrick Harrington, MetaRouter’s head of AI and machine learning, to unpack generative engine optimization (GEO), what it is, why it matters, and how marketers can align content, consent, and speed to win in an LLM‑first world. Drawing on experience at Walmart, Block, and Workday, Patrick lays out a clear path: treat the first mile of data as your control plane, capture consent at creation, and design information LLMs can confidently retrieve mid‑conversation.

We break down the real differences between SEO and GEO, from keywords to the geometric structures models use to understand meaning, and cover how to tighten provenance, validation, authoritative domains, and timestamps to make content both trustworthy and discoverable. Along the way, we tackle hallucinations, guardrails, and why keeping models close to your data preserves sovereignty while enabling real-time decisions that boost ROAS.

Finally, we look at the broader market: LLM platforms, retailers, brands, and the advertising ecosystem are all in flux. The advantage goes to teams that can pass actionable context across channels without losing control—durable first-party IDs, consent-aware event streams, and omnichannel signals that reflect real shopping behavior. Optimization must start in the first mile, where decisions about routing, personalization, and protection happen in milliseconds.

If this helped sharpen your GEO strategy, follow the show, share it, and leave a quick review with one takeaway you’re acting on next.

Read the full transcript below.

Dom Burch: Welcome to the First Mile. This is the podcast where we explore how identity, privacy, and AI are shaping the future of customer data. I'm your host, Don Birch, and thanks for joining us on episode one. Today we're diving into a hot topic: generative engine optimization. We've moved from SEO to GEO, but what does that actually mean for marketers? How are large language models changing how people search, discover and buy? And what impact will that shift have on the first mile of data, the foundation everything else depends on? Well, to help me unpack it, I'm delighted to be joined by Patrick Harrington, MetaRouter's head of AI and machine learning. Now, Patrick's career spans modern treasury, block, workday, and earlier in his career, he helped build Walmart's media data platform, shaping many of the ideas that inspire MetaRouter today. And now he's helping companies unlock even more value from their trusted first party data. Okay, let's dive in. Patrick, welcome to the first mile podcast.

Patrick Harrington: Thanks, Dom. It's great to see you and really looking forward to the conversation today.

Dom:  We've got to start with the name of this podcast, right? Because it's a term that you coined. What does the first mile mean in practice and why does it matter so much?

Patrick: When you look at how any consumer, you and I, interact with anything from Spotify to Asda to Google, et cetera, when we interact looking at things, searching for things, listening to things, the data associated with those activities between the business offering those services and the end consumer, that data is created in what's called the first mile. And the first mile is unique relative to say the last mile or last kilometer, right? Relative to where that data ultimately goes. Many of us get retargeted on Instagram, elsewhere on the web based upon what we do with the economic incentive to try to bring folks back to complete their transaction. And that data also goes into companies like Spotify's large data arsenals, their large databases, their data lakes to then mine that data to extract different patterns, to turn around products, capabilities, to help buoy the experience and hopefully all boats float with incremental revenue. So first mile is really that moment where identity anchored data is first born and then ultimately where it goes. So high level, first K, first mile is largely that impetus for the moment of creation, we'll call it, the genesis of the data being created by whom. And MetaRouter tends to facilitate a unique position in market by occupying that interstitial medium in the first smile and then helping companies route where that data goes. So first smile.

Dom: And that's important, right? Because I guess a lot of companies are generating data. You know, there might be retailers that are capturing customer data. There might be brands that are like, you know, people are hitting their site, whether they're a toothpaste brand or whatever it might be. And unless they're appreciative of the value of that data and how they organize it, and then who they give permission to to use that data, potentially they're not one, maximizing the value, but also, I guess then there are issues around security, around compliance, around actually, you know, holding that data in a way that customers would expect them to.

Patrick: Yeah, there is our European friends were first out of the gate with respect to acknowledging that uh there's a concept of residency, sovereignty, and ownership associated with data and a privilege of the end business company, et cetera, who is processing that data generated from behaviors of the end consumers. So GDPR in the EU, uh, the California Consumer Protection Act, here stateside, there's more legislation to move things forward. This is a really nuanced but key point. There's all these litany of different data solutions, privacy, compliance, identity management, consent management that happens once that data, call it message, think of an SMS, where that data flows down to some appropriate destination. What consent did me as the end consumer bestow upon that message about how companies can use it? And the typical uh, you know, a la cart offering here is anything from their analytics, how many people are on my site, how many people bought, or advertising purposes, or increasingly data going into their own AI models and so forth. And so it's easier to, it's hard to unbake a cake. Let's put it that way. So when things are in the last mile, that message has already been sitting there, time takes away. Maybe my preferences have changed, but in the moment of creation, that concept of consent. You know what? I'm happy to bestow uh analytics purposes, but I don't want to be advertised with this or have my uh message, so to speak, go into AI models. That's captured in the moment, which then affects where it can go. So you avoid the unbaking of a cake phenomenon and are just left holding the raw ingredients, and you're left to do what you want to do with them.

Dom: You know, it's interesting to me, right, because I think back to when I was back at Walmart, what, 14, 15 years ago, and I had this, I had a challenge, right? Because I used to work in the PR department and I cared passionately about the quality of the content, the communications that we put out. And then this kind of SEO gaming world came along where some of the digital marketers were like, we just need to create content because then Google will find it and that traffic will come to our site. And it always felt like a it's like a zero-sum game to me. Like, yeah, you might get an initial win, right? You're gonna get some more traffic. Then they're gonna turn up to your site and go, This is just Ipsy dorum, right? I mean, it mean it's meaningless. You so you found me and you brought me in, but now I'm like, whoa, really? This is the experience you want to give me? I'm gonna jump back out. So we're moving now, we're sort of on the cusp, aren't we, of moving from SEO to this, I don't even know if I like the phrase, but GEO, but you know, generative engine optimization. I mean, one, what is that meaning? And can you optimize, you know, a large language model? Surely that's kind of not the point of having a large language model today.

Patrick: Those days of SEO were fun, right? They were very much cat and mouse. Uh, you know, we we we share that background about 14, 15 years ago with Walmart, where we were a large search engine optimization shop, right? Like, why not piggyback on folks searching for goods and services and having the appropriate landing page content to drive that transaction? And the world is changing though, right? Like as you alluded to, the GEO moniker, which I believe some Silicon Valley VCs coined um, like they like to do, is really how to think of aligning the representation of goods and services such that the AI labs, OpenAI, ChatGPT, Anthropic, etc., where discovery is happening, how do you make that piece of content representing the goods and service most thematically aligned with how natural language is being spoken about the intent associating with trying to find something? And so it's a little different than how Google in the past would reward or penalize, call it SEO gamification. Um, this era it's a bit different and much more mathematically nuanced because you're trying to align your content, reflecting the, let's say go back to the Walmart example, the item of interest, say an iPad, right? Where what type of incrementally unique information of people speaking about uh different facets of the of that device or that item uh align to the mathematical structure of how LLMs process text and conversation. And you're trying to basically reverse engineer this black box in a much more nuanced way to reflect incrementally new valuable signals that the LLM can process in the context of that conversation mid-flight and retrieve it and then put it in front of you, hopefully relative to others, and then ultimately gain that click and then ultimately that transaction and that relationship with the end uh customer.

Dom: And and there was a time, wasn't there, like where Google kind of saying, actually, listen, we want to reward sites that are quick, that actually can serve that search. So somebody's looking for an iPad, you know, we want to get them to somewhere that can sell them that iPad. And if we know enough about that Google customer, we might know they're price oriented, so we're gonna send them that direction more often. And those keywords, there was a relationship, wasn't there, between retailers, brands, and Google saying, let's collaborate so that we can optimize this and make it better for the person searching. Because, you know, to some extent, Google were competing for that search traffic, they did a really good job, right, of being the best in the world at it, um, particularly for a Western audience. Now, we're not at that point yet with LLMs and the ChatGPTs of the world, but like every day I turn on you know my phone and there's another announcement that ChatGBT are gonna extend how they allow people to stay within that environment, right? Whether it's looking for a holiday and then being able to buy that holiday or looking for an item like an iPad and been able to do that sort of inflow, but we're not at the mature stage that we were previously with keywords and and working with the likes of Google. How do you see that developing? And perhaps Matt, you know, and that's a crystal ball answer, right? So, how should it develop so that technologies and retailers and brands can work better to give customers ultimately what they want, right?

Patrick: This is the trillion dollar question. And I I've spoken with some leaders who who are a generation above me and have been kind of seeing the advent from an executive role at the at the hyperscalers circa, let's say 15 years ago up through the present. And so their perspective is very unique here. And the perspective is we're in a fundamental inflection point that it that is a function of every other past innovation in the past 25 years of the evolution of commerce on the internet. And that wave is hitting this beach in this era of AI. And what do I mean by this? When we look back to our days at Walmart, where desktop traffic was mainly the source of uh you know people searching at Google, coming to Walmart, buying things. And then there was that shift to mobile, where conversion rates were different. We have mobile app, mobile web, and every team is scrambling to try to maintain revenue parity across uh you know surface mediums. Same thing is happening now. The difference is that discovery that where people are interacting, searching, having a more richer conversation than call it the three or four words that they put into a Google search bar, which is quite shallow and narrow relative to a more long-form back and forth conversation like you would be having with an expert, a la the LLM. Brands and retailers need to maintain how to represent our content to play ball in this richer, more expert-based uh medium. And so we're moving kind of through the cat and mouse games of kind of that transient with desktop web into mobile app, mobile web, deep linking, et cetera, now into this new paradigm where you have a much richer structure of a signal characterizing the intent of the end consumer of what they want, which can be influenced by the back and forth of the LLM, which is reaching into its arsenal of content to put in front of that consumer in the context of their richer inquiry. Travel's a great example, right? Where travel, it's hard to pin down the unique uh taste in a Google search bar, but it's much easier. And many of us have done this with ChatGPT and others to where you know you're looking for a northern Italian trip of a particular aesthetic and a particular thing, and you got kids and this and that. And it's hard to kind of put that into a search bar without you then doing all the onerous work to then boil it down to a subset of possible trips for your family where you want to make sure that goes off okay, especially with young kids, relative to maybe uh true north is a little bit easier to establish in the LLM domain. And so companies right now are trying to figure out how do we maintain parity and then growth and not uh cede control to some other entity or that relationship associated with that guidance. And so this is the transient we're in right now. And what's really exciting from our standpoint is because we're front and center in helping supporting our customers, major public retailers and brands to navigate that transient.

Dom: Yeah, and and I guess it's sort of you know, circling back, that's why it's important right now to figure out how do you organize your data, what data do you actually have? What's the best way to segregate it? Can you get tags off your site so you can speed up your e-commerce, all that kind of good stuff, right? And and once you're in a better shape, then as this thing starts to mature, you're gonna be better equipped, aren't you? To then go, right, how do we start to optimize then for GEO, right? Because marketers will be sat at home now scratching their heads, right? And you know, good luck to them because it was only 18 months ago we're going, oh my gosh, everyone's moved to TikTok. They're not even using Google now to go, where shall I go in Tao Mina when I'm on holiday in Italy? It's like, I'll just go watch a TikTok because I trust that influencer to say, show me the 20 best restaurants in Tao Mino or wherever it is, right? Here we are then. So LLMs. What what do you see as the opportunity though, right? Because yeah, we're in flux and it's not going to figure it out yet, and you know, every day is a new day. But what kind of opportunities does GEO offer to marketers, do you think?

Patrick: This is what I hear from every single CMO. There's a lot of anxiety about this as well. There's a lot of fear, and there's a lot of uh unknown unknowables, and it's a hard mathematical problem, right? And when you want to try to optimize any business or any product or any engagement uh funnel that is associated with people laying on the site and they click and they buy, blah, blah, blah. You want to optimize that funnel. That's what digital commerce has really uh given to us. But you and in order to do that, you need to really work back from first principles. And the first principles of how LLMs work are not ubiquitously known across the general public or even practitioners in call it the marketing content optimization domain. And that's really what GEO is. You are trying to optimize content to align to a different multi-step, multi-conversational paradigm. It's almost like ordering a, you know, a meal of multi-stages versus ordering, call it a cheeseburger at the counter, and that's off you go, right? There's a more rich back and forth associated with how people characterize the product. It goes beyond just small little reviews, but ultimately how you represent and optimize the images, the text, uh, the different modalities call it, of input signals to align, to call it a geometric representation of that content, which is how LLMs work. They work in a space of taking inputs, text, images, et cetera, voice, you name it, and representing it in a geometry, in a space such that birds of a feather flock together. Similar things are in that similar region of space. And that sounds kind of meta, and it is. Um, but GEO is really trying to programmatically, mathematically align and optimize that content to how these LLMs receive and process that intent such that when the conversation is happening, it's not incongruent with what the conversation is happening. It's call it collinear with how that intent is shaping in that back and forth.

Dom: Okay, no, I like that. And I like being able to visualize. I like the idea that you know the kind of content that ends up on my result from a conversation with ChatGP, ChatGPT, has all been sort of gathering, coercing around a space like a hotspot. And then as us content creators, I'm thinking like, how do we become one of the trusted places that it pulls in, right? Because, you know, increasingly over the last 18, 24 months, ChatGPT is just so much better now. If I ask it something, I feel more confident that the sources it's pulling in and the relevancy of those sources, I feel less compelled to have to click on the links and just check, was that like in the last month, the last six months of last year? Is it a credible source? And by the way, it hasn't just made that up, right? You know, I think this sort of notion that it's still hallucinating feels a little bit last year, but marketers and retailers will worry about that kind of stuff, right? They'll be thinking, what's the quality of this result? And not only how do I need to shape up and be ready to provide that content in a trusted way, but also be thinking about my PR team. Like, get me my content also on trusted sources again. It's almost like we've gone full circle. Like, third-party endorsement of some of my content is going to allow those LLMs to pull in from multiple directions a result that helps me be more relevant in that answer. I mean, that feels like I mean, I'm saying that from a PR perspective, right? That's my old hat on. So I would say that.

Patrick: No, you're you're not wrong. I mean, Google in the late 90s solved the quality problem, which was called PageRink. And the high-level call it uh motivation there is more links from higher quality publishers that reference a particular site help Boo We call it the quality juju of that domain of interest. So if you got you know New York Times and BBC pointing at a particular piece of content, you know, by proxy, when you kind of expand this over the internet, this is why Google crawls, they could develop a quality score called PageRink and then use that to guide the quality of the results. And so for us older folks like you and I, remembering what search was like prior to that, it was pretty bad, right? And so it they did solve a problem of quality. What you're talking about now, the hallucination, in a way, is a quality problem, right? And there are different, arguably a little bit more mathematically complex ways than what PageRink actually is in addressing when content is generated, call it the truthfulness of that content. There are ways under the hood for these LOM providers to, or any practitioner who is fine-tuning models on their own domain data set, et cetera, to impose guardrails, call it, around the confidence of the output and also the state of the model, and if it's in a trustworthy call it configuration. And so there are different ways here, but you're hitting on a key point here. When the world is moving to not just a static piece of content and not just personalization, but pure subjectivity of what is that content, that experience for me, separate from you, Don. And that those are capabilities from like call it a generative image, multimodal video. We're seeing it with like Sora and some of these video apps and things of that nature, where brand and marketers are going to be like, well, if I that's a new knob and lever for me, can I just generate the page in that moment to optimize that experience for this individual based upon the tone, the conversation, the structure of this high intent, high uh longitudinal conversation that goes beyond just three or four words in a Google query, right? And so we're we're kind of like trying to analogize the world these days to how the world call it approached discovery 20 years ago, and then commerce emerged. We're in the same position. It's just that those knobs and levers are not yet clear because we're in that transient today.

Dom: I like the phrase guardrails though, because I think that does bring us back a little bit to let's talk a bit about MetaRouter and the tech and how it actually works, right? Because I think putting some rules and guardrails in, and then I love this idea that the way that we do it with a customer is like we then give them the capability, right? It's still their data, but it's on their side, like before they give it away, and our technology is kind of like embedded on their servers, right? So it's like giving them the tools and the skills and also putting some framework around it that just means it's better organized, it's safer, it's compliant, all those good things. But right, that's my kind of like layman version of it, without getting too technical, just help people understand it. Because if we sit at home thinking, okay, this sounds good, it sounds like we need to work with these guys, but like help them get it. How would you describe it to somebody? I want to say down the pub, but maybe not down the pub.

Patrick: A pendulum has swung in the past few years where companies that made their my they had their servers in the basement, right? They had their servers in their own data centers. And then there was the push to go to the cloud where you don't own the hardware per se, but you have call it a corraled region where your data lives and who can access, you know, this and that. So you kind of have a security posture. And many data processing, many marketing plays would offer goods and services like SaaS-based applications from their call it region of the cloud that end customers would then have to send data to. That pendulum is swinging back because the C-suite is wisening up on the residency and sovereignty of your data. And GDPR is a component, CCPA, but it's also just a risk, an exercise in risk management. Where now those SaaS-based applications, those call it uh infrastructure as a service capabilities, which is what MetaRouter is. We are customer data infrastructure that occupies the first mile. No longer are we just hosting MetaRouter's own capabilities, of which customer A, B, and C route their data through. Our software offering all the different bytes that comprise our what we do is deployed in their environment that they own, they control, and they their data never leaks. And this is really a key. And I think the advent of Q1 2023, when ChatGPT coming off the holidays, it was the talk of the town of every corporate leader. And I was there while during my time at Block and Square and working with a really uh impressive leader who kind of occupies that uh exact perch who's seeing kind of the advent and growth of the public clouds. Where how do we make this a reality? Well, I for me to use ChatGPT, whether it be for employee productivity, I have to send my data to their models in their cloud. Well, when you have call it operational efficiency, revenue capabilities, and then on one hand, and then on the left, you have privacy, compliance, all things risk. They're like, whoa, whoa, whoa, whoa, we're not comfortable with this. This is now really kind of had over the past two, three years, corporate uh behavior has swung it back to where it's in my kingdom, it's in my moat, in my security posture, my data's not leaving. I can bring models to my data versus sending it out of the kingdom. And so this is really what MetaRouter does. We enable our end customers to own their own destiny, to have a dog in the fight, to help them receive and interact in this modern medium of back and forth with Google and the typical traffic referral that happens, but also the new advent of discovery, which is happening in these multimodal LLM-based mediums. And so it's in the kingdom, you own it, it's a car in your own garage, right? And it remains in your in your home.

Dom: I like that. Okay, well, I am gonna force you to get the crystal ball out or put your visionary specs on, right? So climb that ladder and look over the horizon for me, Patrick. But where do you think, you know, let's I'm not gonna push you too far, but like next 12, 24 months, where do you think Geo will be? Where is it taking us?

Patrick: There's gonna be a push and pull where, and I'm already hearing it talking to our you know, CMOs, where there is anxiety around losing that relationship with the end customer and the margins that are associated with that relationship versus a peer fulfillment play. Um, I think there's hesitation to seed that relationship, that experience, that intent funnel to just let commerce happen within an LLM type experience. Um there's no free lunch, right? And they want to maintain, they always have wanted to maintain that relationship, whether it be with how the physical store strategy uh of the 70s, 80s, 90s, etc., where do we put the next store, right? And you know, Walmart being an example, very rigorous with the looking at forecasting traffic patterns, growth patterns, domestic migration patterns, and whatnot. Same thing with the web. You're trying to optimize how many people come, you know, how do I optimize that experience? How do I get a larger lifetime value out of that relationship? Well, nothing changes in this new media, right? And so I think over the next 24 months, there's going to be an interplay between call it the three persona types, which is one, the LLM, like an open AI, ChatGPT, uh, the retailers who are downstream of Discovery, travel providers, anyone, you know, where there's intent-based goods and services search, and then call it the advertising ecosystem of where these brands and retailers have an economic incentive to spend money where there are eyeballs to then bring folks back. And so I think there's going to be an interesting push and pull in this three-legged stool, so to speak, of these persona types of how to flex call it the budgetary influence on spend to drive towards a desired experience. And I do not see retailers simply rolling over and seeding the entire experience to a within LLM-based experience that they don't know. Right. And I think now it becomes the GEO is how can we receive this higher structured intent? How can we pass it along when folks don't convert? The same way that, hey, this person uh abandoned a cart and you see some relevant content on Instagram. What is the higher, richer signal representing that intent? And so we're going to see kind of this con this concept of passing around context between data islands. Because that data is not just flowing away. Privacy compliance legislative principles are trying to lock down that raw data, but at the same time, for these entities to operate efficiently, they need some signal, right? And I think we're going to see that play out over the next uh call it 24 months.

Dom: I hear you because I as I think as a customer, right, if my loud language model starts to become a bit clunky, or I feel like the offer that it gives me at the end of my inquiry isn't seamless for me to then go get it, or it became, you know, or I don't trust it. I'm gonna start reverting back out of there to do what I want to do, which is the shopping in the first place, right? So it's gonna have to work for the customer, it's gonna have to be better than the alternative. And the alternative's pretty good, right? Because if you go direct to a retailer, it's a pretty good experience, or you're not gonna go there again. If you go to Google, it's a pretty good experience, right? Because they've been doing it for 20 years. So I think that's really interesting.

Patrick: You know, I think I think retailers' brands are have to rethink how they represent experiences and content and think beyond, call it classic websites, which really haven't changed that that much past 25 years.

Dom: What do they need to do in this changing environment?

Patrick: Maintaining that relationship with the end customer is mission critical. And that is going to involve uh a modernization of how they interact with their end customers that maybe go beyond the classic websites that really, even you know, with the quote unquote personalization of the past 15 years, they all relatively look similar-ish over the past, you know, two decades. And so they have a significant economic interest in maintaining and growing those relationships. And so it's gonna come down to how do we interact with our end customers? What are the different modalities now that we should be interacting with our end customers? Uh, Omni channel is a strength, it's not a weakness. The LLMs are not crawling with robots around in stores, at least yet, right? And so, how do you leverage that online, offline, holistic understanding of the customer and interactions to be more proactive and intelligent and less one size fits all, so to speak, like a classic item page on some website, right? And so I think there's an opportunity uh for these companies to lean in and leverage their omni-channel presence in particular as a strength.

Dom: Help me understand the difference between the first mile and how you're organizing that data and the last mile and why it's so important.

Patrick: This is such a key point. There have been tens of millions of dollars per year spent on last mile, call it data assets. These are the usual suspects that you hear in market who are there to churn through petabytes and terabytes of data to try to draw some uh establish some business logic at the end of the day, or train some machine learning model, uh, some lifetime value model, or uh audience suppression model, or build an audience segment to then push to Facebook or to TikTok. The first mile, though, is interesting because that data, that experience is going on mid-flight. We're not waiting for the plane to land. We are in transient. And there is a question you can ask to support the business, to support our customers. What type of business logic can we perform in that first mile that affects the experience, the routing of that data, the suppression of campaigns towards higher return on ad spend? And this is really critical, especially when we start looking at customers who have retail media network. Revenue and very aggressive growth targets over the next few years, which are high margin, high uh you know, growth items for them. And so there's a degree of enrichment, business logic that we can perform on these data messages while they are in flight versus when they're at rest in the last mile, respectively. And that allows us to optimize over time that the last mile assets cannot do. They're late to the dance, they're late to the party, they're hours late, and that intent decays with time. So the first mile is critical, not just as a passive pipe of plumbing, but as a central nervous system that affects kind of the experience and then determines what should happen with it towards better experience, better economic outcomes. And this is uniquely appropriate. We're seeing this with customers with retargeting efforts, real-time RMN efforts, where we're able to route or suppress where that experience should go outside the walls of the kingdom, a la Instagram, TikTok, et cetera, Pinterest, you name it, all by deploying machine learning and other business logic in that first mile. So it's a really interesting distinction, and it's a really important one because you're leaving money on the table otherwise.

Dom: That's so good. I love that. So, Patrick, listen, I love how you frame that. A first mile isn't just plumbing, it's the central nervous system of modern marketing. I'm having that. I'm having the first mile and I'm having that phrase, right? It's where data meets business logic in flight, in real time, not sitting at rest in some data lake. And that's especially critical for retailers and their retail media networks, where every millisecond matters. It's the place where you decide whether to root or suppress, personalize or protect. And that intelligence can't live in the last mile assets. It has to happen upstream, right at the source. And that's what makes the first mile so powerful. It's not just about collecting data, it's about understanding it, acting on it, and doing so with trust, consent, and speed. Patrick, thanks so much for joining me on the first mile. And to everyone listening, remember the smarter your first mile is, the stronger everything downstream becomes. But for the time being, Patrick Harrington, thank you so much.

Patrick: Thanks, Dom. Appreciate it very much.

To learn more about MetaRouter, reach out and let's chat!

‍

‍