Pod Insights Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast Fan out analysis, Local rank checks in AI About this episode 80% of AI-cited sources don't appear in Google's top results. Karl Kleinschmidt, founder at Data Marketing Group and 18-year SEO veteran, has developed systematic approaches for LLM optimization across enterprise-scale local SEO programs. The discussion covers fan-out analysis methodology for mapping user intent beyond traditional keywords, multi-LLM data collection frameworks using Claude projects and Gemini validation, and local rank tracking strategies that account for geographic personalization in AI-powered search results.
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
Transcript 0:00 Speaker 00 The Voices of Search Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast? I Hear Everything delivers podcast production, growth, and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit ihereeverything.com. Welcome to the Voices of Search Podcast, a member of the I Hear Everything Podcast Network. Ready to expedite your company's organic growth efforts? Sit back, relax, and get ready for your daily dose of search engine optimization wisdom. Here's today's host of the Voices of Search Podcast, Jordan Cooney.
0:53 Speaker 01 AI is changing what it means to create content because in AI-powered search, it's not about who writes the best article, it's about whose content gets cited. In fact, recent research shows that 80% of sources cited by AI systems don't even appear in Google's top results, which means ranking isn't the same as visibility anymore, and that creates a new challenge for SEO teams. So the question is, in a world where AI summarizes everything, how do you create content that actually gets chosen? I'm Jordan Cooney, and joining me today is Carl Kleinschmidt, founder at Data Marketing Group, an 18-year SEO veteran focused on local SEO, large-scale data systems, and
1:36 Speaker 01 LLM optimization. Today, we're talking about fan-out analysis, local rank tracking, and why LLM optimization is really about intent, not keywords. Carl, welcome to the Voices of Search Podcast.
1:51 Speaker 02 Thanks for having me back.
1:53 Speaker 01 Hey, Carl, it's so much fun having you on the show. I mean, obviously, we've known each other for a long time, but you have
2:01 Speaker 01 Constantly been innovating. You don't stop innovating and this is an industry that's going through probably its biggest transformation. I just remember like a few years ago, you were showing me some like heat mapping with ranking data. And now we're looking at a whole new world with LLMs and how LLMs work. And you know, LM optimization is really all about intent. And I want to know from your innovator mindset, what does this actually mean and how should individuals put this into practice?
2:37 Speaker 02 I think the biggest change is you stop thinking about what people are searching for and what they instead what they want. It's no longer the right approach to say, okay, they're gonna search. I want a red running shoe. They want a shoe that helps them while they're running, that looks stylish, their favorite color is red. And so you need to figure out what the intent is of what people are actually searching for. And then you can figure out how do I best write the content to fit that intent.
3:11 Speaker 01 And this reality of like mapping to like what users are looking for and what this intent is. Is this a purely data-driven thing? Is this a survey thing? Is this a guess and trial and error experiment thing? Like where do you get that insight? I think that's one of the things that, you know, historically as SEOs, we used to do these like content briefs and we used to do all this other stuff, right? But now this is changing a little bit and I'd love to know your perspective on like, where do you get the knowledge, the data source?
3:43 Speaker 02 So my approach has been to basically, I don't know if teach is the right word, but basically, provide LLMs all of the data around a company and then say, I want to figure out.
3:59 Speaker 02 What are the sub-intents of a query, right? Most people call it fanout. Um, but the interesting thing is there doesn't seem to be a clear definition of what fanout actually is, right? Like, how you get sub-intents seems to change. Every person does a little bit differently. Every L L M does a little differently. Um, a lot of the researchers that are really going deep. Um, I think they're getting a better and better idea of how uh, fanout actually works, but I have not seen any amazing tools that are like, this is how you consistently do the perfect fanout.
4:36 Speaker 01 Right, I mean, and like fanout is like one of these components of intent, right? But I mean, even like taking a layer back, like let's before we get even deep into fanout like, you gotta know what the prompt is, right? To some extent to even get to the fanout. And like, I think a lot of the prompt itself has a significant amount of context to intent, right? Like, if you don't know that, then you you're gonna have a really hard time kind of breaking down intent. So like, where where do you start with that? Is it is it understanding your audience? Is it understanding your users? Like, where where do you get some sort of knowledge on that on that front?
5:11 Speaker 02 So I take it from two approaches. The first one is who am I promoting? What are they known for? And what do L L Ms think of them? And so I have basically built these skills and tools that go at the very highest level. So if we look at, let's use Allbirds, they're in the news. Um, are they the best shoe company in the world? I think most people would agree, no. They are not the most relevant best shoe company for everyone in the world. Are they the best running shoe company? No, they don't really do running shoes. They do like lifestyle shoes. They do really good waterproof ones.
5:55 Speaker 02 So now we're at the waterproof level. And then you see what do L L M think? Uh, think about, are they one of the best waterproof shoe companies? Currently, probably not, because they're now mostly gonna be known for A I. But you figure out what am I currently? And then you have to make this decision, do you go up? Do you go, okay, currently I'm the best waterproof shoe company? I think I do make the best running shoes. And so then you have to write the content to convince everyone and get the citations and get, do all the authority building to become the best running shoe company? Or do you go wider in the vertical? Do you say, I'm the best waterproof shoe company?
6:40 Speaker 02 What are the uses for waterproof shoes? Like, what to do when it, what to wear when it rains? Maybe I partner with a raincoat company. Maybe I do some research about um what are the best waterproof clothes? I just go that way. So you have to first make the decision, are you going wide? Or you're gonna try to move up a level?
7:01 Speaker 01 And I mean, I think this is where it gets interesting, right? Because then when you go down into the fan out queries, that kind of starts to give you a little bit of that kind of a fragmentation of what a user is trying to do as it relates to the primary prompt, whether it be branded or it might be discovery based or informational based. And and and so from from what you were sharing right from the beginning, like, moving beyond just keywords, going into prompts, but then more importantly, fan out queries, like these fan out queries are essentially keywords. So are we moving beyond keywords? What what what is the value of that data, which is fan out queries?
7:41 Speaker 02 I I think I treat fan out queries sort of like keywords, but then you have to do in-depth intent analysis on each one. Right, like traditionally, if you have a set of keywords and they're tangentially related,
8:00 Speaker 02 you have to do some intent analysis, but you don't have to stop, like do a complete intent analysis. For fan out, there's a chance that the uh you are not relevant at all to the fan out. And then you either have to say, okay, I I I'll be relevant for sixty percent of the fan out, but this forty percent I'm not relevant. Or if you're not relevant for seventy percent of the fan out, that's a pretty good signal, you're in the wrong space. Um that L L M's believe the intent in this specific query, the intent breakdown is so different from what you think it is, then some something went wrong and you need to revisit your core assumptions.
8:43 Speaker 01 So then, in this, in this relationship of trying to get to what a consumer's question is, how should we be thinking about the mapping then? Is the mapping based on our business objective and then the relationship of those intents? Like, I'm trying to sell shoes and so there are certain buyer or transactional intents? Or is it really the inverse of trying to understand how the broad market is trying to do things? How all users are trying to do things and where the volume is? And then working your way towards that intent? I think one of the biggest challenges that we have today, that search volume was able to give us in the SEO world,
9:28 Speaker 01 was it was able to give us some direction around the life cycle, the customer life cycle, like how someone begins the journey to the end of the journey, right? But today, with, within these AI models, what we're essentially analyzing is answers. And so sometimes we're just getting to the conclusion, but we're not truly unpacking where the user is and their buyer journey. And so my question is, as we look at this data, how are we doing the mapping of fan out query information or context to the end customer or user we're trying to reach?
10:06 Speaker 02 I consider fan out like a, a thought exercise of trying to figure out where are users coming from? And especially, right, like with AI overview and
10:22 Speaker 02 also just pinging the LLMs, you can test theories. So for instance, like I was working on a project to figure out, does the location of where someone is searching from influence the answers that LLMs gives? Specifically in the auto repair space? Cause I was interested in researching, like,
10:47 Speaker 02 If I live in Phoenix versus if I live in Minnesota, will the answers be different for something related to AC repair in the car?
10:57 Speaker 02 And so you can do these context tests where you're like, if I give you the context that I live in a specific city, or if I give you the context of something else, you can get pretty good answers um for what the L L Ms, what makes the L L Ms change their answers in the expected ways. And so when you're doing these fan outs, you are just doing like assumptions. And then you have to validate that the L L Ms think the same way that you do and the changes come, follow your assumptions. And then you have to use all possible data from actual user interactions, right?
11:41 Speaker 02 We have G S C impressions to get some data. We now have um Bing's search console to have some data there. So we're starting to get a little bit of data, but that is still the hardest part. You have to keep on simulating possible intents, collect data about what is working, what isn't. And we, that is still the hardest part. I don't think anyone has cracked yet. How do you actually connect those two? But you just gotta keep on trying and see what moves needles from the data that we have.
12:14 Speaker 01 Hundred percent agree and and and and one of the things that I think is really interesting about our conversation so far is that we haven't actually answered why this is so valuable. Like, I I don't think fan out queries are going anywhere. Like, I fundamentally believe that L L Ms and A I models, including Googles, whether it's A I mode or whatever other features and capabilities they release, are going to be um connected to how search is traditionally done and they're going to continue to ground themselves and use RAG as a foundational way to complete certain prompts or requests from users in an A I driven discovery process.
13:00 Speaker 02 I would agree with that, yeah.
13:01 Speaker 01 Okay, okay, and so if you agree with that, that's where I was gonna, I was gonna go in that direction. So you answered that question. Thank you, Carl. If you agree with that, that this isn't going anywhere, what is the, what's the biggest thing that marketers should be doing then? Because to your point just now, like, there isn't a lot of great evidence. There isn't a ton of data sources. What should we be doing in order to best utilize the fan-out data that we either can capture or are capturing?
13:34 Speaker 02 So the, the number one thing is as early as possible, collect all the data that you have. Make sure that your analytics platform has a filter set up for LLM queries, including conversions all the way through. So you know, are you actually moving the needle? Then ingest the search console data via the data import, not just the UI. And then for sure, create a Bing Search Console account. So you, because that is the best data we have right now, uh, from all of the LLMs.
14:09 Speaker 01 So then like, Carl, what else should you be doing here? Like, what are, what are the other activities here?
14:13 Speaker 02 So the main thing is, um, do as much fan-out analysis as possible. Pick, every level of fan-out is going to get harder and harder and more specific and seemingly less valuable, but also gives you a further understanding. And treat every single one of your major clusters of keywords as its own fan-out, um, process. So I, I have recently started working with some customers and they have, they have multiple major clusters. And so I have been struggling with this whole idea of like,
14:57 Speaker 02 at some point, they start overlapping. And then you have this really interesting exercise of which cluster do you, uh, which fan-out do you go with? What is the primary, right? Like, if in a pure LLM world, which we don't live in, you would probably create some sort of content for every single fan-out activity. And, but then there will be duplicates because, right? Like, if you sell running shoes and...
15:29 Speaker 02 Hiking shoes, you're still gonna ask about sizing and do we think the intent around hiking sizing versus running sizing is gonna be that different that you need, you need content for each one. Maybe, but then the next level is men versus women. Is men sizing versus women sizing different enough that they each need their own intents?
15:53 Speaker 01 Right, yeah.
15:54 Speaker 02 And so you keep on going down and you do it for everything that's important. All of the things that make you money, you need a cluster and a full fan out for. And then you have to overlap them all and see where is the unique intent? And you're, for all of those fan outs, you need the data from the L L Ms, the rank checks and the A I overviews on a regular basis to see what is changing.
16:24 Speaker 02 Go ahead.
16:25 Speaker 01 Let's talk about these, these clusters, because I think this is actually the, the next big theme here, like, for, for those of that are listening here, like thi- thi- this is a, this is a very deep conversation as to where and how you're analyzing this L L M data. And I don't think that a lot of teams are doing this level of analysis today. I think so many people are still hung up with the bells and whistles of visibility. But unpacking these fan out queries is foundational to understanding what consumers want and need. And even more importantly, connecting them back to your content or web strategies. So tell me about clusters and how you're connecting clusters to what a team is doing, what your actions are,
17:08 Speaker 01 and how you're investing your resources against them.
17:12 Speaker 02 So let's imagine we work with one company and they only have one cluster. They only do one thing. You basically have a topic keyword at the top and then you have one level of fanout. I, by default, I do ten. I think my Claude skill says eight to twelve, but most of the time it ends up being around ten. For each one of those queries, then I um, I need to figure out what is the intent behind this? Is it? Are they looking for more details? Are they trying to make us specific decisions? Where in the buying process are they and who is searching this? So when I do fanout analysis, the questions that I'm trying to answer is. Why are they searching?
17:58 Speaker 02 Who are they? Um, what triggered them to search? And what type of content are they looking for? And like, if I say, how do you, if someone searches, how do you tie your shoes? They want. They, they got a new type of shoe for the first time and they want a step-by-step list of how to tie your shoes. But if you are searching, um, what is the rubber made of in these shoes? You want informational content about. And you, you, you're a random person that bought these shoes or researching these shoes. You're interested about the makeup of the rubber. And you read about something or you thought of something and now you want more information.
18:42 Speaker 02 And so if you collect those pieces of information for every single fanout, you have a relatively good understanding of what you need.
18:50 Speaker 01 Yep.
18:51 Speaker 02 And then you have to look at what.
18:56 Speaker 02 You overlap it with GSC data. So basically what I do is I pull in all the GSC data and I um, connect each fanout with semantically related GSC queries. And so then I have, okay, what is my average ranking impressions clicks for every single fanout?
19:17 Speaker 01 Right.
19:18 Speaker 02 Then I can be like, oh, here's a complete blind spot for me. I have zero clicks, zero impressions for this thing.
19:30 Speaker 02 Let's look more into it. Am I just not the right person to talk about it? Do I have no content for it? What is going on? Um, and so then you basically do that for every single one. And then you have to make the decision, what is the um, what is the right way to answer that intent? And so like, do you need to cre- create a whole new blog post about this? Do you need to add an FAQ question? Do you need to refresh a page? Do you need to add more citations? Um, whatever it is to improve and better fit the intent of that specific fan out.
20:09 Speaker 01 Yep. All right, let's, let's shift our conversation because there's another main topic that we want to talk about when it comes to AI discovery. And that's local and how local consumers are seeking out information. And for me personally, I, I, I love this because uh, this topic, because this is something that, that's near and dear to me. I've been using ChatGPT quite a bit lately to do local based discovery. Um, I've been doing it for restaurants. I've been doing it for uh, different uh, service providers in my local market. I've been doing it for hotels uh, and lodging. And one of the things that I think is really fascinating is that it is a far
20:53 Speaker 01 less capable experience than traditional search, right? What, what goo- the empire that Google has formulated around video, multimedia, maps and map packs and map directional like information. Um, review and review concepts that are highly local and, and tailored.
21:17 Speaker 01 It's just pretty remarkable and it, and it's hard to replicate. Um, one of the most frustrating things is when I need a map experience in that discovery process in ChatGPT, that is awful. It's bad and it's frustrating. So Carl, give us, give us a little bit of like the lay of the land, my, my own personal beefs aside. Like what's happening in this shift around local? Specifically, let's talk a little bit about AI overviews or AI mode versus LLMs and what's happening in, in, in different models? When it comes to local discovery.
21:55 Speaker 03 Time for a one-minute break to hear from our sponsor, Previsible. So you're looking for SEO help, and you've got a couple of options. You could start replying to spam from agencies that claim they can get you to rank number one on Google. You can pay an hourly rate for a consultant who will inevitably nickel and dime you with hourly charges. Or you can work with a cookie cutter agency to quickly launch a strategy-less project with low success rate. None of those sound very good, now do they? Well, that's where Previsible's integrated consulting model comes in. Previsible draws from a collective 40 years of SEO and digital marketing experience to unlock your organic growth opportunities. They build custom solutions that combine strategy, technical expertise, content, and reporting to effectively operationalize SEO for your business.
22:40 Speaker 03 Previsible's four-stage approach ensures that your SEO programs thrive by starting off with a "strategy-first" approach. Then they support you in your efforts to create quality content, help you identify technical issues, and most importantly, to work with your cross-functional teams to integrate your SEO strategies to make sure that your SEO budget actually drives results, not just your agency's bottom line. So join brands like Yelp, eBay, Canva, Atlassian, Square, all who rely on the SEO consultants at Previsible. For more information, go to previsible.io. That's previsible, P-R-E-V-I-S-I-B-L-E dot I-O.
23:23 Speaker 02 Sure, so I'll give a little bit of a caveat. I've now done a lot of research in three to four verticals. I don't know every single vertical in local. And the data seems to indicate that some verticals, there are AI overviews everywhere. And in other verticals, they are almost nowhere. And so whatever vertical you're thinking about doing local SEO for, that is the number one question you need to answer at the beginning is, are AI overviews there by default? Because they will change the interaction so frequently. I agree with you, from a local perspective, the experience is really bad. And my theory is two things.
24:08 Speaker 02 They're still really bad at figuring out where you are.
24:13 Speaker 02 And so, right, local, uh, Google Maps knows where you are really well. You have, you are giving them some crazy permissions.
24:21 Speaker 01 Right.
24:22 Speaker 02 The L L Ms don't have that, especially on desktop. If, in my experimentation so far, if you give the L L M enough context about where you are, they're pretty good. But no one searches like that, right? They're not, you don't search, find me the best pizza near, here's my home address. You just, Google has trained you for so long to just say, near me.
24:49 Speaker 01 Right.
24:49 Speaker 02 And then the L L Ms will not be good at that.
24:53 Speaker 01 The creamy part is that Google also knows what style of pizza you like. So if you like thin crust or New York style or Chicago deep dish, it already knows.
25:01 Speaker 02 Exactly. And the second part is that local websites are probably some of the least structured, um, least rich content websites out there. So the L L Ms are guessing a lot and relying on business directories a lot. Like, some of the verticals I've looked at, Yelp is just the source of truth. Like, if you look at who's getting cited, it is just Yelp, like. And if you're not in the top three or not top five at Yelp's, or if you're not paying Yelp, you're not showing up in AI overview or in the L L Ms.
25:45 Speaker 02 Which is really scary.
25:47 Speaker 01 I mean, it is scary, right? I mean, it gives, it gives a lot of pause to like, what's gonna happen near term versus long term? When it comes to local discovery. It, it also though, uh, I have to ask this question because I think fundamentally, this is like the big bet. Right? It's, it's the, you know, the Sam Altman effect, which is like, all this is gonna change really, really, really fast. And I'm betting long on me winning faster than anybody else. So like, we just need to wait a couple more months and all this local stuff's gonna work really well?
26:21 Speaker 02 Um, I don't, the location portion, yes.
26:26 Speaker 01 Okay.
26:26 Speaker 02 The rich content about where locations, what, all the information about a local business in an easy to understand method is.
26:42 Speaker 02 I uh uh there is still a lot of work to be done and so far I have not seen anyone do it amazingly um at scale.
26:53 Speaker 01 Yeah, yeah.
26:55 Speaker 02 And the right local SEO has been separate from organic SEO for a while now. But I think local SEO is coming towards uh organic SEO. And that is going to bring in a lot of complexities that I don't know if the local SEO people are going to be ready for. Um, there's right fan outs at a local level are even crazier.
27:24 Speaker 01 Right, right. So.
27:26 Speaker 02 Yeah.
27:26 Speaker 01 So we're we're now getting into this like weird vortex, right, which is like what users want need versus what the technology's capability is. And and and and it is quite quite uh friction, right? Like, I mean, Google's had a twenty year game at trying to help. You know, consumers find answers to these to these very local and and often very specific types of requests. And so, you know, one of the one of the key concepts that that comes to mind with with local is that local really meets at this crossroads of personalization. That that's what makes local work so really so well.
28:06 Speaker 01 And I'm curious to get your perspective on how personalization impacts local intent. How personalization impacts the result sets? And is this just one of those areas where Google's just gonna win faster than everybody else? Because they have a lot of that context?
28:26 Speaker 02 So like, I'm still of one of two minds, I'm still not a hundred percent sure on the right approach. Like, on the one hand, personalization has been in local forever.
28:38 Speaker 01 Mm hmm.
28:38 Speaker 02 And the traditional local SEO approach for personalization has been I can't control where people are, so I'm going to just figure out where they are likely to be and target those areas.
28:53 Speaker 02 L L M's are taking that to but are taking personalization to a whole nother extreme. But you still, as a marketer, can't really do anything about it. It is not a positive ROI exercise to start going around asking individual people their pizza options. So that you can write them the right pages on it. So if they search for their pizza, you show up. You are still taking guesses at what are people looking for? Where are they? And what is the intent behind their search? It just gets way more complicated. You have to do way more research into every single keyword.
29:38 Speaker 02 And the data that I have collected so far seems to indicate that. We are going away from. This is a type of business to most local businesses have five to 10 clusters within them. And every single one requires a cluster, a full fan out analysis. And in local, you then have to also talk about geographic intent. Like, how far are people willing to travel for pizza? How far are they willing to travel for a doctor? You have to do that for every single cluster in your business and then make an informed decision.
30:20 Speaker 02 Write different content about the intent of each different cluster.
30:25 Speaker 01 And so when you, when we, when we start to see this like future uh for, for local. Is one of the thing is, is something that we should be thinking about today? The change in measurement? Like, should we be measuring local in a different way? Like, right now, I see a lot of noise around measuring calls. Right? Like, that the best way to measure local is on the number of phone calls. Right? Um. I'm curious to get your perspective on that. Obviously, it differs based on the types of businesses, right? Like, you know, some businesses, it's great to have phone calls. Other businesses, it's like, no, I just need people walking in the door. I own a coffee shop. No one needs to call me. They just need coffee between like six AM and two PM.
31:12 Speaker 02 So Google Business Profile interactions, I think, right, Google is still the number one way people are gonna find you. Gemini heavily relies on um Google Business Profiles, AI overview features. Um, Google Business Profile um in their, in their sections. With some exceptions, I have now saw, found some verticals where Google is not featuring GPPs. Which, that's gonna open up a whole other thing where if people can't interact with you, they have to search your name and go to your website. So branded has always been important. It's probably in some verticals, it's gonna get even more important.
31:57 Speaker 02 Um, in other L L Ms, they're gonna send you to the website. Um, or the people are going to Google your name. So, um, I don't think measurement changes drastically. Tracking phone numbers have always been a thing. Um, I'm of the opinion that almost every single local business should have a phone number and. Right, I don't know if AI phone services are at the point yet where you can use them if you don't wanna answer the phone. But you should have a phone number that at least says, I'm at this address, here's my phone number, uh, here's my. How you order, you want me to text you your domain and stuff.
32:36 Speaker 01 Yeah, absolutely. I mean, this is a, this is a good segue to kind of our, our last theme though here, right, which is like. I, I fundamentally think that things are going to be shifting quite dramatically and, and you've been a huge proponent of building custom tools and building your own kind of technology. So as we think about the shifting landscape, as we think about discovery changing from keywords. Um, to prompts and changing from rankings to uh, what are, what are answers. But eventually, what I think all these L L Ms wanna do is they want to get to a conversion. They want to get to an event. They wanna be able to complete the action. How should we be thinking about building custom tools, the platforms that we use?
33:20 Speaker 01 And, and the data that we are collecting uh, in order to best uh, inform ourselves about doing any form of digital marketing.
33:29 Speaker 02 So my recommendation is always you should use all of the L L Ms. Assuming you you have the budget for different things. In my experience, each of them is good for their own thing, right like Claude is really good at. Collecting a lot of data and making decisions based on that, right? Uh, but the way I do it is for every single one of my clients, I create a Claude project. And I have all of these skills that basically ingest in a standardized way all information about my clients. I create the projects and then it has as much context as possible.
34:08 Speaker 02 Claude, on the other hand, though, is really bad at. And I still don't have a great word for this, but like. Bringing down concepts to a single word, which is like. I'm sure everyone has had this L L M um experience where you're you're trying to get it to do something and it. It is just not understanding what you're trying to do. It keeps on trying new stuff and it's just not getting it.
34:34 Speaker 01 Hundred percent.
34:35 Speaker 02 That's where Gemini.
34:36 Speaker 01 Yes.
34:37 Speaker 02 Gemini is amazing there. Um, if you give that to Gemini and say like, hey, I need a word. Or a phrase that explains this, this is my goals. If you then give that phrase to Claude, Claude becomes way better. And so I have, whenever I run into those situations, what I do is I just. Ask Gemini um and then. Give J uh Claude the the phrase and say, oh. I meant this and then it's like, oh, yeah, now I know what you mean and just does it the way I wanted to. And so.
35:12 Speaker 01 Yeah.
35:12 Speaker 02 That's been super fun um. I use Claude code a lot, but Codex, I've seen a lot of people say Codex is getting really good. I am mostly using Codex as a validation tool. Even the free plan is enough where you're like, I created this Python script, critique it. I I have found that Codex is one of the harshest critics uh of your code. Uh, where like Gemini loves to do like, this is the best thing I've ever seen. I don't think I've ever had Codex tell me that. Uh, it's always like, here's five things you could be doing better. And then you obviously have to make the decisions, but um.
35:54 Speaker 02 I like putting as much content into Claude project in the most structured way possible. With standardized like read me's, what's this engagement like so on? And then also building more and more repositories of data that L L Ms. Um, are really good at reading. I for the first time yesterday tried. A content brief specifically written for an L L M instead of a person. And I was really surprised how good they are. If you tailor it to them. And you make it a markdown file with some X M L in it.
36:35 Speaker 02 You can give it to like the cheap models and they do a pretty good job.
36:41 Speaker 01 Okay.
36:41 Speaker 02 And so I think we're all still trying to figure out, but that's currently where I'm at.
36:46 Speaker 01 I mean, here's one of the challenges, right? Like I think. Uh, buying something off the shelf, it it feels a lot more comfortable, feels a lot easier to do, right? And and and all the noise, all the hype right now, Carl is like, hey, you can replace your whole toolset. You don't ever need to buy this tool or that tool to do tracking, ranking. I feel like there's a new A I visibility tool that gets launched every single thirty five seconds. And so the functional question I want to ask you, because you've been a builder for a long time. You've been an advocate of bu- building your own tools and data. Like, where should the average S E O, G E O marketer today start? Why should they be building their own tools?
37:28 Speaker 01 And should it always be fully adopted inside of an organization?
37:34 Speaker 02 I don't think so. The.
37:39 Speaker 02 The right, in my opinion, G E O is still eighty percent S E O. And.
37:48 Speaker 02 I would not, if you are doing SEO right now. And you're just trying to improve click performance. I don't think you have to change anything drastically right now. I do recommend to everyone that you give the L L Ms as much context. As you have and ask it questions and brainstorm. You're not replacing any tools, you're not building tools, you're just like, I have all of this data. I am trying to do this. Any recommendations? And you're going to find that they have recommendations for you that will be helpful. Especially as you give it more and more information. Um.
38:33 Speaker 02 I, I have tested a bunch of A I um. Tracking tools. But in the end, I just go back to just pinging them directly via um. A P I or I build a little, I pay for a little bit of a middle layer, something like open router. Where you're just like, model switching is easier and you can have some additional things on top of it. Or. And so. Most of the tools that I'm still using is. Give me the. Most reliable data for the cheapest price possible. Um. I'm, I just want to consume data, I just want the L L Ms to consume data, but.
39:17 Speaker 02 L L Ms also hallucinate, so having a source of truth for. I don't want Claude to try to get SERP data because it's really bad at it. Sometimes it can get right, but if I go to uh Scale Serp. Get the actual search results with everything nicely formatted and give that to Claude. It's gonna be way better than it's trying to get all of this information, so. I wouldn't recommend you start replacing tools. You don't even have to build tools. Just start testing.
39:49 Speaker 01 Yeah.
39:50 Speaker 02 The smaller plans are cheap enough that it makes a hundred percent sense to just test, give it data, see what it thinks.
39:59 Speaker 01 Yeah, I mean, in all of this noise about like AI replacing jobs and taking people's jobs, I just, I just keep reminding myself, I'm like, this is gonna be interesting, because over the next five to 10 years, we're not necessarily gonna replace jobs. We're just gonna create new jobs, right? And these new jobs are gonna be very complex. They're gonna require some technical depth to be able to do them. And in order to scale, you're gonna, you're not just being a human in the loop anymore. You're probably like a human who's evolving the way these models function or adapt to the way a business is doing things, right? And I don't think that that's simply gonna be, uh, at this point, gonna be replaced with AI. So I want to transition, Carl, to my favorite part of our pod, which is the lightning round.
40:44 Speaker 01 So I'm gonna ask you five questions, uh, from our episode today. You're gonna give me a thirty to sixty second response back on each of those, each of those questions. You ready?
40:53 Speaker 02 Yep.
40:55 Speaker 01 Okay, Carl, what's the most overrated SEO tactic right now?
41:01 Speaker 02 LLM manipulation, where you keep on pinging the LLM and argue with it and say, no, I am actually the best.
41:09 Speaker 01 I love that. I love that. And, and why is that overrated?
41:14 Speaker 02 Because, because I am confident it will give you short-term results, but it is, it's like CTR manipulation, right? Like, you search the same thing over and then click on the same results. Maybe in the short term, you get results. In the long term, you're not bringing your client any, uh, good results.
41:33 Speaker 01 Absolutely. What's one signal that matters more in AI discovery than Google Search?
41:40 Speaker 02 What triggers the event? I I think like in SEO, if you wrote good content and had all the important words, you could rank. In an LLM, if you don't have are fundamentally addressing the concern of why they are searching this thing, you are going to struggle to get mentioned or cited.
42:10 Speaker 01 Okay, on a software or product perspective, build or buy? What do most teams get wrong?
42:20 Speaker 02 So foundational data, I think we still need to buy. But they buy too much reporting. Reporting is so unique to you. You should be building your own reporting because you know way more about your business need than any SaaS company out there.
42:36 Speaker 01 Yeah, I love that insight. That is, that is one of the things everyone wants the canned reporting, but in reality, that's just, that's just so inherently unique to each business and team in many cases. All right, Carl, one mistake companies make with AI-generated content.
42:53 Speaker 02 Thinking they're the best prompters out there. I I I see people say like, I have this amazing content delivery system where I have built. It's eight stages long and I have perfected every single prompt.
43:08 Speaker 02 That might work for now and it might work for the next month. But people that write really good, unique content with unique data, unique perspectives and unique expertise in the long run are going to beat you. No matter how good you prompt you are, and there's always a better prompter out there.
43:28 Speaker 01 If you had to focus on one thing for AI visibility, what would it be?
43:33 Speaker 02 I think content brief creation. I I like the.
43:39 Speaker 02 SEO people are not content writers in most cases.
43:43 Speaker 01 Right.
43:44 Speaker 02 And in this LLM world, content creation is such an important part, right? Like, the amount of inputs that go into a content brief are expanding weekly, right? Like, these numbers are a little extreme, but it used to be a content brief is two pages. Then it got to 11 and they keep on getting longer. And you you are giving the content person so many inputs that their job is getting harder and harder, and therefore you need better and better people to write content. And so, brief creation, giving them that content in the the context, in the easiest
44:27 Speaker 02 to understand um way is a superpower, because if they get all that context, they're going to write amazing content.
44:35 Speaker 01 Yep. Okay, that wraps up this episode of the Voices of Search Podcast. A huge thank you to Carl Kleinschmidt, founder at Data Marketing Group, for joining us. If you'd like to contact Carl, you can find a link to his LinkedIn profile in our show notes or on thevoicesofsearch.com. You can also visit his company website, karlk.ai. If you haven't subscribed yet and want a daily stream of SEO and content marketing knowledge in your podcast feed, hit the subscribe button in Podcast App or on YouTube and we'll be in your feed every week. Okay, that's all for today, but until next time, remember, the answers are always in the data.
Previous The most important channel for AI visibility Next The most overrated SEO tactic right now Page updated November 14, 2026 · transcript and summary are machine-generated