Vollständiges Transkript anzeigen (9.455 Wörter)
Ross Hudgens: Josh, you are our AI search wizard. You came on once, we talked about AI SEO, it went very well, so I figured, let's juice it up with the in-person version of this. So, welcome back.
Josh Blyskal: Thanks for having me, man. Absolutely, what a pleasure.
Ross Hudgens: So, you put out some recent research that I thought was a good place to start. There's been some studies floating out the inner webs from various people, seen it from Kevin Indig, who we were speaking glowingly of right before this, other people as well. How often should people be tracking these things for accuracy? What's your, what's your take on that?
Josh Blyskal: I'll put it simply and say that if you're trying to track a wide breadth of prompts, the the relative value of repetition is much lower. So if you're going wide, you don't need as much repetition. If you're trying to track a small number of prompts, repetition will help you. So it really depends on the prompt portfolio. We did some research. We found that the difference between prompting once a day and like 10x a day was only .25% difference in visibility. A tiny, that's not an actionable gap. And we did a second test and found that the bigger change that will actually change your visibility score day to day or moment to moment is the prompt portfolio you're tracking. Right? If I change my prompts, my visibility score is going to change way more if I expand the prompt set or, you know, decrease it, right? That's like, think about it as like, is that, right? So, intellectually, honestly, like, ProFound can run repetition, and we do for some customers. But there's tons of people out there that are paying for more repetition and more executions than they need if they have, say, 500, 1,000, 1,500 unique prompts that they're tracking, because at the end of the day, pragmatically, marketers are trying to understand a wide subset of the market. You can do that in multiple ways. Um, it's a methodology question, really.
Ross Hudgens: Yeah, no, that's great. Uh, and I appreciate the, I assume you guys can charge on a model of the more you run it, the more expensive it is for people, right?
Josh Blyskal: Happy to. But we, we try to give, we're trying to go highest signal with the least number of executions so that you can do more with those with those executions that you do have, that you can only say you can only run 100 executions a month, I'd much rather you you get a wider sample of your your market than having, you know, 10 little samples 10 times over. Right. It's important. I mean people, people will do different things though. Everyone has a right to their own opinion.
Ross Hudgens: Yeah, makes sense. I think of um in classical SEO there'd be people with maybe just one head term that's just like the bulk of their their world. Yeah. That would be for that person, it sounds like run that, or I guess, what is, what do you think the right frequency for that person with let's let's call it classical SEO had one page with 90% of the search volume?
Josh Blyskal: Wow. My my gut there is that you should run it between 5 to 10 times if you're literally tracking, I mean, daily? Yeah, a day, if you're tracking literally one head term, one prompt. Let's do it multiple times. If it truly is like that for your business. Okay. But what I would say is if you're Nike and you're doing shoes, maybe there's actually many query fan outs or many prompts that you want to track. And in that case, let's be intellectually honest about the different ways our audiences are actually searching, and if it's bigger than just what's the best shoe company, then, you know, you're probably actually looking at more prompts, even though you might think in traditional SEO, hey, it's actually one head term. That world doesn't exist. That might actually be 200 prompts. Depending on how granular you want to look at it. Like, you should go with the prompt volume first. So, you see 90% of the search volume around this particular page or you're seeing 90% of your referral traffic from ChatGPT on this one page, it's not enough anymore to say, oh, what's the keyword for that page? And then that's our AI search keyword. Not to be this is not a plug by the way, but like with our prompt volume tool, we can see in the wider world we have 1.5 billion prompt volumes, we have 1.5 billion conversations. We get the user's submission, but then we also get the model's response and with the response, we see global citations. So we can see, we can if you put a URL into ProFound, we could be like, your URL was cited, we can back into what prompts a URL is actually being cited for. So if you're like, hey, Josh, nike.com/shoes-men uh is our number one page by far, I could be like, listen, okay, there's actually like 20 main clusters that this is getting cited for. Here are the topical clusters, here's the distribution. If you say, hey, um it is actually uh adidas.com/shoes-women that's cited, I could be like, oh, you know, this is very different than anything I've ever seen. This is actually there's only two or three prompts that are really uniquely showing up, two or three fan outs. So depending on what that actually is, would dictate I think what I would recommend for you to track. But it's cool, but it does depend.
Ross Hudgens: Uh how about on the other side of the spectrum? It's like a classical e-commerce portfolio, probably maybe they have 200 SKUs. Yeah. That are major of of significance. What's their recommended frequency range?
Josh Blyskal: For e-commerce without being too generalistic, because there's just so many shapes in that space, it depends on how often you're showing up in those shopping results overall. Like it depends whether or not you're tracking ChatGPT shopping, it depends if you're working with Etsy, Shopify, uh any of the search providers there, or if you're doing it on the web yourself. As far as frequency goes, my perception is that the actual shopping results are more consistent than the citation results from ChatGPT's web browser or Cloud's web browser, for instance. That when it's pulling from that API, there's something deterministic going on about how the fan outs work and I think the APIs, the shopping APIs themselves are actually relatively limited. So the catalog of results is much smaller relative to the entire internet. Um, my fundamental belief is that even for someone who wants to be relatively repetitive, more than 5 to 10, you really start to see a decreased returns on investment. And I think for 90% of marketers in the world, you probably don't even need more than one or two. Two being like, a pretty nice extra bonus there. I mean, you're you're doubling the number of executions from 1 to 2. It's people, people think 1 to 2 is a very small step change, but it's actually a huge change to how much analytical weight um you know you're demanding out of your provider and how much you're going to have to pay.
Ross Hudgens: Well, it's interesting, and I I assume you guys can charge on a model of the more you run it, the more expensive it is for people, right?
Josh Blyskal: Happy to, but, we we try to give, we're trying to go high as signal with least number of executions, so that you can do more with those with those executions that you do have, that you can only say you can only run 100 executions a month, I'd much rather you you get a wider sample of your your market than having, you know, 10 little samples 10 times over. Right. It's important. I mean people, people will do different things though. Everyone has a right to their own opinion.
Ross Hudgens: Yeah, makes sense. I think of um in classical SEO there'd be people with maybe just one head term that's just like the bulk of their their world. Yeah. That would be for that person, it sounds like run that, or I guess, what is, what do you think the right frequency for that person with let's let's call it classical SEO had one page with 90% of the search volume? Wow. What would your get, I mean my my gut there is that you should run it between 5 to 10 times if you're literally tracking, I mean, daily? Yeah, a day, if you're tracking literally one head term, one prompt. Let's do it multiple times, if it truly is like that for your business. Okay. But what I would say is if you're Nike and you're doing shoes, maybe there's actually many query fan outs or many prompts that you want to track, and in that case, let's be intellectually honest about the different ways our audiences are actually searching, and if it's bigger than just what's the best shoe company, then you know, you're probably actually looking at more prompts, even though you might think in traditional SEO, hey, it's actually one head term. That world doesn't exist. That might actually be 200 prompts, depending on how granular you want to look at it. Like you should go with the prompt volume first. So, you see 90% of the search volume around this particular page or you're seeing 90% of your referral traffic from ChatGPT on this one page, it's not enough anymore to say, oh, what's the keyword for that page, and then that's our AI search keyword. Not to be, this is not a plug by the way, but like with our prompt volume tool, we can see in the wider world we have 1.5 billion prompt volumes, we have 1.5 billion conversations. We get the user's submission, but then we also get the model's response and with the response, we see global citations. So we can see, we can if you put a URL into ProFound, we could be like, your URL was cited, we can back into what prompts a URL is actually being cited for. So if you're like, hey, Josh, nike.com/shoes-men uh is our number one page by far, I could be like, listen, okay, there's actually like 20 main clusters that this is getting cited for. Here are the topical clusters, here's the distribution. If you say, hey, um it is actually uh adidas.com/shoes-women that's cited, I could be like, oh, you know, this is very different than anything I've ever seen. This is actually there's only two or three prompts that are really uniquely showing up, two or three fan outs. So depending on what that actually is, would dictate I think what I would recommend for you to track. But it's cool, but it does depend.
Ross Hudgens: Uh how about on the other side of the spectrum? It's like a classical e-commerce portfolio, probably maybe they have 200 SKUs. Yeah. That are major of of significance. What's their recommended frequency range?
Josh Blyskal: For e-commerce without being too generalistic, because there's just so many shapes in that space, it depends on how often you're showing up in those shopping results overall. Like it depends whether or not you're tracking ChatGPT shopping, it depends if you're working with Etsy, Shopify, uh any of the search providers there, or if you're doing it on the web yourself. As far as frequency goes, my perception is that the actual shopping results are more consistent than the citation results from ChatGPT's web browser or Cloud's web browser, for instance. That when it's pulling from that API, there's something deterministic going on about how the fan outs work and I think the APIs, the shopping APIs themselves are actually relatively limited. So the catalog of results is much smaller relative to the entire internet. Um, my fundamental belief is that even for someone who wants to be relatively repetitive, more than 5 to 10, you really start to see a decreased returns on investment. And I think for 90% of marketers in the world, you probably don't even need more than one or two. Two being like, a pretty nice extra bonus there. I mean, you're you're doubling the number of executions from 1 to 2. It's people, people think 1 to 2 is a very small step change, but it's actually a huge change to how much analytical weight um you know you're demanding out of your provider and how much you're going to have to pay.
Ross Hudgens: It's interesting and I I curious if we go the opposite direction and you have a very measured take on this. We have some of your competitors have a an option that's basically it's either daily or weekly. Yeah. And are you would your take be weekly is never significant enough?
Josh Blyskal: No, every, everything is significant enough with time. Yeah. But you have got to, I mean look at like SEMrush for instance. I mean, they don't update their keyword ranking stuff every single day for every single keyword. You'll see things that update, you know, oh, 8 days ago, 5 days ago, 14 days ago. You can refresh them by hand, of course, but paying, you know, by credit. But I don't know every, I don't think that every SEO does that. I think a lot of people, you know, go with the flow and you know, they're not sitting in SEMrush refreshing keywords. Um, hey, shout out if you are. Um, Semrush thanks you. Um, but the key is time. You'll get signal weekly. It'll just, you know, we found that to reach stat sig on daily, you've got to run your prompts for about 2 weeks. 99.9999% of marketers in AEO and GEO are doing it for more than 2 weeks at a time. They have, they have bought our software and they are tracking for more than 2 weeks and if you're not doing that, just wait 2 weeks. But for weekly, it's a little bit longer. I don't have the date, otherwise I would tell you the date, but I imagine it's on the scale like months, it gets, it gets more difficult. But it makes sense on the same same wavelength, maybe there's some, if you're looking at a portfolio view, you have 1,000 SKUs and e-commerce, maybe that would, that would, that would come down based on what you're saying. It's basically just a, a corpus wide analysis, really.
Josh Blyskal: Yeah, of course. I think the bigger problem in the space though is still just attribution. I think the frequency debate is, it's a cool debate to have, but it's hard to attribute. Even with frequency, even without, like, I almost think frequency right now is a bit of a distraction to the industry.
Ross Hudgens: On that, you shared something really interesting at breakfast today when I asked you about attribution. We've talked about it at lengths uh on this podcast. So uh we you don't need to tell us to do, um how did you hear about us as a drop down form, but you gave a new way that people are starting to attribute. What, what is that?
Josh Blyskal: Well, human and agent attention are actually analogs. Agent attention is an analog for human attention, and what we're seeing for some of the more um you know, radically adopted, people who have really jumped into the framework of AEO and GEO, um we are seeing them discount agent citations at a certain rate and kind of attribute that into clicks. So let's just say, you know, one in basically you take all the citations you've got from uh agentic user agents. So you monitor your network logs, you see ChatGPT's user agent, cite you, let's say 20 times. You apply, okay, we're going to discount it by 95%. We're only going to keep the 5% of citations, we're going to call that 1 click. 20 citations is a click or 100 citations is a click, because agentic attention itself is really going to um be downstream of It's going to be downstream of humans demanding content, right? Yeah, humans are going to say something, agents are going to be downstream from that. Agents are going to go downstream to the website further than that and then you're going to find out that an agent viewed your site on behalf of a human. At some level, there's human interest for you, but it's through an intermediary. And you can apply that kind of conservative reduction in the in the overall value of that, let's say, hey, I don't, I don't think a one citation is necessarily worth a click. Let's be conservative. All right, let's discount it by 95% and assume that, you know, 5% of people that see that citation are going to be influenced. They're going to have top of mind, because with the citation share, we're not talking about clicks anymore, right? We're talking about your appearance in the answer. We're talking about someone deciding to go to Google and, you know, search your name. We're just talking about mindshare at that point. You know, if you're cited 20 times, one of those people will see you in the response. One of those people is going to think about your brand, one of those people will eventually downstream to say, hm, okay, I am going to go check out Nike, okay, I am going to check out Skims.
Ross Hudgens: Yeah, uh so, just for clarity, I'm realizing I'm thinking through this myself, it's not that you're a source URL necessarily, but that your brand was explicitly cited. Exactly. Okay, makes sense.
Josh Blyskal: Makes sense. The brand was explicitly cited by the answer engine in that case.
Ross Hudgens: But you but it is the it is the case of being a source URL.
Josh Blyskal: Yeah, right. So, if the answer engine visits your page, you'll see it in the network logs, right. Yeah. You know that you are likely to be cited in that answer. It's visited the page, it's grabbed the content. You know, the chance that you're uh visible there is quite high in the answer and the text and that's why we apply the discounting as aggressively as we do, 95% discounted. Again, these are these are the ways that we can start to understand the value of an agent's time and attention on your brand. The value of an agent scraping your pages.
Ross Hudgens: So you're determining this via server logs, or what's the network logs. Network logs. Okay. And your visibility data, so understanding like for every time you're cited, how often do you appear in the answer. As well.
Josh Blyskal: Oh, okay. Okay, makes sense. That's like a secondary thing you can add on top. Like people are getting quite advanced with this, too.
Ross Hudgens: Right. Yeah, one of the things the logic that was going in my head was there's a there's a type of site that gets is a source URL, but is not a citation often like a Yelp, probably. Yeah. Or maybe I'm saying that backwards, but essentially an affiliate type site is influencing without being mentioned, without being mentioned. Yes. Yes. So that if that company did the click model, they're probably not getting clicked very often, either way. Either way. Yeah. So then you have to assume, basically, depending on what your discount and depending on what you believe about how often you're getting clicks, you've got to factor in your referral traffic. So, if your referral traffic is zero from ChatGPT, that changes how much influence you think you have for, you know, what that rate should be. Obviously, for affiliates, I think it should be lower than 5% for people who are on branded content or expect their brand to be in the answer, expect people to follow up with them as a result of the search, I would expect it to be higher, but we are seeing referral traffic from ChatGPT up 60% since mid-May. Yeah, that's great. Yeah, it's crazy. Um because ChatGPT wants to build a click model. They want to understand where people are clicking, so now they've made it easier to click and now people have started to click. If you're, I mean, everybody who owns a site right now should go check, I mean it's actually very interesting to see ChatGPT's referral traffic. It's almost universal of the companies that I've talked to this far, one of my favorite things to do in meetings right now is to say, have you since May seen an increase in your ChatGPT referral traffic? Without fail, I go to Fortune 500s, I was just at a Fortune 500 um yesterday before this. I, I mentioned it to them and they said, yeah, we actually saw more than 60%, we saw an 80% increase. I was like, wow. Why? Because they're doing CPG, um you know, customers want to immediately click, B2B, I think it's a little bit harder or if you're, you know, a distributor or if you're in industries where you're not a consumer packaged good or you're not a consumer facing brand, I think you'll see a little less of the tailwind, but for CPG it seems very strong.
Ross Hudgens: Yeah, well, we, we saw that as well, and saw that across our content, we were talking about the AEO sensor from HubSpot which is great, you could look at for kind of some industry trends. Yes. On, on that side. Uh yeah, it's interesting where I'm sure there's some agencies out there who would like look at that, be like, we're doing a great job, look, we jumped 60%. Of course you need industry relevancy on the side. Yes. Exactly.
Ross Hudgens: One thing we talked about today that hopefully by the time we publish this, will not be outdated. There's some contentiousness, potentially, or it's a negotiation stance by Reddit. Yes. With, with with Google in particular. That obviously has big implications on AI search, potentially.
Josh Blyskal: It does.
Ross Hudgens: What is your thought process on what's going on there, where it may lead?
Josh Blyskal: Well, I think at any given time, Reddit is trying to ascertain the value of what they're providing to these answer engines. I think there is this kind of larger valuation problem that um Reddit has to approach. Everyone knows that Reddit is a cornerstone to ChatGPT, a cornerstone to Perplexity, incredibly important across many of these different search services. What really isn't talked about is, you know, what Reddit's end in this is going to be and how Reddit's going to commercially benefit in a way that benefits their advertisers, uh their own model, beyond just getting clicks or beyond just influencing the answers out of the goodness of their hearts. Obviously, last year we saw ChatGPT and Reddit, um, had some huge fluctuations which people attributed to negotiation between the two parties, uh where, you know, Reddit. Oh really? I didn't, yeah, makes sense, so. Reddit went from like 9% of, the negotiation piece was the, the light bulb for me, yeah, yeah, yeah. Where the, there's a lot of, because you saw Reddit, ChatGPT, the traffic source, Reddit has this massive, massive, massive, massive, massive, massive amount of data, right? And so people are like, oh, does ChatGPT dislike Reddit? I'm like, no, I actually like, it's hard for me to explain or it's a hard to understate how, and I love Reddit, but unimportant it is that, you know, Reddit particularly has to own the answer. What ChatGPT is really indexing on here is UGC. Yeah. Right? When, when Reddit goes up in ChatGPT, it's not ChatGPT saying, we love Reddit, we just want to, you know, crank Reddit and make Reddit more important in the answers. It's saying, this information class, the portfolio of the answer, might be better if more of that pie chart is dictated by UGC and what consumers are saying, so let's see what happens if what consumers are saying, what conversations online are saying are larger in the answer, and let's dial it back, let's test it. And so when Reddit goes away, I don't think the need for UGC necessarily declines. Mhm. But the sources definitely have to change their landscape. You'll see Quora coming in, you'll even see folks like, I mean, Wikipedia is not a good analog for UGC, but you do see like industry forums, Wikipedia, they try to find it from wherever they can. Um, but all that to back track and say, a year ago, we saw this with um with Reddit and OpenAI. And now we're thinking about seeing it with Reddit and Google. But, um it remains to be seen, you know, how that relationship's going to go. I think for marketers the lesson here is that AEO, GEO is still a relationship business at the channel level. Yeah. These channel strategies are not you know, they're not a result of nature, these are results of partnership meetings and negotiations and discussions and these things could change rapidly. For instance, Claude cites Reddit almost zero. Claude does not touch Reddit. That's not because Claude is incapable, it's just because the agreement isn't there. So as as professionals in this space, we've got to watch these like a hawk because this dictates how we're going to start moving for customers and clients. We've got to know that, okay, if you're trying to win in Claude, if you're, you know, trying to market to you know, white-collar software developers, and you're doing your DevOps product, and you want to do a Reddit campaign, that's never going to show up in Claude. And if you think your audience is using Claude, maybe you have to rethink your Reddit strategy, not because of, like I said, not because of what's going on, you know, fundamentally with the answer engine, just because of what is accessible. Yeah. It's a big walled garden problem.
Ross Hudgens: Yeah, and that whole content partnership thing is is subject to change rapidly.
Josh Blyskal: Every day. For sure, I mean, yeah, every every everyday publishers are max. Uh, I've heard it increase in tension of as the, the news publishers are getting hit, they are effective, some of them are the, the lack of track traffic is making them want to more be like, why don't we just completely pull the plug on this thing? What do we have to lose? Maybe it'll help our negotiation, right. It's a walled garden, right? Yeah, yeah.
Ross Hudgens: Yeah, so it'll be interesting to track. I guess one of my my where my brain goes on UGC, one observation I had recently is that I I had this thought process that LinkedIn would become a next variant of Reddit.
Josh Blyskal: It is.
Ross Hudgens: What I, all right, I mean I'm curious what you're seeing recently, but what I just ran in the last couple weeks is it was skyrocketing in Q1. It is kind of come back down to earth, as at least a Google traffic source, maybe I haven't run the prompt data um for it, but obviously them like many people are seeing a glut of AI slop. Yeah. And I think that's part of the reasons corrected in Google, but uh yeah, what's your current view on LinkedIn's it depends on the answer engine. It's like you said, it is it is hard to say because it's so volatile right now. You'll, what you'll see is you'll typically see, it is it follows this kind of quite bouncy pattern where an answer engine will quietly test a channel like Reddit or like LinkedIn, it'll spike in citations. Everyone's like, oh my goodness, here's the new channel for AI search. Um, and what's going to happen is naturally the test experiment's going to end or the partnership trial period's going to end, or whatever's going to happen and you return to like kind of a normal state. But with LinkedIn particularly, we're starting to see it pick up steam. I mean they are the number I think it's number one source uh for B2B UGC, at least across. Oh okay. even above Reddit. uh for B2B UGC. I think right now it's like Reddit, G2 and LinkedIn are all like fighting for that crown pretty tightly. But LinkedIn was peaking up there for a little bit. It was quite impressive.
Ross Hudgens: Yeah, I mean it's interesting to think about a world a scenario a world where Reddit is less influential for B2B, especially, like what are we're now back to BoFu first party roundups, probably. So I don't know, Reddit I don't know if Google is going to like that one, because there's not a lot of like great sources after that besides LinkedIn I guess. They basically and then we're going to be littered with AI slop content in the search results, so there's not a there's a lot of ways this is going to go.
Josh Blyskal: I mean it still has to follow, you know, I still have faith that answer first high utility content is the content that wins in AI search. Content that actually adds to an understanding of an industry. My, my perspective is always that the opinions, biases, takes, thoughts of a piece of content are the thing that differentiates slop from not slop. I don't care whether or not an LLM created the final piece of content. Yeah. What I do care is that it has a reason or a purpose for existing. I can't tell you how many pieces of content I have read, that don't rank in AI search because it's just like looking at a checkerboard and just moving pieces around aimlessly. We're getting nowhere, we're doing nothing because it's just a vague post about men's running shoes that doesn't talk about anything and it has no perspective. Okay. But when you bring a perspective to the table, you have all of these little assets, you have all these little kind of beads, you know, if you're thinking about it like a necklace, you've got all these beads on the necklace and the only job the LLM has is just to run the string through it and say, the ams, the ors, the the, the buts, run the string through the beads, tie it together, boom boom done. You have a good piece of content. I don't care about all those filler words. As long as I have a strong outline and I believe something, uh and it forwards the understanding of an industry, you're doing work for an LLM. Like, if you think about the LLM's work as like a pie chart, 100, it's like, you write you press enter in the keyboard. The LLM understands 0% of the industry that you're talking, what's the best CRM? The minute you press the keyboard Enter, the LLM understands 0%. The pie chart's totally red. As the LLM starts to search, you try to fill it up with your sources, moving as much of that pie chart to like a green pie chart as possible. I want my source to basically teach the LLM as much as possible, to do as much of this work of teaching the LLM what the CRM industry for instance is. Um, in the fewest searches possible, to give it the most value. That's why like listicles and stuff still work so prominently, because they're doing the caloric work of contextualizing an industry, offering a framework for it, even if there's this like counter punishment right now with Google, the answer engines are still demanding it which is probably a one of the most fraught conversations in the space right now. Yeah, yeah. So painful, because I want nothing more than to have these things in more alignment. I dislike the LLMs and Google kind of asking for different things out of both sides of their mouth.
Ross Hudgens: You, you kind of hinted at it, yeah, but what yeah, what is your take on those articles? Uh, I know tools the tools want them, it will recommend them, I forget if you guys do that or not, but some people have had contention with, maybe it's not you, I don't want to speak for you, but yeah, what's your opinion on those?
Josh Blyskal: We still, we still have, I mean, we still recommend them because in answer engine search, if you look at the citations, they are the sources that I would be dishonest for me to say ChatGPT wants anything else at the moment. It would be me trying to build in other and like, if you purely want to win in ChatGPT, it is still the most viable strategy despite the pains associated, and so we need to think about how to do this tastefully. Like the industry as a whole has to mature beyond this like self-promotional nonsense angle. There is a point where I think we're close to reaching criticality there and we're about to crest the hill. So for everyone listening, like you know, write it down, July 2026. This is the time where, you know, we need to be thinking about how we do the work of answering these questions and contextualizing these industries in a way that's copasetic both to humans and agents, and in a way that's tasteful about the future of our content. At the same time, winning today is a totally different mandate, and I hate when those things are in tension. Right, you don't quite know where it's going to move. But you know what the data is pointing to today and you know what the data is probably going to point to. If you've been in the SEO industry for any measure of time, you know what's going to happen, which is that with any strategy that people adopt, the strategy loses value and it gets deprecated and we move on. Yeah. But we're in this in-between phase of what is the next strategy, what does work, what does keep our content as cohesive as possible? No one wants to hemorrhage tread search content and no one wants to hemorrhage those clicks. At the same time, the mandate from every board and CMO in the world is to win AI. Yeah. Um, earned media is a great, a great play there. Um, earned media is doing a lot of heavy lifting in the space right now in a place where a lot of people are very um you know, weary of the owned site plays. Social is also helping tremendously. Like Reddit, LinkedIn couldn't be any more relevant in the time. When you say earned media, sorry to interrupt you, but, I think that's that's a Please. what I've seen a lot in citations sets is X company was named the best blank, and they'll do like a press press release as a as a component. I don't even, I don't even know if that'd be counted as earned media versus just a press release done by the PR team, but going in more detail, what kind of earned media you're you're talking about?
Josh Blyskal: I mean anything where, I mean it's a wide spectrum, right? So anything where you're getting placed in press articles where you're generally actually going to the market with a story, with an insight, with research, with data, with a market perspective, all the way to even just sponsoring a piece of content. I've seen them work in both contexts. So it can be you know, relatively low lift you know, in terms of contacting a journalist, working with them, getting included in other lists or roundups. It does tie back to E-E-A-T. Look, we're all the way back to trustworthiness. Who has a right to be ranking men's running shoes? Who, who has the trustworthiness? Who has the authority there, right? Who does? You've covered running shoes a lot. Yes. Yeah, do you have repeat.com? No. They do. I mean they have a lot of great citations. Shoutout to Runner's World um by the way. You don't you don't sponsor me, but maybe you should. Yeah. Maybe we'll give you links, we'll delete your topical authority by LinkedIn to you, I'm sorry.
Josh Blyskal: Um, no, I think um you know, we're lucky enough to live in a time when YouTube is also tremendously important. Like for people that are like, what do we do besides listicles, it is time to get creative about placing on our own pages uh things that are slightly different than just like a roundup, but also at the same time thinking about different channels, right?
Ross Hudgens: What I've seen is it's mostly third-party affiliate-esque roundups on YouTube. Is that is that, I mean I've seen exceptions where people have like created their own video where they round up other people. Seemingly in less competitive markets. Yeah.
Josh Blyskal: You know, I see a lot of product tours and a lot, like if you're in I'm talking now about like B2B and like software, but like it's a lot of like contextualization. Another good job to be done within AI search is just simply ensuring that the information the model needs to know about your product exists in a way that it can access regularly. I see a lot of people jump right to listicles, just I'm like, no, you don't need to do that. What you really need to do is make sure your product pages, make sure your about pages, make sure like the fundamental facts about your business are accurately represented. And so even before you're like, oh, do we I hate being put in this position where we have to make these, you know, comparative pieces. Maybe what we do is we actually just start working on, you know, YouTube videos about what our product does, bringing in customer testimonials, doing you know, just the 80/20 work of ensuring that when you ask when ProFound was founded, the right information shows up, what ProFound's products are, their right and updated information shows up. A lot of what brands are doing today revolves around accuracy. Even. It's quite shocking. Last year, very much visibility focused. Everyone wanted the Viz. Now people want accuracy, sentiment, comparisons. It's quite interesting. Yeah. You shared a data point of like beyond best X articles, what is the second most search for commercial query?
Josh Blyskal: So G2, um put out some insane research on this. 33% of B2B commercial queries and this is B2B, were best in category questions. What's the best CRM? 31% of B2B commercial queries were X versus Y.
Ross Hudgens: Right and right now there is no, you know, strong product take yet showing how you directly compare in a competitive framing with your consumers. ProFound is obviously working on this. Uh, I'm not going to go into too much detail, it's a product we're um you know, beta testing at the moment with a few customers. But the cool thing there is that your answers are going to change completely depending on the, the competitor that you're framed against. It's not as easy as just saying, tell me about ProFound. If you say, tell me about ProFound, you get one answer that kind of politely summarizes our products, but you say ProFound versus competitor, you know, who's the best for um, a gentic marketing, you're going to get a framing that says, well, for this, ProFound is great, but for this, competitor is great, and if you, you know, wanted an ultimate choice, go with ProFound, they're going to do this and this and this, right? But knowing those battle cards is so important. I mean think about B2B procurement. Once you're at the evaluation stage, you're going to be like, all right, I have to buy a CRM, I'm down to my two choices, Hubspot or Zoho. Hubspot versus Zoho. Pop into ChatGPT, see the results. Every person, every CMO in a B2B business, every CMO in software, knows that is what keeps them up at night. Those are queries that are maybe the most commercial, the most important to win, especially in AI search. You are mere millimeters away from capturing that conversion, as long as you can show up properly in those comparisons. And that's what we're going to do.
Ross Hudgens: Yeah, I love that. I we ran a similar study where we looked at all of our clients' commercial content and then linked it against who is correlates most with traffic. Not which content type drove the most traffic. Yeah. But did the client have the content, and then how much AI traffic did they have? And the content type of the traditional alternatives versus and best, surprisingly, versus had the highest spearman correlation of actual traffic. Yeah. So it was really interesting when you brought that up because it was like I feel like that, you're completely right and I'm with you on that, and I'm glad ProFound's building that product that this the next layer is evaluating those things and influencing them to a massive degree where you can. Yes. Exactly. It's exciting. Yeah. So you touched on, if you want to uh optimize for ChatGPT, or you better have best listicles. I'd curious if I, if I framed it as a divergent generalization. You're optimizing for ChatGPT, or you're optimizing for Cloud. July 2026. What are the differences in a general fashion between those two right now?
Josh Blyskal: Cloud dislikes UGC. Cloud never cites Reddit and Cloud likes listicles more than ChatGPT.
Ross Hudgens: Okay, so it likes it even more?
Josh Blyskal: Yes. It likes it even more, Cloud is like ChatGPT on steroids. Okay. So it likes it even more. Cloud is perhaps the easiest model, currently, to market to and influence, depending on how you would quantify those things.
Ross Hudgens: Surprising, right?
Josh Blyskal: Because it is, how what I hear from how you just waterfall that is there maybe a tighter corpus it pulls from, so therefore, if you influence one of those corpus areas, you swing it. Is that fair to fair to say?
Josh Blyskal: It's 70, right around 70% of all Cloud responses and citations are basically, are sorry, 70% of all Cloud citations are are directly from Brave's top 10 search results. So, Brave Browser, if you've never heard of it, neither have I, uh basically I love them but there, it's a great browser product by the way but they're not well known. It's an incredible, it's an incredible. Is it getting more adopted or are you just hearing it? Is it it's privacy focused, right? Yes. Yeah. And it's just it's excellent, it's very performant. It's a strong browser, but it's like, it's like Opera, you know. It's one of those like, how's it different from DuckDuckGo which sort of has a similar feel? It's just like a very, um, you know, if you're in the know you know, it's a very high quality browser, anyway, whatever. It's it's really quite cool. Um, but not well marketed, at least in my opinion. They could be so much larger than they are right now. Um, but Cloud's doing a really good job at bringing them in anyway, into fruition. But 76, 72 sorry, 72.6% of all Cloud citations are directly from Brave's index, and as a result, you can track where Cloud is pulling those results if you track your Brave rank. Your Brave rank really, like, you don't even necessarily have to track Cloud in any of these analytics tools yet if you want to do it scrappily, you can use the Brave search engine index tool, there's like an API, five bucks per 1,000 requests. If you really want to understand what's going on, you can gather the fan outs that are important for your prompts in Cloud. By the way, the fan outs in Cloud, they're also extremely consistent. ChatGBT's are almost entirely inconsistent, but Cloud's fan outs are roughly the same across like 65% of fan outs remain the same day over day. They're tighter out of curiosity? They're very tight. Yes. So you can build strategies on top of those fan outs. And as a result, you take those fan outs, you can plot them into Brave, and you basically apply the same kind of click map. Well if I'm rank 1 on Brave, I'm this percent to be cited, if I rank 2, I'm this percent, I'm this percent, I'm this percent. Okay, let's grab those 1,000 API requests, let's be honest, ask Cursor or Cloud code to write me a program that grabs all the API requests and cleans it and gives me my overall percent citation share. And you can reverse engineer it. I mean like, the the key here is that with these answer engines, the strategies that apply are more often dictated by the indices, the channel agreements and the the prompted kind of biases put into the models, than necessarily by the content that you might be creating. We have to approach the models with our content in the right way, rather than creating general content and hoping that the models pick it up. Claude's entire content retrieval system is different than ChatGPT's is different than Google AI mode. And respecting that as marketers is a really important step in really running a critically uh acclaimed GEO-AEO program.
Ross Hudgens: Yeah, you your mind makes me go to a conversation we had at at breakfast again, uh like how you future align this thing. How you just described it, how I would think about future aligning that, is naturally how you would have a marketing stack and you'd almost think, I mean you would have a lean towards the things you know influence AI search but ideally you would operate agnostically to AI search in a way, uh such that like if your audience is on Reddit, you have a Reddit presence and as a byproduct, when it becomes an influence up and down, you're there. But if you're doing it purely for AI search influence, you may help the executive kind of conversation short-term but you're probably more ripe for volatility and like you're not building audience off those top layers because you're distracted by the bottom layers. Does that make sense when you push back on that?
Josh Blyskal: The only place I'd push back on that is the arbitrage opportunities where people find things that they wouldn't know are actually critically important for showing up in some of these search services, that that don't exist in the trad media landscape, don't exist in trad SEO. I can't tell you how many times I'm working with like a Fortune 500 and I'm like, these are your top 10, you know, earned media sources that are dictating your narrative and they're like, we have never, and my job is media planning for this company, we have never heard of three of these ten sources. The other seven find cool, whatever, but like seven, six and three are weird individual blogs that we should work with. Or these subreddits I would never have known that these subreddits were actually the ones dictating my conversation. We saw, um we were working with an investment company, I cannot say who because it's a little bit funny, our /povertyfinance was one of the most like, the critical and they were in like R /stocks, R /investing, you know, like how you'll see in accounts, places where you think you would go. It was actually R /povertyfinance that was dictating, right around like 70 to 90%, depending on the type of conversations that this brand really wanted to target in Q1, by the way. Just Q1, they were really pushing on a new product. It's a credit product, um and they were like, oh my goodness, okay, okay, this change is a little bit of how we're going to go to market with this narrative if this is what's actually being discussed and these conversations are happening. You don't often know that until you can see the zeitgeist that like a million miles out. Once you see it though, I think the disciplined thing to do in execute, like execution and execute, like executing on this, is to approach it from first principles. I think if you start going to Reddit and saying, I'm here for AI search, it's a beautiful way to get turned around and sent right out of Reddit. You're going to get ejected. Um, you have to be like, all right, so we know it's important, but almost let me cleave my brain in half and now let me just approach Reddit like Reddit. Let me just approach media like media. Let me go make good content. Let's not try to, um you know, you know, hack the system so to speak, but you have to know where you want to show up. I mean, the data here is, it's rapidly still changing.
Ross Hudgens: Uh, on that kind of like arbitrage thing, a trend I've seen, I'm curious if you have any comment on it, I I saw for a few clients for the first time, there's like these arbitrage like best X software, it was kind of a legacy SEO play, it very clearly looked like they bought a site with some domain rating, this topical authority for maybe research or something, Yeah. And propped it up as a citation source and it would rank probably on Brave, and maybe it ranks on ChatGPT, but not on Google, just because they figured out that kind of manipulation. But for a moment, we're in a period where that works for some AI search engines. Yeah. Have you seen any of that at all? I mean I don't know the I wouldn't recommend it, unless you like, unless you like I think that's one of those arbitrage things, it's probably right now, yeah.
Josh Blyskal: It's like going to the casino, right? Like, I mean you are living on the next roll of the dice with something like that. You can win, sure, I think there are ways to win there, but the more you invest there, the more I think deep down in your gut, you know you're investing in quicksand with something like that. It will be taken away from you, it really is just a question of if, rather than when. I mean, everyone in the world has an incentive to strip that influence away from you if you put yourself in that position. The way to think about this is really coming at it quite self- it, selflessly, rather. The question the thing I like to post to marketers is, think about what ChatGPT, what the engineers at ChatGPT, what the engineers at Anthropic would look at and say, wow, I wish our answer engines had more sources like this because this makes our answer, answers so good. This makes our answers so good, this makes our product look genius. This is so smart, this is so great. The customer, the user is getting such a well-reasoned answer about running shoes or about CRMs. Dang, if only people could understand how our answer engines retrieve content, and how they want a balanced well-reasoned answer about this piece. Um, man, I really wish you did this. At the end of the day, if you put yourself in, just like with Google, this is no I mean, everyone in SEO is listening to me speak and they're like, duh, Josh, of course, align with the interest of the search engine or answer engine, but there's a lot of people coming in this industry who think it's a game of hacking, it's a game of manipulation. It's not. It is a game of aligning with the providers and making the providers look like geniuses. The sooner you realize that, the sooner you start getting clicks delivered back to you. Yeah.
Ross Hudgens: So they're not saying expired domains in those meetings?
Josh Blyskal: No, they don't. I think we're okay.
Ross Hudgens: Maybe, maybe that's where I go, maybe maybe. No, I shouldn't go there. But uh, one of the overall things about all this and I mean I think we hinted at it along the way but to put a bow on this thing, there's a three, there's short term and long term plays here. Like what is, what would your summation of like, we're tracking, what I want to build a strategy that yes, drives incremental, now, yes. but is around three, three to five years? What what would that be?
Josh Blyskal: One, I think the most important answer engine to look at right now is uh Google AI Mode and Gemini and AI overviews. Um I think these are I think people will be surprised in 2027 to see how, um, important those models are going to be. They're quietly, quietly accumulating billions of users. I think that they are a very strong search service. I think ChatGPT is pretty much your very close second for a pure play answer engine. As far as content goes, I think you have to think about that ideal mix versus your content mix. So you have to think about where the arbitrage opportunities are, both in earned, both in affiliates, um, but also with owned, figure out what the demand is around your owned, figure out alternative ways to play the listicles. So, I think there's still this need calorically to create understanding around your industry. Maybe that's in product roundups, maybe that's in, you know, product release, you know, notes and and diaries, maybe that's PR releases, but finding ways to give the same caloric understanding in different formats. Thinking about first principles, what works, um, natively for answer engines, super important. Markdown is going to be a key play for content creation, creating content that's kind of native or at least kind of rhymes with Markdown formatting. You've got your headers in those places, semantic chunks are still here, you know, summarizing with data first into extractable chunks of text that answer engines can pull and cite, very important. Um, ads are going to be important as well. I don't think anyone has to worry, you know, today about their business going under because they didn't place enough ChatGPT ads, but I do think that ChatGPT ads, however nascent, are going to prove to be one of the most, uh, important innovations in advertising in the last decade. Um, I think it is going to be the most targeted ad platform we may have ever seen in history. Nice, tell Google as a version of it.
Ross Hudgens: Yes, until Google has a version of it, and then we're all, then we're all going to change our approach. That's a great mic drop, Josh. I should have led with this, but you're uh writing a chapter in our book on GEO coming up. So I really appreciate you participating in that. As great chapter on the technical side of of this whole thing and you did a great job on it and that that's publishing in middle of October. so excited about that and Uh, yeah, you are a great follow on LinkedIn. Yes, sir. Mostly because it influences AI search. You said it, not me, really appreciate. Thanks for having me on man. Yeah. Uh, X, anything else you should, people should check out for from you guys?
Josh Blyskal: Follow me on LinkedIn. Follow me on LinkedIn, follow me on LinkedIn. It's great. All my research is on LinkedIn, all the best stuff. Um, really excited. Thanks for having me on.
Ross Hudgens: Yeah. Thank you, Josh. All the best. Cheers. Yeah.
