← Content and Conversation: Organic Growth Insights from Siege Media
Content and Conversation: Organic Growth Insights from Siege Media
AEO Playbook: How to Optimize for AI w/ Profound’s Josh Blyskal
00:50:15
Josh Blyskal, AEO Engineer at Profound, joins Ross Hudgens to dive deep into the emerging world of AI-native search optimization—and how brands can win visibility across platforms like ChatGPT, Perplexity, and Google’s AI Overviews. They unpack Profound’s Conversation Explorer, fresh data on generative search intent, and the real-world tactics that drive clicks in the new AI-dominated SERP. From schema tips to 2025 URL hacks and LMS.txt files, this is the tactical roadmap for modern marketers navigating the future of search. Show Notes 0:08 – Why AI-native visibility is heating up 1:00 – Profound's Series A and platform momentum 2:15 – The moment for AI search is here 4:40 – How Profound built the largest dataset of answer engine queries 7:12 – New ad tech categories marketers aren’t tracking 9:40 – What “generative” search intent actually looks like 11:25 – Tools vs. blogs: what answer engines really cite 13:00 – Why traffic quality from AI sources is insanely high 15:05 – The r...
Loading summary
Transcript69 lines
- [00:00]
Josh
Foreign.
- [00:10]
Ross
Josh, I'm excited to have you here. Your company is taking the world by storm. This, this whole EO thing of course is as well, or however you want to describe it, which you may have some opinions on as well. But clearly I'm getting asked by our clients, like daily. It seems like a new client is asking, what LLM tracking platform should I recommend? You're not for everybody. You're an enterprise upmarket offering. But I've been recommending you a lot and seems like demand is through the roof. Things are wild over there and people are talking about profound like, yeah, how are things currently?
- [00:48]
Josh
It's been incredible right now. So I think just the last two days, we've just announced our Series A, so led by Kleiner Perkins, Nvidia Ventures, we've got Coastal Ventures jumping in there. It's been an incredible run right now. I think, you know, it's actually interesting. We've got clients across from, you know, small businesses all the way to enterprises. And you know, I think the moment for AI search in marketing is really starting to reach its momentum and kind of really starting to find its rhythm. You know, marketers are really trying to do something. We're moving from that point where people are thinking, okay, this is a space to stay aware of all the way to the point where it's like, okay, as a business, we're seeing that marketers need to take action. Marketers want to know what's next, what are we going to do in search? How do we get our traffic back? How do we think about these new systems and how they're going to influence things? I mean, for context, profound. We came out of stealth in August 2024, so fairly new in the world of software, I guess. Relatively mature in the world of startups, one could say. But it's been an interesting ride, right, with AI mode and the evolution of all these technologies as they to kind of evolve and we continue to track them. You know, this space has really started to find its, its place and it's kind of its home within this wider search world. So it's been really interesting. It's been beautiful, really.
- [02:13]
Ross
Yeah, A lot is happening all, all at the same time. We'll probably do another podcast on this, but it almost seems it does tails into the demand you're seeing is we're, I think our whole industry is seeing this reporting need shift where we're thinking of like, traffic's not the thing anymore. We need to be visually other elements. And of course a lot of what you all do is that but just as of this last week, we were like making a more profound, no, no pun intended, shift to highlighting chat GBT and our reporting and thinking about. We got to like, put traffic a little further down in. In the order of operations. But clearly exciting times a lot can be definitely exciting for you. Maybe not a little nervous for a lot of SEOs, but still, I think a positive change keeping us on our toes anyways.
- [03:06]
Josh
Yeah, I mean, I think the search world always evolves. Like, SEO eats everything. There's like that, that, that little, like, you know, there's like a. There's like 10% of my. Of my heart that's like, yeah, I really want AEO to be really different than SEO. But I know, like, even if AEO is different, it's gonna. It's like as long as there's an engine that people are searching on that we need to optimize for, it's gonna get eaten. It's gonna be within that larger sphere. It just depends on what that. Like, I always try to model it with like, two different curves. Curve of like, adoption of like, all right, how many people are on these AI chat flows? And then there's a curve of like, difference, right? How different are the chat flows? And so like right now, the way I think, that's like how I think about the adoption of this whole market. Right? Like, so let's just say adoption. Let's just pretend adoption's really low and it's not really interesting, but the difference is huge. That validates that the space is still worth kind of tracking and being in. Right? Like, at least at a minimal level. Like, okay, if the difference is massive and maybe 5% of all searches are done through AI, then, you know, at least those 5% of searches need to be understood because maybe they're super high intent or whatever. If you're in a situation where adoption's huge, but the difference is still low, obviously that's super valid as well. Like, so if it's very close to Google search. But also 95% of searches happen on AI search, but it's not like, exactly aligned. It's like we're in a world where we still need to track it. Basically, I think that the takeaway is that we're reaching a world where the adoption is getting higher with Google AI mode, all these different tools, and it's different. And that's really the hair on fire problem that profound is built to solve. And that's what I do all day. I'm an AI strategist. I guide our customers through basically all the different strategies, all the things that they need to do to actually tangibly improve and start to work on their AI visibility. And that's what I'm excited to share today. I mean, it's going to be great. We're going to kind of get deep in there. So. I love talking shop. Nice. Yeah.
- [04:58]
Ross
I mean, you made me think of your. One of the immediate questions was your new Conversation Explorer tool, and there's those estimated search volumes in there. Can you share what, where first, where those come from and to what you just said. Are there clear. I'm guessing you're also hinting at both how we optimize, but maybe the type of searches are different. But are there differences? Is, I mean, there are, but yeah, maybe just talk through some of that variation and also where you're getting that initial search volume data as well.
- [05:32]
Josh
Yeah, what I'll say is that the search volume data is really like, we're really proud of that. We have a number, we have some of the best data providers, some of the best data agreements in the industry. We have the biggest single database of queries across these answer engines that exists outside of these model providers. And so we have this pipeline. It's just this beautiful pipeline that pipes in tens of millions of queries every single month across, you know, you know, your ChatGPTs, your perplexities, eventually Google AI overview, Microsoft Copilot, as we start to really think about how to scale this out. But it's, it's legitimate, you know, answer engine queries. It's the first in the industry to have it. So we, you know, no one's really built this before. If, you know, if we were thinking about doing it again, we'd probably have brought data scientists on, you know, right at the founding. That way we could have been even earlier on this because it's been, it's been a huge value add. Like you have, you have. There's so many different platforms out there that do kind of what Profound does. But I think the advantage to what we're doing here is that we're, we're, we're going headfirst into the, like, into the mess. Like, it's all the mud. Like, it's like, all right, how do we get the query data? How do we get the legitimate, like actual information that people are searching for now that we're able to model that, we're starting at this point where you can go in as a marketer and start to understand the key topics that people are actually asking about within your industry. So you could ask like, you know, how many people are asking about budget laptops versus, you know, gaming laptops or how many people are asking about, you know, corporate credit cards with lounge access versus or not. Yeah, I guess. Yeah, sure. Corporate credit cards with lounge access versus corporate credit cards with employee spend management. You could see the differences there in real time. Like it's a week by week view, which is incredible. And so, you know, the way we got it, we've been working on it since November. It's incredibly intense. You know, we love our data partners. Here's what I can tell you. What I can tell you is that there are basically two whole new intent groups that marketers are not watching out for. Like, this is definitely a breaking story. We're working with our insights team right now to categorize and figure out how we talk about this insight. But, but something like, depending on the way you slice it, something like 50% of the intent in the. I mean, and it sounds kind of like, duh, obvious when you hear it, but it's 50% of the intent in these answer engines is just like, it's generative queries or it's just people like conversating and interacting with these models, which means that your informational type queries, your navigational, your transactional, the commercial stuff, all of that existing stuff gets actually squeezed down into a much smaller proportion, a much smaller share when people are actually asking to do things. I mean, and it makes makes sense why, like answer engines, like, what I can say is actually I can even give you the. Which is really interesting. I have this up right now. This is our like visualization right here. Something like 40% of queries are just purely generative. There's this new kind of area called no intent, which is just people saying like, yes, please do that, go for it, try again. So it's this, it's this interesting world, but like, if you see the, you know, the SEO world's kind of in green over there in the bottom right hand corner and you've got the AI search world kind of coming out pretty new. This is something that, you know, we think is absolutely worth tracking because brands are going to be mentioned. Even if you're starting from a generative position, you know, how do you position your brand to inherit a new type of query? These new kind of, these generative searches. What does ChatGPT actually use in its rag model when it's going to go find a generative piece of content? Like something like, you know, let me edit a PDF. All right, so what Pages are well positioned to win that particular piece. If someone wants something that's, you know, obviously not ChatGPT, so long as ChatGPT does, you know, ChatGPT image, you know, change the game. Like, you know, Canva's got a great like, image creation plugin right now in ChatGPT. Obviously, you know, mid April, when ChatGPT images came out, maybe that entire world shifted. Maybe other tools are going to be inside ChatGPT in the same way. Or how do we make sure that our tools, you know, exist and continue like microapps, things like that. Even in a world where ChatGPT is handling most of these generative queries, how do we position ourselves as like the next experience? So there's lots of interesting stuff. Yeah, go ahead.
- [09:48]
Ross
Yeah, super interesting. So when you just restate it back to you, otherwise could be said, go do something. For me, action, like take an action is what generative means, essentially.
- [10:00]
Josh
Yeah, it's like create, you know, create me a spreadsheet of my employees hours, you know, create the schedule, you know, stuff like that. But, you know, if someone wants to do something one step deeper, who's going to win that? Click. Who's going to win that search? Right.
- [10:17]
Ross
It reflects what I think a lot of people are seeing. And I know Ahrefs did a study on this as well, like their, their traffic. And I know it's true for us. We have like a contrast ratio tool. Tools are getting a huge amount of that volume and I'm guessing that's where that's represented a lot of the time. Is that fair to say?
- [10:36]
Josh
I think that's like a really good moat right now in AI search is thinking about how do we create experiences that are so specific and so high value that they overwhelm whatever basic level experience that ChatGPT can create. Like thinking about microapps, thinking about micro experiences. That's a really good way, even in the traditional world. I was starting to hedge your bets on the traffic side of things. Even if you don't want to even start thinking about aeo.
- [11:01]
Ross
Yeah, it seem. I, I know it's. I, I was talking to Kevin Indig about this as well. It's like in traditional search, you're not really getting AI overviews on those kind of searches too. So it's like in some ways the most defensible. And yeah, maybe it's actually much the search volume is growing significantly because people are unlocking like, oh, I can do things and they don't.
- [11:22]
Josh
I don't.
- [11:22]
Ross
This is a new mode of Searching to your point that we haven't been thoughtful about previously.
- [11:27]
Josh
I mean like the whole, the whole search world is transitioning from like a means to an end to like the actual end itself. It's like, you know, when you were searching, you know, pre LLM era, you know, the reason you would search through all the blue links and you would look through all the information is to find the thing that you wanted to understand so you could go do a thing. Search is now just becoming all right, I want to do the thing, let's just do the thing, do it, create it, do it, let's go. And so it begs an important question about the shape of websites to come. Like what's the incentive for top of funnel? How do we think about top of funnel? How do we make that whole world work? At the same time, I think it's still in all the AI model makers interest to try to keep those incentives alive as long as possible. I think they all know that training off of all this content that's been kind of generously provided to them is the core to why they're able to disrupt traffic as much as they are. So it's a really interesting world where, you know, human written high value content I think is only going to get higher value and more scarce. It's going to be kind of like finding, you know, a diamond in the rough eventually if incentives don't change.
- [12:33]
Ross
Yeah, the business model, I mean I don't have access to ChatGPT's economics or any of the other LLMs but it just doesn't seem sustainable unless they eventually go to a world of an advertising model of some sort. Which you'd think would they would want to create some incentives to click out to have that be an effective model. Same with Google. Like they don't want to lose that model obviously they don't want to get disrupted at the same time but hopefully there's like some crossing happens where ChatGPT introduces that Google.
- [13:03]
Josh
Yeah.
- [13:04]
Ross
Still can retain it and where you still have an open web that's fruitful.
- [13:09]
Josh
Yeah. With the ad model I think, I think that's the moment where the like ever so clearly once ads are supported within all these AI search responses magically I think we'll start to see click through rates start to improve and conversions start to improve as well. I just think the incentive is so, it's so it's nasty but it's aligned against you know, the people who have the website right now. Like the, the trustworthiness of a ChatGPT answer is so High people trust them like they're, you know, they're bible truths. Same thing with, to a lesser extent, AI, AI mode, AI overviews. Perplexity, though. Perplexity is doing a great job, I think. Like, if you're going to look at any model and look at maybe what's to come in the future, Perplexity is kind of the canary in the coal mine for this stuff. Because they were, you know, first to shopability. They're doing sponsored queries now. So you can go in as a brand and you can sponsor those. You can sponsor follow up queries. So you write something like, I don't know, you know, what's the best non permanent. What's the best non permanent marker for my seven year old? Great. The next one's gonna say maybe the follow up query is how to wash permanent marker off of walls, right? And then it's sponsored by like Clorox or something, right? Like that's an interesting situation where, you know, as a brand you kind of want to beg the question, like, how do I know? How do I track that? Like, where's the search console for that? Where's the ad platform for that? What's that gonna look like? Like what, how do I put money onto a keyword in AI search? Right? Because if someone puts in a seven paragraph query like, oh, I have a, you know, there's, I want to buy a car, but I've never bought a car before and I really want to learn how to negotiate to buy a used car. And I like Honda, but the new Hyundais look really cool to me. Also, I live in New Jersey. What car should I buy? What's the keyword? Right? Like, so I think there's also this world where at the same time as we're trying to kind of understand what people are searching in AI models and what the searches look like and how to kind of group them up. We're hoping that the AI models actually meet us kind of halfway and start to try to build that understanding themselves and start to kind of build their own platform. It's a question I get all the time is like, do you think profound is like cooked if, if, you know, ChatGPT releases a search console, like actually to the contrary, I think then we're in a much better position because then we start to actually have units of measure and, and things that we can actually kind of latch onto to start saying, all right, that's, you know, that's a brick, that's one unit. All right, cool, we can plug this in and we can start to optimize for it. It's the same way with like search console.
- [15:38]
Ross
Yeah, I perplexity in that four links at the top model. I don't know if it, it doesn't have nearly the demand of Google for any of us and there's no search console to my knowledge of like click through rate. I'm curious what the click through rate is on that. Do you have any sense of what that is for perplexity is or quite.
- [15:57]
Josh
Bad from the four links to the top?
- [15:59]
Ross
Yeah, yeah.
- [16:01]
Josh
The interesting thing about Perplexity that we see from our own data is that perplexity has like between like a 6 and 10x higher click through rate than ChatGPT does. For what it's worth, even though perplexity has like a minuscule perplexity is the Bing of AI search. If ChatGPT is the Google of AI search, perplexity is the thing. But the conversions from perplexity are excellent. Like we'll see, you know, let's just say 10,000 views, 10,000, you know, visits from Perplexity out of you know, a hundred thousand scrapes and we'll see you know, a million scrapes from ChatGPT. We can see this within answer engine. I'm just making up fake numbers. We'll see a million scripts from ChatGPT and we'll see you know, 150,000 conversions from you know, ChatGPT. So it's in a, it's, it's, it's a really big difference across all these different models. I think the quality of traffic is going to be a really important factor though. Like I think the one bone that has kind of been thrown maybe unintentionally, maybe. I mean obviously kind of as a consequence of the space is the fact that like the traffic coming from these answer engines is extremely high quality. Like they are ready to buy, they are qualified. So when you get someone who clicks, a click is a conversion for the most. I mean I've heard of traffic CBR as high as 20 to 30% from ChatGPT. That's a high end. Like I'm not going to say that's normal but for sites where you can convert right on the site, you can sign up for a trial, you can demo 20 to 30% isn't strange and I've heard that more than five times, less than 10 times, but it's a pretty good sample size across maybe a few hundred customers who I've asked the same question to.
- [17:35]
Ross
I mean our demand, we hear it mentioned all the time. That we. We were found on recommended on ChatGPT and perplexity. I I think it's kind of interesting that some historical pessimism about Google results and Bing results and like what a lot of like these best roundups would have. Like they're all paid and all those things and a lot of these models to my knowledge are using those exact same things. But they're taking it as science when it comes through those. I wonder if that will persist. I'm not sure. Maybe it'll just get better.
- [18:07]
Josh
I don't think so. I don't think we're in a world where anything here is solid ground. I think everything. This is like quicksand Bill. Basically. Yeah. Listicles comparative content we have. I have a data set of like 177 million citations that I use here internally at Profound. It's like the. It's like the king data set of everything. Listicles comparative content are 32.9% of all citations. That's the number one most cited single type of content. The number. The number two is all blogs and opinion at 9.9%. So that gap is humongous right now within AI search. Yeah.
- [18:45]
Ross
I remember seeing. I think it was a man. I should just. We just wing this off top of my head. But it was like research affiliate products are getting sourced a lot but blog content was like actually very low. Like 1%. I guess I is that it sounds like you still have a little higher than that in your numbers.
- [19:05]
Josh
Yeah.
- [19:06]
Ross
But curious if you've seen that study X maybe X funnel.
- [19:10]
Josh
I think it's accurate. I think like directionally they're accurate. I think you know, it depends how you slice everything up. I mean you know they could get 1% but if they're. If the basic finding is that like listicle affiliate content's winning and blogs are kind of a sad number too. Yeah, it's true. Yeah. It's this great X funnel.
- [19:26]
Ross
Oh yeah. I guess I do see blog here at 9%. So that's basically what you're saying. Maybe I was misrepresenting that.
- [19:31]
Josh
That means my data pool is pretty good. Yeah, I brought that up. That was. That's my old like Brighton stat. But yeah, I mean the situation here with that is that like right now the answer engines are predominantly still using. We're still kind of. Most searchers are still using those 4.0 models. They're not doing the deep research. As long as we're on that like one search and done kind of methodology. And AI mode's challenging that like I'll, I want to like put that off to the side. We'll talk about AI mode, but like, as long as we're on those forum type models, when they have to do one or two searches, they're always going to bias towards the listicle to try to get the widest breadth. Like, kind of like the marketplace of answers. So if I ask an answer, best men's running shoes, right, it gets to throw one, you know, ChatGPT gets one query over to the Bing API, maybe two queries over to the Bing API. It's, it's going to be limited in what it's going to be able to pull and it's going to want to try to find someone's article that lists everything about the industry out all the top players. It's going to grab those top results and then it's going to throw those back into the user and be like, all right, yeah, the top list of the top men's running shoes are 1, 2, 3. You'll go to the listicles that it sources and it's like, yeah, no surprise, it's number 1, 2 and 3 in the listicle as well. Like, it, it's that like marketplace that everything's being forced kind of like into. Basically. I'd say like right now, answer engines want to make a marketplace with deep reach, research, O3 and the query fanning and AI mode. I think those things are going to change. I think we're in for a big change there because those answer engines actually have the time to go visit the product pages and go visit the review sites and go see every individual page. Fingers crossed, hopefully. Because I want them to be able to come up with their own opinions. Like if I, I basically personally stopped using like 04 with search on. Cause I was like, I don't want to read, you know, the summary of a listicle. I want to just like, you know, go through a deep research if I'm buying anything, like serious, serious. Yeah, it's at least what I do. Because I mean it's, it's. Right now we're in a very transitory stage. Like we've got this rag model with OpenAI, it's heavily reliant on Bing. Like for anyone who's listening, who doesn't know already, Bing is the foundation of ChatGPT's retrievals, right? So if your site's going to appear in ChatGPT, it has to get indexed in Bing and then ChatGPT will index it in its own index. But if you're not in Bing, you're not gonna make it through that first filter. But this rag methodology that's using Bing is, it's, it's really kind of made by academics. Like you've got a bunch of PhDs in the room. They wanna make a performance system. They're not thinking about marketers or SEOs or AEOs or whatever to call this space. They're thinking about how to like how to scale something that gets web pages that are reasonably accurate in a reasonable amount of time that provides value to the user. I think that's going to shift that whole thinking over at OpenAI. As long as they want to continue being a pretty serious search engine right now it's going to have to shift pretty seriously in time to start thinking about people who are going to be optimizing for these systems. A great case study right now is like looking at those listicles. I don't think that can't be permanent because then every brand is just going to Release a top 10 comparative listicle. As with any SEO strategy, it's in vogue for a year and then it's done. Fun.
- [22:49]
Ross
Yeah. I mean they, they work unfortunately so we'll keep doing them while they do. But I mean do you think the pricing. I mean I'm no tech expert. I know generally things tend to exponential or get very, a lot cheaper over time and I know they already have already will that. I know the fan out mode of AI mode for Google does some of that. Do you think this will be something that'll be accessible? Just my mom.
- [23:23]
Josh
Oh. Like deep research and stuff like that.
- [23:25]
Ross
Yeah. You just on a like consistent basis. Yeah.
- [23:27]
Josh
I mean it kind of has to be. I mean here's my, here's my real thought. I think, I think in the future I think LLMs are going to get dumber. I think that's my like take that I always tell people is that we've got this, we like we're in this awkward space with LLMs where they're doing this job that they're not supposed to be doing, which is like if I ask an LLM who the Count of Dusseldorf was in 1443, it's gonna like reasonably try to tell me that information. That's not the LLM's job. The LLM's job is to model the language. It's just its job is to be the interface that I can understand. And in an optimal world I would have a very small, very optimized LLM that's like super good at modeling language. And then it would be connected to A really smart brain that lives somewhere else that does all the knowledge stuff for it. So like I like, if you're trying to use like ChatGPT on your iPhone, if anyone like downloaded that like ChatGPT extension, when Apple Intelligence first came out, people were like, oh, you can use ChatGPT on the iPhone. You can ask, hey, ChatGPT, search this. It's just doing an API call. My future vision of this is that you'll be able to run hopefully an LLM on your phone locally. And that LLM is going to be relatively robust. It'll connect like a Google search or a Siri. Like, I can't believe Siri is not incredible yet because they've got all the different technology, all the pieces. They just, Apple has to build it in house. But they'll run a query. That query will do deep research. That deep research piece will be handled by some sort of intelligent engine, some sort of search engine piece. Then the LLM, the answer engine, whatever call it what you will, is going to actually go in and write that answer to the user. And that's the piece. It's like that, that user relationship piece is really where the LLM starts to really own and the knowledge kind of gets offshored. As far as I feel with like pricing and things like that, Google is just like. I mean, Google is just driving towards turning queries and compute into a commodity in this space. Like Google's like, yeah, you want to charge for tokens? OpenAI? How about free? How about just free search, 50 searches, 80 websites? How about you pay nothing? That's hard to beat. That's just a hard economic problem for anybody else, which is, I mean, it's a great business move. Right? It's exactly what I would do if I was Google.
- [25:35]
Ross
Yeah, it reminds me of the Amazon model that made them so efficient that they just crushed everybody because they could lose money on certain categories until they won the category because they had Google. Google has the search ecosystem to do so much of that.
- [25:51]
Josh
I forget who it is. There's like, there's even like this research group right now hiring with. There's like a research group in the AI space. It might be meta. I think it's meta. The signing bonus for the top research talent in the world right now is a hundred million dollars per person with it's 75% stock but 25% liquid, which is crazy.
- [26:07]
Ross
Yeah, yeah, I saw they were trying to recruit people from OpenAI. That's pretty wild story. We'll add that to the show notes. But you're sort of maybe you're touching on. I don't even know if you're recommending those listicles to clients but like thinking about this dichotomy and of AEO to SEO or SEO for AI or however you want to call it, but like doing optimization for these tools compared to search, what do you think a strategy right now is if there are any that you would do for specifically LLM benefit that you wouldn't do or maybe you wouldn't prioritize the same way to make it helpful for people.
- [26:46]
Josh
I think there's, there's some really interesting stuff right now with like stuff I wouldn't do with SEO that, that, that makes it really difficult because the way I like to contextualize the relationship between AEO and SEO is like, like here in AEO land we are still working out the same muscles. Like we are using meta descriptions, title tags, schema, you know, FAQs we're using, we're still writing for featured snippets, but maybe albeit with a little more data based insight from what LLMs are doing. We're still thinking about our URL slugs, we're still thinking about eat in content, but we're doing all of those things. And the way I say it is like we're doing slightly different exercises but for the same muscle groups. And so we're basically thinking about how to use those same units in a different way. So there's a few things that I would, I'm gonna lead with like kind of this is the ultimate optimizer mode thing. Like if you're a crazy optimizer and you don't even care about SEO, you want to like let it all burn. I want to just be visible in answer engines. This is one thing you can do immediately. But I think it's, it's, it's not best practice in SEO because of redirects. I think you might know what I'm, I might, some of you might know what I'm gonna say, which is that you can put 2025 in your URL slugs, in your title tags and in your meta descriptions and see an upwards of 20% improvement in ChatGPT citations just by the inclusion of 2025. And for everyone asking why, the reason is because ChatGPT appends 2025 to the end of search queries it sends off to the Bing API. So you say something like I want to buy that new car, blah blah blah, all that other stuff you said. And then ChatGPT reads that seven paragraphs. It does something it basically calls regularizing the query. So it basically turns it into the query that it decides it's going to send to Bing. So it asks, you know, best used cars, best used car dealerships, N.J. 2025. A pending 2025 instills a bias in both the Bing API, obviously and in ChatGPT's model to basically pick and drive towards resources that just so happen to include that string 2025 in the end results like, that's why I say this is the ultimate optimizers play, because you're going to be dealing with redirects in 2026. What are you going to do? What's the plan for any of those things? Like do you create the 2026 page? It's kind of awkward, but we've seen customers who are saying, you know what, I'm going all in on AI search. I'm just going to do this, I'm going to try this. The pickup is insane. There's some great research right now as well on LinkedIn Pulse, like building consensus right now in models. So people are saying if we're going to create a blog post, it really does help to cross post that into LinkedIn Pulse so that when ChatGPT does search, or, you know, your perplexity, your AI mode does search, it sees a few different sources saying the same thing. You know, maybe your blog, maybe over at LinkedIn saying the same thing. And then we can go as the model and feel pretty confident that, okay, multiple sources are saying this, this is something that's known in the world. Got it. I'm going to show this to the user. There's a few more I'm going to hit on, maybe two more. I think semantic chunking is a really big win. Semantic chunking with data, to be very specific. So what does that mean? What that basically means is that when we're writing our blog posts, when we're writing for answer engine pickup, what we don't want to do is we don't want to create paragraphs or areas of our content where the exact meaning or maybe the exact insight from that piece of writing that paragraph is kind of muddied. We want to keep insights really snappy, really data based, really data Dr. And we want to almost think like thinking back to how ChatGPT answers our questions, can I format an answer as kind of a self contained paragraph within my. Within my content? Exactly. Because when Perplexity wants to go and look through its vector database of all the searches, when ChatGPT wants to understand in its Index what source to cite. It's looking for hyper specific answers to long tail queries. And so building those hyper specific answers is really, really, really important. It's really powerful being the person who takes data from, you know, all the best listicles in one space and creating like the master like crazy listicle with like, you know, your competitor has seven different dimensions. You have you know, three and then your other competitor has seven more. You grab seven, the seven dimensions. They're comparing the seven dimensions from the other competitor and then add two more on top just for good measure and then create like a killer structured table in HTML for answer engine pickup. You're going to see citations just as a natural result. Like the answer engine, you want to create the best marketplace, the best venue for a lazy, tired answer engine to come and be like, all right, great. Finally the place to learn everything about men's running shoes. Cool, I'll just. Thank you so much, Ross. I'll just take your answer and I'm just going to go show this to the user. Cool. Takes it, steals it, throws it to the user. That's the, I mean that's the name of the game for.
- [31:44]
Ross
Those are great insights on One of the things I've been personally curious about is like do you think there's actually a very clear uplift to implement schema.org on? Like I think if you describe the semantic chunking like it maybe it's like an FAQ list at the bottom of a blog post or something. Is that notable? I. I was just chatgpting and googling if I. Is there any actual data that supports this rather than just like structuring the article? Well, curious your thoughts on that.
- [32:15]
Josh
I have experiential like I have experiential data with it. We've refactored a few websites with customers. The thing with doing that kind of stuff, if like, if I'm working with someone who doesn't have schema and schema markup, I'm also probably changing a bunch of other stuff on the site. Like I'm going to be very honest. Like there's also a bunch of other things I'm probably going to be doing on the site. At the same time I don't have a rigorous study into schema. I think the best uplift I've seen so far, surprisingly is author schema, which is really weird. Like really building out like a nice author schema has been really nice for answer engine pickup. That one I'm still. That's like, that's a two or three week Old insight. I'm still trying to dig into it in my spare time, why that is. But we were working with a cybersecurity company. They had a few cybersec experts commenting, you know, writing some guest blogs. We just built out a little more author schema. Just really kind of started to standardize the way that that was going to be placed on the site and pickup increased drastically from existing pieces of content. So that was interesting. I think the biggest thing right now is there's two kind of areas that I would really start to think about. I would think about your URL slugs. Semantic long descriptive URL slugs. When an AI search that is clear to us at least like the data is overwhelming in that regard. So if you have a page that's like, you know, Ross.com corporate credit cards and it's a listicle that's comparing the best corporate credit cards that slash corporate credit cards URL alone. Just assuming the same content across both pages is going to lose every time to some like total nerd who's looked at the data and can see that you should be writing for answer engine pickup. Best dash corporate dash credit cards dash compared to 2025 in the URL slug. Even without the 2025, the performance uplift is a pretty significant magnitude. I would say it's probably between like 3 to 7% citations just from a really nice URL slug. Because when these answer engines, especially ChatGPT, looks at those Bing API results, it is trained specifically to look at URLs and look at results that it's highly confident in. It is supposed to only click on things that it believes contains after the click the exact answer it needs. So if it's looking at 10 pages that say corporate cards, corporate cards, corporate cards, best corporate cards compared. Don't mind if I do click. Same thing with meta descriptions. It can see the meta description. So in a lot of ways when you're thinking about being cited for answer engines, it's kind of a two step game. So step one is like kind of marketing the COVID of the book. It's like, all right, we want our URL slug to look really good. We want our meta description to be really sexy so that the answer engine thinks that there's a lot of value behind the page. And then when we click on that page, we want that page to be presented in a structured, detailed way so that the answer engines click is not wasted. It gets the information it needs. We actually do influence the search and then that information is brought forward to the user.
- [35:11]
Ross
Makes sense some like follow up questions there. One thing I've been it's been top of mind for me in this direction is when you do a lot of these articles on site you end up with 2025 and like all of your your titles and it looks spammy because you have like nine related links. They all say 20 guidance to the team has been just kind of indicate this in the title tag not have it in the post level the post title and generally because we are going more evergreen don't want to redirect URLs. We we wouldn't have a year in the URL. But you're specifically calling out the URL a lot there do you think? I I'm guessing this is just like classic Bing optimization is just like not as good as Google. This is good but any thought on the title tag versus URL?
- [35:55]
Josh
I think the title tag is like the title tag is the normal person path to do this. If you're crazy the crazies are doing the URL just because if you're going for broke it is the area where very few people are still touching. It's the blue ocean for this kind of thing. I think everyone basically the moat to putting 2025 in your title tag is just so much less significant than putting it in your URL. You're full sending it if you're going to put it in the URL. You're really buying in that this is going to be this one thing. I would say the title tag for 90% of people is going to be just fine and an excellent choice. I think there's like this subset of like ultra optimizers who are just like. Like we work with customers who don't care about SEO anymore frankly. Like they're just like okay or even like they haven't started. They've never done SEO before. Like new companies fresh. Like I don't even really worry about SEO. We're in an industry with massive incumbents. I don't care. I don't really want to fight against, you know, building this content out. The interesting thing is like they say they want to only do aeo. In reality they're doing both at the same time with a little AEO kind of sprinkle on top. Obviously. But the name of the game is just like just thinking about how to make things super obvious for these answer engines. I mean internally, profoundly call that agent experience. It's like it's kind of everything, right? It's the operator being able to go on your site and click around you know, seeing all the different menus, making your menus understandable. It's your LLMs txt, LMS txt, whatever people call it these days, that uplift people are so skeptical of that. I think to set. I always want to set the record for that because there was this moment, I think in March, April where everyone was like bullish, we're all going to try it. We tried. It didn't change that much about our referral traffic. It didn't change that much about our crawl rates. If we looked at our network logs. I think that's an accuracy play. Like I think LLMs Txt is really a play where we're trying to make sure if you know, if a developer is using cursor or if someone's asking about the relationship between our products that that information is portrayed appropriately. I also think like if we're talking about llmstext, if I'm going to be really quick here, explicitly allowing LLMs txt in your robots txt, even if it's totally redundant. It's like absolutely redundant to do that. But we've seen it drastically increase the amount of actual pickup from answer engines like rl. Oh really? Yeah, quite significantly because of that. Also there's this thing called LLMs full. Txt. If you're a software that has a lot of documentation, the play with LLMs full. Txt and we work with a few really, really, you know, smart teams who have done some good research into this even beyond what we've done. They found that you go to your docs page. So like imagine try profound.com docs. Then you would do LLMs full. Txt. It would be a markdown file of all of your documentation hosted there. We see that getting picked up something like 5-10x more often than your standard LLMs txt because of those, the specifically Cursor and Windsurf.
- [38:57]
Ross
Okay, interesting. Yeah, I know there's been some, some people just saying there's like it almost was a comparable example I think of amp. I don't know if it's the best example of like no one's actually adopting it. Like do you think it's going to matter long term? This is going to continue to be something.
- [39:15]
Josh
Yeah, I think it has to matter long term. That's the interesting thing about this. It's like for going to this world where the answer engines are arbitrating true versus false or what products matter, we've got to create. I think there's going to be a basically a two tier highway. We've got to create a way for these answer engines to access information consistently in a standard way across our sites that doesn't rely on the old paradigm because you know, I think temporarily at least JavaScript isn't rendered temporarily at least there's some blockers with crawlability. People are still restricting answer engines from being able to see certain pieces of content. Rightly so for some brands, some companies, no shade there. I think it's a great play, it's important for some folks. But I think there's also a world where we want to make sure that that maybe second layer of the freeway is still open and clear so the answer engines can go get that information. Like in my perspective on the future of search I have a pretty optimistic one is that answer engines allow us to really free the website up to be a really exciting, maybe unoptimized place. Like maybe the user facing.com website is actually like radically designed and is not super intuitive for answer engines. And it's almost like kind of like a piece of art rather than an actual. Because websites right now to my opinion look very samey right now Websites are very optimized. There's a very similar structure in kind of a basic sense across websites allowing answer engines to confidently go in and navigate some sort of backend architecture be it LLMs, Txt, be it whatever it is allows us to do crazy things again with our websites. I want to see weird websites again. Like I want to go back to the, you know, early days of the Internet and see some of those like archive sites. Yeah, right. So I think you know there's, there's a, there's plenty of different areas where things can go but I think that that avenue more than anything else, I mean like weird websites aside, that avenue still has to be clear. We have to decide, you know, how to answer engines access pages and I think just kind of like shrugging and saying well they'll just do it the same normal way is kind of a lazy answer. I think that you know, creating a markdown file is not the craziest thing for brands to want to do.
- [41:22]
Ross
Yeah, makes sense. One, one follow up question on the time freshness. Have you seen just from a lift on just timestamps. So it's like the post date on the article that's that's pretty massive as well.
- [41:37]
Josh
Huge. Huge. Yeah, humongous. There's a decay in almost all content in answer engines. Unless it's like it's your first week on, you know, on publishing is great. You'll we see content get picked up in the scale of like 2, 1, 2, 3 days, you'll release a piece of content, it'll get cited, it'll be 2% of the truth of like a multi billion dollar industry. Like it'll be like, you'll like, I don't know, Chase bank will release a piece of content two days later, it'll account for 2% of all citations in the industry and then it'll start kind of dwindling and dwindling for the next month until it reaches like 0.5% which is really where it's at for the next like three years.
- [42:13]
Ross
Do you think that's the decay cycle is one month?
- [42:15]
Josh
I think the decay cycle is anywhere from like a month to two months in my experience. Yeah, it really does take like a month or two. Just really find your like your actual area. The interesting thing is that those timestamps, they're dependent on the size of the industry, right? Like if you're writing about feather caps in Texas, you're going to be fine for the most part because you're going to be the only person doing that. But if you're talking about something like I don't know what's a great one? CRMs. I came from HubSpot's SEO team. Shout out to HubSpot. But if you're talking about CRMs, you're going to be covered up by some rando writing another article, you know, the next week. Especially if everyone knows that listicles are the way to do it. Right. Sorry. Content quality is just gonna have to get really important soon.
- [43:01]
Ross
I we have a general stance like it's about relative freshness on that search result. When we're thinking Google of like you if it is a query like best CRM software, often that can be a quarter. Like everything's a quarter or less old for something like best credit cards is actually gonna be under a month. Best headphones could be like one week I think. I would guess it would be very similar. Like it's probably using those similar results to inform. But do you think there's any variation? I mean clearly it's going more exacting and. But is it even more profound in that way?
- [43:36]
Josh
Yeah, I mean everything is kind of accelerated right now in AI search is what I'd say basically on the unlike the. On the data side, like things go up, things go down much quicker. If you're seeing things on a week with headphones in traditional search, I'd say that that sounds perfectly accurate to what I'm seeing within AI search as well. Like things like, you know, industries where things are just always moving the way to stay defensive. There is commercial landing pages just because, I mean there's always going to be. And weirdly enough, Wikipedia, that's like a side thing. Wikipedia. Let me, let me just append that Wikipedia, commercial landing pages and Reddit are like really some of the strongest evergreens right now in AI search, just to be very clear. Because those specific resources are kind of like totem poles for these answer engines. They just, they ping them out of necessity. They feel that that is an important place to check the box at. They will hit a certain Reddit thread, they will hit the Wikipedia page. Wikipedia page is super underrated. The problem with Wikipedia is just like you can't really. It is what it is. I really, I am not, I don't get into the Wikipedia world. I just, it's, it's like if you want to. The only as far I go as far as saying you should create a Wikipedia page if you don't have one. But I don't think that there's any real structural way to think about optimizing. I mean, I'm sure someone's out there. Someone out there is like listening. That's like tearing their hair out. They're like, yes, there's a structural way, I'm sure. But I can't even get into it with Wikipedia. It's just so difficult to really break through and do that consistently and create a structure by doing that.
- [45:11]
Ross
I mean, moderators may flag.
- [45:12]
Josh
They'll kill you. They'll kill you. Yeah. And that's their job, like, rightly so. I don't want, I don't want Wikipedia to be a marketing channel. It just so happens the answer engines pull that way. So I think that's also just an interesting piece.
- [45:24]
Ross
If I'm reading between the lines of some of the things you said. Like, it sounds like you from a future proofing standpoint, you're recommending developing real long tail commercial landing pages.
- [45:38]
Josh
Yeah.
- [45:38]
Ross
On app, like eventually the deep research models may go that direction. Is that sort of what I'm hearing? That feels like a core. Maybe that's newer where previously you may have just done a long tail blog post, rounding up the best options and putting yourself first or something like in.
- [45:55]
Josh
The, in the ideal world right now, if you're thinking like how do we win for the next year or two. Right now it's about creating a landing page for any kind of like, it's almost like problem oriented. It's like a solution oriented landing page. Like if I Map every use case to a landing page for the same product almost. And that sounds insane, but it's exactly right. Like long tail clps are a great evergreen way to stay ahead because there's always going to be that moment where the answer engine just needs to at the baseline know which solutions exist in the industry. Like if you're thinking about, let me just think of it like I don't know. No Zero trust proofs. I'm just making this up. Let's just say some company is out here creating zero trust proofs, which is like, it's a crypto thing. You know, you would just create like Josh crypto.com zero trust proofs. When people ask about zero trust proofs you would, your page would show up because it's one of maybe five. There's Ross Crypto, Josh Crypto and Bob crypto dot com. They all have their zero trust proofs pages. Right. And the enter engine is going to hit each one of them. It's kind of like that. And you can go longer tail. I mean you can go crazy. Right? I mean it's, it's as deep as your products go.
- [47:02]
Ross
Yeah, makes sense. And there's ways to have that be user friendly where you have like a all use cases site type thing that goes off the core. I've seen that be effective. Well, I feel like I could go like an hour and a half with you, Josh, but I want to make sure, be respectful of your time. Any other like research or product launches or anything else that's upcoming that you're excited? We already talked about some of it. But anything else that we should chat about?
- [47:28]
Josh
I mean we are, I mean we're excited to tease basically what will be coming in the next few months, which is our, our actual actions piece of the platform here at Profound. So right now some of the biggest feedback we get from marketers and users alike is that this data is incredible. The data is, you know, world class leading. You know, no one has this data. But how do we actually understand, how do we understand for our site, for our brand how to actually take action? One of the things that we're building right now, maybe I can even search it up, maybe not. But I'm not even, I'm not going to do it right now. But one of the things we're building right now is this ability to actually create content briefs and create content within the Profound platform itself. So no longer are you going to have to go out to an external source or use a ChatGPT deep research. We're going to be able to use this massive, hundreds of millions of rows of data, huge data sets to create real, actionable, AEO optimized content direct for your brand, which is going to be insane. I think, you know, driving towards the future, the future is basically, you know, autonomously starting to think about what the solution is for your brand, for your, for your industry, for your company and then bringing that directly to you. Like maybe in the future there's not going to be a need hopefully for that much AI strategy. Maybe that's all going to be visible in the data. Maybe I can go do something like maybe I'll go be a fisherman in Bosnia or something, I don't know. But it'll, it'll be an interesting evolution in the space I think and I think it's going to really, I think it's going to redefine like what it means to start kind of optimizing because it's just, it's pure data based optimization. Learning from the, the best patterns in the data across this massive data set. So everyone's going to have the ability to really go in there and start thinking like a top tier researcher who's got all this data like just thinking about all these different factors. So it's going to be very cool. I'm excited about it.
- [49:18]
Ross
Nice. I'll look forward to checking that out. I know you, yeah. Have historically been a, I mean you're up market but you have a self serve option now that's 500amonth at the time of this.
- [49:27]
Josh
Yes.
- [49:27]
Ross
Recording.
- [49:28]
Josh
Yeah, yeah, yeah.
- [49:29]
Ross
It's more, more accessible. So that's great. Yeah. You're putting out some great research. Yeah. You put in some good stuff out on LinkedIn.
- [49:36]
Josh
Any.
- [49:37]
Ross
Anything else? How should people find you?
- [49:39]
Josh
Oh man. Yeah, yeah. Find me, find me on. I mean, yeah. Josh on LinkedIn. Find us on Tribe. Profound. Find us on Chat. GPT perplexity. Yeah. I mean please. Yes. Sign up. Check it out. Email us. We are really interested to show you what's going on with an ao, what's going on with your brand. Yeah, it's really interesting space. Appreciate it.
- [50:00]
Ross
Awesome.
- [50:01]
Josh
Thanks for coming on Ross. Absolutely.