Top SEO myths debunked with real data | Ethan Smith (Graphite)

0:00 Ethan, welcome to HR's podcast. >> Thank you for having me. >> I think you're now the expert when it comes to answer engine optimization. You've been everywhere like all the Laney podcast, tons of webinars with a lot of companies on LinkedIn, you're wildly regarded as the guy and yeah, I I support that claim because yeah, we've met back in San Diego when we were doing HFS Evolve. we get to talk and I learned how much effort you're putting into this and how how much data you research. So you're not just sharing opinions, you're sharing data.

0:35 And I believe today with this episode, we're in for a session of SEO necromancy as I would call it because we'll be resurrecting SEO from the dead. you have some interesting data research to to refute to debunk some of the SEO myth myths and I'm very excited to dig into this. Thank you for the kind words. Excited to be with you. >> so why don't you lead the way? I believe you want to present a bunch of SEO myths that you want to debunk.

1:07 I've seen your notes. I have some of my own comments there, but yeah, what do you want to start with? >> Yeah. So I I think the last 20 webinars I've been on have only talked about answer engine optimization. I haven't talked about SEO in I don't even know a year maybe >> because everyone's interested in answer optimization >> but SEO SEO is huge and I would say that there's this increasingly consistent narrative that SEO is going down. So you basically have these two curves and they're you know they're they're surpassing each other and the key assumption is that if something goes up something else must go down. It's zero it's the zero sum bias. and we saw this in, you probably remember this in 2010. In 2010, the app store launches and people said web is going away.

1:58 Mobile apps are going to take over. Teenagers have stopped using the web. only old people use the web. So, you better, you know, you better get on board. Don't worry about web anymore. websites don't matter. Just use mobile apps. And that was a zero- sum mindset. Now, it is true that mobile apps blew up. Like, the promise was real. eventually happened, but what did not happen is that web did not go down. And that's exactly what we're seeing now where AI is actually blowing up and the promise is I don't think it's overhyped.

2:29 But the premise is that the web must go down if AI is going up. And so that's the main thing that I want to debunk. And I always knew from looking at our clients websites like they keep going up and people are telling me it's going down, but I I keep, you know, or not like every single project that we work on, but like consistently I'm seeing charts go up. So clearly this this can't be true. And so I've been digging into it and there's a couple I like where does this narrative come from? And it comes from two different things. The first thing is that media it is more interesting to say things are being disrupted than to say things are not being disrupted.

3:04 >> It always works. >> Fear mongering works. Yep. If you have a new product say that the old product doesn't work anymore so that you have to buy the new product. The more disruptive the better. And then the second thing is that as I've been digging into it, I've seen several pitch decks from new software companies that you know say, "Well, you don't need HFS anymore. You need you need this new thing." and a key thing are these converging charts and studies that say that >> SEO traffic is going down. And interestingly, I I looked into it.

3:35 There's multiple market research firms and all of them, this is premised on self-reported surveys. So, it's me sending you a survey and saying, "Hey, Tim, are you using AI?" And you say, "Yeah." Or like, "How often are you using AI? How many times did you search last week?" And you're like, "Oh, about seven times." And and then, "Okay, well, SEO traffic is going down by 25%." Like, that's the the actual premise of of of of these studies is you quantifying your previous behavior. And anyone in psychology research, my background is in human computer interaction. Anyone who has a background in that knows that asking you to quantify your prior behavior is a terrible research method.

4:12 You know, it's kind of like if I said, "What's the search volume for chicken recipes?" Oh, well, I'll ask Tim what what he did last week. Like, of course, you wouldn't do that. You would just use >> panel data or like search volume from from Google. >> So, if you have actual quantitative data, like clearly that's better than a self-reported survey. So I started looking into this and there's the key myths are number one that SEO traffic is down 25%. Number two is that AI going up causes SEO to go down.

4:40 >> The third is that AI overviews is reducing clickthrough rate which is true. And then the fourth is that Google stealing more and more clicks. And all of those are wrong. >> Let's start with the first one. SEO traffic is going down. So it is down for hrefs.com I'm afraid. So it is down for href's blog I'm afraid. So we we suffered from it and we're not alone. I I like earlier this year we we caused a little bit of drama with HubSpot famously when Ryan Law, our head of content, posted that oh look at HubSpot blog which has always been regarded as kind of the kind of northstar of all content marketing and it was losing traffic dramatically.

5:27 We're not seeing that dramatic traffic losses for HF's blog and for HF's website as a whole, but it is going down and they see a lot of other people reporting this, but at the same time, I do also see people who are bragging about their SEO traffic going up. So, it's not like doom and gloom across the board. So, I wonder what data have you studied in this regard? >> Yeah. And so the the biggest stories are about Monday.com and HubSpot and they going down. And that's true. They did go down. And I saw you I saw a a comment of yours in another post where >> one of our friends said, "Well, I looked at like 15 sites." And like guys, 15 sites. Like that's >> you can't look at 15 sites and generalize based on 15 sites. I mean, it's it's not it's not wrong, but you need a large sample. And so all of the other studies that I looked at are like my one site went down. True.

6:21 >> Yeah. >> So what I did was I looked I I I worked with I looked at your data, number one. And then number two, I worked with Similar Web. And so Similar Web >> has a bunch of panel data of total traffic. And I we looked at the top 40,000 sites. So not 10 sites, but 40,000 sites. It's the 40,000 largest sites in the United States. I looked at it over 2024 and 2025. So, it's basically Similar Web's organic traffic to all of those and I also independently validated that it was directionally correct because I correlated it with some of our internal data and then I also correlated it with some of Google's data. So, I evaluated the similar web data was was directionally accurate. So, I looked at 40,000 sites and of course, some sites went up, some sites went down. And so, what I saw was that total traffic is down, but it's down 2.5% overall >> and not 25%. And I could just show you a quick example here where I have that.

7:22 So, this is this is summed across all all 40,000 sites. Mhm. So again to be clear because there's a lot of ways we can look in the data. What we can do is that we can measure the growth or drop of traffic per website and then we can aggregate it to an average or we can sum up the traffic of all websites and also track like the total traffic to all websites like which one did you do? Did you did you do an average per website and then aggregate or or did you do the sum of all traffic?

8:03 >> I looked at a few slices. I did not look at the median like the the percent the average percent change within site is what I think you're saying. I didn't do that, but I did something kind of like that. So, you know, >> Reddit is huge. So, is Reddit skewing this whole thing? >> Yeah. >> And I did look at that. So what happened was the the top 10 sites actually went up and the smaller sites went up, the middle sites went down. So the middle sites, and by this I mean I'll just give you like a little sneak here, but like if you look at all of these sites, I basically looked at all these and then I sliced it by size of site.

8:38 >> And so the middle of the pack went down about 6 to 7%. Reddit and the big sites went up, but also the smaller sites went up. So smaller sites went up the most actually. So so there's basically you know this dist this change in clicks where it's going a bit more to the big ones and a lot more to the small ones moving actually the middle of the pack I would say to the smaller ones. So you know our clients for example are probably going to be in this middle pack. So the kinds of companies that we work with would be in that middle. But but yeah that's kind of how I looked at but but I I think that the point is that Google isn't sending less traffic.

9:14 like maybe there's a shift in and you know companies that you and I would be working with did see a bit but even that you know it's six 6% 7% so it's not not like a massive decrease. >> I have I have two more comments about this. First of all we have this nice website chatgpt versus google.com and what we're seeing there is that we're tracking there the traffic share. So we're not tracking like growth or decline we're track we're we're tracking traffic share. So out of a 100% that websites are getting traffic.

9:50 What is the traffic share of Google and is it growing or declining and yeah it's been relatively stable. So it has moved singledigit percentages. So it's not that it dropped by 15% 20%. No, it was like six or 7% since the time all the kind of LLM and the AI chatbot craze has started. So this is the first one. It kind of your data checks out with our data and I think right now we track it over like 70,000 websites which are connected to HFS and every month more websites connect so we have a bigger panel to track. So this is the first thing. The second thing is that in terms of hrefs.com losing its own search traffic, what I've noticed is that it's mostly for the kind of terms and queries, wellformational ones that have become almost like common knowledge. So it's hard to get traffic for things like what is SEO when this is something LLMs can mention can can give you answer for easily because there's so much content about this. But I think to your point that new websites you saw that new websites got more SEO traffic.

11:06 I think one of the reasons for that is that new websites might be going after net new topics where there is like lack of information for LLMs to just aggregate you an answer. So they have to reference this website and yeah we we have even in our field we have a lot of new emerging topics like AEO GEO AO visibility a AI tracking and all those things are on the rise and whoever is kind of pioneering their content on all these topics they would be getting more search traffic.

11:37 >> Yeah and there's several several things in in what you're saying. So the first thing is that your the AHF's team did a study on the when AI overviews is appearing and it's appearing >> way more onformational queries on queries that you didn't make money on. So perhaps you got some clicks going away but you didn't make money on those. The conversions are not going away. The transactional queries are not seeing AI overviews everywhere. That's being disrupted a lot less. The second thing is the chat GBT versus Google share that you have that is a a 100% stack graph for you know it's like share of 100%.

12:10 But what's actually happening is the pie is getting bigger. >> Yeah. >> the third thing is that >> the that's referral traffic and most AI answers are zero click. So that's actually not fully capturing all of the AI piece. So the AI piece is probably even bigger. But again the pie is getting larger. It's not it's not zero sum that the the the pi is getting larger. So that's generally what I saw and one other thing. Yeah. So this is another interesting thing. So is Google going down and basically it's not. So what's happening is there is a reduction in click-through rate essentially from AI overviews and your team sent over this updated thing. So 30% of the time AI overviews is appearing and it does you know your your team also did this one here where you showed it was a 34% when when AI overviews appears it reduces clickthrough rate. So like yes that's true and yes it's true that it's showing up and it's showing up for informationational queries and the clickthrough rate's going down but it's not a result of fewer people using Google. It's just yes Google is taking some more clicks with AI overviews for informationational queries essentially which is you know not that big a deal in my opinion.

13:18 >> Yeah. Yeah. we're seeing the same like you said some of this data is coming from us actually. So I also saw the the study a while ago by Rand Fishkin and datas another kind of company that has panel data click stream and all that and he also saw the same thing that u a lot of people like the the adoption of AI chatbots like chat GPT perplexity Gemini the adoption is growing but not at the expense of Google. So it's not like that more people using chat GPT doesn't mean less people are using Google and it kind of checks out with my own habits. Yes, I use chat GPT a lot. but I also still using Google almost just as much as I used to. do I click less? Probably in a lot of informationational cases. Yes.

14:10 But if I need to go deep, I would always go to the source. So I I would always click. And just like with mobile apps, do you have you did you do you use mobile apps? You use them all the time, but that doesn't mean you you don't use you know your laptop and the and the web like of course you do. You use both and you use both the sum is way more than you probably used it in 2005 and that means the pi is getting bigger and then once again the pi is getting bigger with AI.

14:36 >> Yeah. I also like your argument in the notes that you sent that the use cases are different. So the way that you're using Chad GPT is not the way you're using Google. And yeah, a lot of the times I use Chedgpt to do things. I I ask it to I don't know refine my copy to I don't know generate something for me. But in Google I typically want to navigate to some website. I want to find I don't know website of a restaurant. I want to find a website that does a certain service. or if I'm studying some topic, I know I want to learn something about investing.

15:10 Yeah, I can ask a bunch of general questions to chat GPT, but if I'm researching, I don't know, a given stock, a given company, I want to go to the source and see like who's actually talking about it, what people are talking about this company, yada yada yada. So, there is still a lot of traffic to be had. >> OpenAI did a study with Harvard where they looked through all their prompts and categorized them. I don't know if you've seen this, but they categorized about a third of the prompts are new, behaviors, specifically writing, generation and self-expression. So, help, you know, rewrite this, help me write something, write an article, or generate an image or let me just, you know, let me pretend like you're my AI friend and tell you how my how I'm feeling. Like, these are not Google use cases. These are these are new use cases. And about a third of at least a third are are this group. So it's it's yeah as as you mentioned it's opening up new behaviors that are not taking from search.

16:07 >> Okay let's recap what we have so far. So the first myth was that SEO traffic has declined dramatically 20 to 60% like across the board. It's not across the board. lots of websites are still having gains but yeah a lot of websites like ourselves suffered. But in my case me and my team we're not losing faith in SEO. are still creating a lot of content especially we are prioritizing those new emerging topics rather than trying to go back to conventional things like how to do keyword research what is SEO SEO tips no we're not targeting that much those things anymore we're looking for new emerging trends where there's not not so much content and where you can still establish your thought leadership and be the source the second myth that we discussed is everyone is switching to chat GPT. and this is why Google share is going down again. we're not seeing this in data. We're not seeing it in usage patterns. as you're saying, the pi is getting bigger. So yeah, the fact that people are using chat GPT a lot doesn't mean that you will get a lot of traffic because chat GPT isn't really interested to send you traffic. And by the way, back to our Cad GPT versus Google website. I was supposed to post this on LinkedIn but I was lacking time. I actually saw that in December the traffic the referral traffic from Chad GPT the the traffic share went down because again Chad GPT changed something and they decided okay we we want to send less traffic to website. So they they have those citations but at any point in time they can make those citations more prominent or less prominent. They can feature them more they can feature them less. So they're totally in control. All of these hundreds of millions people that are using chat GPT they like the open AI can turn the knob and websites would be getting more traffic and they would turn the knob back and websites would be getting less traffic. So I believe this is a work in progress and they're still experimenting with the kind of outputs that Chad GPT is giving you. So let's monitor what kind of traffic you can get. But yeah, everyone is saying that answer engine optimization, it's about mentions. It's about your visibility in the answers. not so much about getting traffic to your website.

18:34 >> I agree. Although I if I were designing OpenAI, I would make answers way more clickable. So I would expect that they'll make it way more clickable. You're seeing that with commerce, you're seeing that with local where it, you know, like when Google started, it was 10 blue links and then they had maps, shopping carousel, all these rich modules. You're starting to see that with with all the LLMs where you have for local you have maps for shopping you have the shopping carousels. If I were asking I want the best you know I want the best SEO tool. I would like to be able to click on something. It's better than opening a new tab. So my guess is it'll become more clickable and you'll be able to track that stuff a lot better.

19:10 >> go ahead. Another interesting observation that I was recently playing with our own tool brand radar the tool for AI visibility and what what you can do with this tool is you can see all your websit's most cited pages by chat GPT or perplexity or whatever. So we have this huge index of prompts. I believe it's like 10 million prompts that we ask to chat GPT and based on those 10 million prompts we can showcase which pages of your website are appearing in citations and what I've noticed is that one of the most cited pages it's like top five cited page from hrefs.com was about our pricing and the date published of that page was 2022 and I I was immediately ringing the alarm to my team. I'm like, "Guys, look, Chad GPT is citing our page about our pricing." so it's not it's not a pricing page. It's a blog post that discusses some changes in HF's pricing and it's from 2022, which means that Chad GPT is most likely giving people answers that are inaccurate, that are outdated, and we need to go and update this page fast and make sure that all the information is recent, is correct.

20:29 So to your point about making answers more clickable, it's almost as you want to display we took this page, we took this information from this page and this page is this old. So please beware that like this information that we have found it might be out to date out of date. So we'll see what happens on this front. a whole new field of online reputation management and online reputation management spec specifically for AI and I think that this is also another interesting subject which is the other myth that I didn't include in this particular study but it's that SEO and AEO are completely different and they're they're different but there's a there's heavy overlap and I would say that there's probably 2/3 to 80% overlap everything in SEO works for free in AEO what's different is that you need to worry about certain things like your pricing page or your help center or your you know your features page or your integrations page which you would not need to worry about for search but the mechanisms to to to modify those are the same as search like we're ranking in this you know we're ranking in the citations for AI for our pricing so that's still SEO it's just that I'm worrying about pages and queries that I wasn't and prompts that I wasn't worrying about before but the majority of SEO applies for SEO The main difference is off-site off-site mentions, but even that is essentially an evolution of link building. Like it's not that different. It's just that it's just that it's a mention rather than a link and it's a mention on a specific URL rather than a link on the New York Times homepage. But I think that that's the other myth which is that they're completely different. And again, this comes from I need to disrupt I need to disrupt thing like if I haven't done SEO before and I'm doing AEO and I'm like, "Oh, this is big thing." I need to persuade you. Like I if if I'm pitching you've never done SEO before, you're going to say, "Well, where are your case studies?" And they're like, "Well, I don't have case studies, but no, neither does anybody else." and they'll say, "Well, this, you know, isn't it similar to search?" Like, "No, no, it's totally different. There's no nothing in common at all. We're all in the same level playing field. You should hire me even though I have no experience." And, you know, same same with software companies like AHFs, don't worry about AHFs. Like keyword tracking, you know, doesn't matter. You need answer tracking. This is totally different. and AHFs could never build something as sophisticated as answer tracking. So this is another what was that you already did.

22:54 >> We already did >> of course but this this is the other class of the myth which is that they're completely different and they're and they're not different. They're they're heavily overlapping. The vast majority of the strategies for SEO work in SEO. It's just that you need to apply them in slightly different things. Instead of link building, it's mention building. instead of you know optimizing how to do X or best tool for X, optimize your product content and your and your pricing page and your you know your about page and things like that, but but they're not different. They're almost entirely the same, which means that you can have one strategy for both of them.

23:28 >> Yeah. Yeah. This is exactly something I tweeted the other day and got a lot of support. I made an argument that if you admit that AEO and SEO are 80% the same, how do you sell your new disruptive next generation AO tool when you have HFS which covers that 80% and then we're building out those 20% as well. So yeah, a lot of people are both tool providers and consultants as as well are incentivized to position it as something completely different because this way they kind of can differentiate themselves from big established trusted agencies and consultants and tool providers that focused on SEO. But the thing is those big established providers, tool providers, agencies are the best suited to do AO as well because like you're saying they're 80% the same. it's just AO has more emphasis on kind of earned media or like on what others are saying about you where SEO was mostly focused on your website. But then again SEO, search engine optimization, your goal is to show up in search. So, what a lot of smart people were doing, they were claiming not just a search result that you own with your website, they were going to other top 10 ranking results. As long as they're not your direct competitors, you can go there and say, "Would you like to be my affiliate?

24:59 Would you like to mention my my product, my service on your page and get a kick back every time I get clients from you?" And this is still SEO. You're owning not just a position with your website, you're owning a mention in all of the top 10 ranking pages. And this is the same that you need to do for AO right now. You need to be present everywhere so that wherever LLM would go and look for answers to a people's person's question, they would find mention of your product on a lot of websites, not just your own website. So this is this is the only 20% difference. And then again, we could argue that that 20% difference existed as an SEO strategy.

25:36 If you think outside the box of your own website. >> Yeah. And the best way to do this is to ask chatbt, ask perplexity, what's the best credit card and search it on Google. The whole page will be pages that say we may have taken an affiliate possibly a commission. if you if you click on this thing that that means that it's an affiliate and for for both Google and for all the LLMs you'll see the exact same thing. I will say I do think that there will be one class of stuff that will be entirely different possibly and that would be autonomous agents. And so I think that the LLMs are going to try to just like you know just like Apple wants you to stay within the phone and not go somewhere else. I think that they're going to try and push the conversions within the the native experience. And I think that at some point that the LMS will optimize for something like plan my vacation. plan my vacation and do it for me and I don't want to deal with it. It's kind of like the super EA. So like book my hotel, book my restaurant, book my flights and do it without me doing anything and I don't need to go somewhere to do it and then the agents will then go convert on those destination sites. We see chatbt launched atlas their own browser and I believe they were marking up their they said mark up your buttons in your forms essentially so that I can perform these actions. So I can foresee a scenario well I mean they want to do that. So if that works that would be a net new different thing and then you would have ads as part of that. So that I think would be completely different but basically everything else is is just SEO v1.5.

27:09 >> Yeah. Also with with all the answer tracking with all the AI visibility there's a lot of questions there on the methodology because the first thing we did at HFS when we released the functionality to add your own custom prompts. So like I said, brand radar is based on an index of 10 million prompts that we already asked and you can tap into this data and see if your brand is mentioned about among those those prompts and those responses. But when it comes to custom prompts, when you want to set up a list of 10, 20, 50, 100, 200 prompts where you want to see if you're showing up or not, it comes with a ton of bias. So your selection of the prompts is influencing your AI visibility more than anything else because you can ask prompts where your where the propensity of your brand or your product or your service to show up is just higher. So the high the the higher up kind of the funnel you go from from the awareness stage to consideration stage to blah blah blah the less your product will probably show up. And the thing is most of those AI visibility trackers I see their packages are starting from like 25 prompts, 50 prompts. What kind of AI visibility? How can you rely on 25 manually selected custom prompts to gauge your AI visibility? And yeah, so so I asked my team, I asked Patrick, I asked Glenn Alup, let's published, let's publish a few articles on our blog on different methodologies of selecting your custom prompts because a lot of people wouldn't put a lot of thought in it. They would, I don't know, go into chat GPT and say, give me 20 questions that people might be asking when, I don't know, choosing a CRM to go for. It would generate 20 questions. they would put it and they would track the their AI visibility based on that. But there's a lot of ways you can go top of the funnel, middle of the funnel, bottom of the funnel. One one methodology that we're going to explore and that we're going to integrate in our product is we're going to take all the content of your website, everything you published on your blog and turn that content into questions. Because if you publish this content on your website, you care about this topic. this is something your product helps with and you want to know if people asking questions about this topic if you're getting getting mentioned. So converting your existing content the content of your website into questions and generating like a few thousand of them if you have enough content on your website can then give you a good visibility into like where you stand. But yeah, I wonder what you think of the of the custom custom prompt tracking and if you've seen like what what kind of strategies, what kind of methodologies people are using to to set them up.

30:03 >> I've looked into this quite a bit. I actually have an 80% done article on my own. So, lots to share on this. So, an analogy would be what if I had no idea any search volume data. So, I'm going to use refs and start tracking keywords and I have nowhere to start and I just start tracking keywords. I'm like, well, I'm position four maybe that like I don't know how many people are looking for it, but it seems like it's good.

30:26 >> So, like what would that be like? And we explored doing keyword research with with the LMS like hey what keywords would I potentially want? Like u an example is for for rippling what key what things might the director of finance search for to decide their payroll management software. And it gives you know all these different themes like well you know need to go through these 10 different steps and here are some keywords almost all of them are zero search volume so like they look great and there's no search volume for almost any of them and that's for search. So like we know that for search which is actually easier that that the LLMs are very bad at making these suggestions even like your own intuition it's very bad if you look at the search volume in NHS for most of the keywords you come up with there's no search volume.

31:11 >> Yeah. >> So we know that it's bad. So then if you're asking for questions, it's going to be even worse. And actually one other thing that is different with LLMs is that the tail is much much bigger because the average number of words for a search I think it's like 3 to seven depending on your data source whereas for qu for prompts it's 60 words. So the majority of prompts have probably probably search volume of less than 10 or like prompt volume of less than 10. So this is massive tail. So then you're guessing about what the prompts might be. So now you have a measurement error. Then you're guessing about the volume of the prompts. Even more measurement error. Then so we looked at prompt we looked at we what we do is we just take search keywords and essentially we just transform them into prompts. So I think that that's actually a pretty good method. This is what I do think that >> Yeah. So I think that that's great because it I I think it's okay if it's not exactly the prompt the correct prompt verbatim. If it's conceptually the prompt, you're probably going to get roughly the same data. And we assume that there's roughly similar search behavior with prompt data. That's not entirely true. You're going to miss the generative stuff. You're going to miss some of the tail, but it should generally be accurate. Now, eventually we'll have panel data. It's like for for keyword tracking, we just ask Google, right? And so Google says here's the source of truth. Maybe at some point the LMS will give us you know their own their own data, maybe.

32:34 >> But in the meantime, we have panel data now. And and this is what prompt volumes is generally based on. So what is panel data? Panel data means I take a subset of of actual first-party data and then I multiply it by something like if I have 1% of everyone on the internet and it's unbiased and it's representative. You multiply it by 100, that's prompt volume. And that and that's legitimate. Like when you do presidential polls, you it's usually survey a,000 to 2,000 people. They're not exactly perfect, but they're actually pretty pretty accurate.

33:02 So that's essentially what panel panel data is. The thing is the p the panel data is extremely noisy. Like I I've I've spoken to to multiple AEO companies. They just don't have good panel data. If you had good panel data, that's actually a legitimate approach if it's large enough and it's unbiased and representative. It just doesn't exist. It's it's too noisy. And so I think panel data today is not ready. It might be ready at some point, but today it's not ready. So I think search volume is is a good method. The other is where are people actually asking questions? So they're ask actually asking questions in your sales conversation on Reddit, maybe on Kora, maybe on Stack Overflow if you're a developer. Go find places where there are real questions. Like that's your panel data right there.

33:46 especially sales calls if you if you're a salesled company, find those questions. Those are probably the questions that I would target. So I would focus on that. I don't think the panel data today is ready. I think it could be ready, but based on based on my research, it's not ready. The last point that I want to make is on grouping and grouping is very important because it's just like keyword grouping. I like as a marketer I I want to do an initiative. So I need to know the group of keywords that I'm going to do the initiative for that I'm going to target.

34:16 I don't want and depending on the group it could be wildly different. It could be like a thousand times bigger depending on that. Like let's say I want to target chicken. Well I'm not targeting every single keyword that has ever contained the word chicken. like I'm probably targeting how to cook chicken, how to do blah and then there's a bunch of other topics and you know you have different strategies. So per for prompt volume again the tail is so large the grouping algorithm is very important and most grouping algorithms are based on just did I see prompts that happen to contain these words like you you mentioned one where or you know like generative prompts is it a is it a search prompt or is it saying write an article about you know this thing well that's not a like I'm not targeting that as a marketer so >> is it matching my intent is the primary subject the prompt so most many prompts will contain words where the primary subject is not the prompt that you're looking for like project management software. Not every pro prompt that contains the word project management and software is about looking for that particular product. So you need to know is it a search and discovery prompt is the primary subject the prompt and then is it grouped properly and we've been playing around with a grouping algorithm based on embeddings. It's actually pretty good. But anyway those are my thoughts on prompt volume. Thanks a lot for sharing all of that. because this is exactly what we're seeing as well.

35:36 anyone who is claiming that they have accurate volume data for prompts they're just essentially lying to you because yeah we we talked to a lot of providers as well. We talked to a lot of providers. A lot of companies are of course reaching out to HFS trying to sell their panel data to us and like you're saying it's extremely noisy for the reasons which we just discussed because use cases of chat GPT and use cases of the conventional search engines like Google are very different and this is why on chat GPT you have long prompts and you have conversations contextual conversations and it has history about you blah blah blah it's it's very very hard to derive insights from it.

36:22 And yeah, we're using search volume, but full transparency, I'm not happy with how we're using search volumes to extrapolate into the prompts that we use in brand radar. But the team is on it. We're working. We we're going to change the system because right now I think we are kind of extrapolating the search volume of a parent topic onto the questions kind of too bluntly. It is like too straightforward. we need to have a more sophisticated model of how we how we do it and the team is working on it. I think in the next couple months we're going to release it. what else I wanted to say about this? yeah, grouping grouping is is a cool thing.

37:04 This is something we're also looking into. Instead of kind of trying to evaluate the individual volumes or popularity of different questions, we would be looking for patterns like what what kind of topics what kind of yeah what kind of topics these groups of prompts belong to and we would kind of try to gauge them by topic. But again when it comes to when it comes to tracking your your AI visibility with a list of custom prompts, this limits you a lot. This limits you a lot. So yeah, it's yeah, it's it's a very hard problem to solve. It's a work in pro progress >> and you have compounding errors. I don't know what the prompts are. I don't know what the volume is. I don't know whether or not they're about the primary subject. I don't know if they're search and discovery or generative. And I don't know, I don't know what the group is. So there like the all these errors just having this wildly extremely wide confidence interval that's almost meaningless. Which is why again I I like search volume. The other thing that I would do if if I were to to try and be foolish enough to to build a tool around this, I tried I tried to build a tool, but it's it's pretty hard to compete with >> with AHFS, but if I were to, I would build >> mining like Reddit and Corora and some of the sales calls like that's actually real data >> that I think that that that's real data that's available right now that I think that would be quite interesting to layer on top of search volume while we wait for panel data to be good enough >> a few comments about sales calls.

38:36 first of all, I wanted to point out, yeah, it's a great source of potential questions. But I feel the questions you get from sales calls, they would fall under the evaluation stage because people are already aware of your solution and they they know a little bit about it and they're evaluating it. not so much discovery stage where people kind of are earlier in their journey and they just need to discover that their solution like yours and they need to like come to your website, schedule a call with a salesperson and ask this question. But yeah, I'm sure that sales calls is a treasure trove of evaluation questions anyway, which are still super important because a lot of people do use AI assistance to help them compare different solutions and explain their case and rely on AI to to give them advice why a certain solution would be good for them. Another thing just quick and slightly unrelated is that I don't think it's it's that easy right now to get aggregate questions from sales calls. I believe Gong has been promising that they would create this kind of like AI transcript analytics where at a scale of thousands of calls they would be able to give you questions, topics and blah blah blah. I don't think they released it yet because just just a few days ago we were discussing getting gong for HFS because we're still not using it. and our sales leader she was saying that they don't yet have this cool functionality that that you want team because I said oh we could use those questions to to help us improve our product to help product marketing team to understand what kind of content we need to create and she was like no it's it's still hard to get this data and I believe a lot of people are using hacks where they get the transcripts from from their kind of call logging software of choice and then they give it to AI agent to process and they extract questions and then they group them. So it's right now it's very manual. have have you seen any kind of plug-and-play solutions for this?

40:47 >> No, definitely not. And we're doing something similar. Almost no one's going to build this. This is why some enterprise company like an HFS or somebody else needs to build this because this is like a huge effort. We take our >> we take our sales calls via Fathom, send the transcript to BigQuery, then we just have a giant BigQuery table and we're not doing anything with it. But at some point we might look at it. So because it's like a whole thing, you know, it's kind of like setting up event tracking in 2003. Like you have to do all this work to set it up. So basically I don't know anyone who's doing this.

41:20 >> We might play around with this with a couple people, but it there's no plugandplay. This is easy. You have to really invest in building this custom yourself. Another thing that I think is becoming super important in this I wouldn't say transition from SEO to AO but rather expansion from SEO to AO is yeah your mentions around the web. and this is something HFS is actually uniquely positioned to solve because we have a freaking index of the entire web.

41:49 We crawl the entire web and we have this cool tool which has always been underappreciated. they feel it's called content explorer and it's basically an index I don't remember how big it is 10 billion pages or something basically it's it's the same as using Google so for example you want to find all pages that mention parenting advice so the phrase parenting advice and you want to find all all web page that have mentioned this phrase so in Google if you search for parenting advice in quotes it would give you what a 100 pages like 10 10 pages pages with 10 results 10 by 10 100 pages. That's that's all you get. if you search for parenting advice in content explorer, it would give you hundreds of thousands of pages because we have the index of the web and we don't limit you to just a 100 search results. And then we have all the SEO metrics. So we have the domain rating, we have the domain search traffic, blah blah blah. So you can filter the results and find the biggest websites that talk about parenting advice. And you can also do it with your competitor. So if you're seeing that your competitor is showing up in AI more than you do, instead of trying to rely on a few hundred prompts that you have set up, go and see how much more their brand is mentioned across the web compared to yours. Of course, it gets a little bit tricky when we're talking about brands which are just regular words like Apple or something. It's it's a lot easier when you have a brand like Hrefs which is a unique word and when when someone mentions HFS on their page, this is us. There is no kind of second guessing. But we're also solving for that for that. So we're going to identify brands on the pages and we're going to expand the index of our content explorer tool. So basically instead of looking at the outcomes when you do when you do AI visibility tracking you're looking at the outcomes you can go to the source and you can understand why why are we having this outcome why this brand is mentioned more because it's mentioned more around the web and you can find all these websites you can find all these pages and again you can reach out to them and say you're mentioning my competitor let's figure out what what needs to be done to to have our brand our product mentioned next to it on the same page. And yeah, one final thought, one final thought.

44:17 Of course, you want to track the citations, the actual pages that chat GPT or other chatbots are using right now to pull information from. But will they be using those pages tomorrow? Maybe they would tweak their algorithm somehow. Maybe they would tweak their fan out queries and they would be using different pages to to pull citations from. So when you're checking the current citations, yes it is important but this is what works today but what would work tomorrow? So you need to go to the source you need to see how you represent it across the web and this is where a tool like HFS which does the freaking crawling of the whole web and can give you all the content that has mentioned you or your competitors or a given topic. This is where people would start paying attention to it. and we're surely going to do more marketing around this because it just makes sense to do.

45:07 >> That's great. And this is actually another thing that is different from search or at least an evolution from search. And and by the way, I didn't actually know that this feature existed. I don't know if I can get a an upgraded account, but I would love to use this. But the reason why >> No, it's included in in like in standard plan in advanced plan. It has been there for Yeah. It's it's very underappreciated. Content explorer is very underappreciated.

45:29 >> Yeah. Yeah. And it's not that useful for search, but it is for AI. And the reason why is because all that matters in search is are you in the top three. I don't care about any any other rank really because you don't get any clicks. Whereas in AI, it matters a lot for two reasons. The first reason is that when you ask a question like what's the best project management software, there's the core model and the rag response. So the rag response is where searching the web get a bunch of citations and it's looking at the citations and it's not just looking at three citations. it's looking at 20. It's looking at many citations. And so position 10 and position 15 of the citation set can matter a lot versus search where it's just the top three. The second thing is the the the the core model piece. So the core model is training on many you know the web and and just as as large a data set as possible. And so if you and it's looking at you know co-occurrences of words like what's the best project manager see at this brand like what's the best SEO tool AHFS. And so even if that's not in the citation set, the core model sees that and the more the better. as you mentioned, is that citation going to be the citation in six months? And I actually think that the answer is it'll be very noisy.

46:35 And the reason why is because if you look at the citations in all the LLMs, they're very relevant, but frequently they're kind of weird. Like how is this strange site appearing? And the reason why is because it's not that hard to build a search engine. So search engine goes into citations. It's not that hard to build a search engine that has a good relevance score. It's very hard to sort. Like it's very hard to build the authority score.

47:02 >> And that's why in all the LMS, maybe Gemini probably has this figured out, but all the LMS are struggling to like we're fine with relevance, but we have all these strange sites showing up and they're going to have to spend a bunch of time building out their own page rank, their own domain authority. And I think that the citations that appear will be very different o o over the coming years which to your point like how do you know if this one's going to appear in six months just but go after all of them and again so it's the core model piece and the rag piece.

47:29 >> Yeah. Ethan we've covered a lot of ground already. I don't think we need to make this episode more lengthier or more packed than it is. I'm sure people's the heads of our listeners are blowing up with information already. any kind of last arguments that that you wanted to make something we didn't discuss but is important to kind of close out the episode. Yeah, I've been looking at is AI overhyped and I actually thought that it was like I'm I'm an a AEO influ I never wanted to be an influencer but somehow I became an influencer but you know is AI overhyped and I thought that it was and it's actually not like as I've been looking at the data it's actually not overhyped in fact I think that it's underhyped so usually I try to have cerebral recommendations about not overstating how big AI is but it's actually quite big and so my general recommendation is there's you know these points in time where there's a new technology and there's a land grab in the first few years to like take over the whole you know take take that over and then it becomes a mature field. This is not a mature field. AO is not a mature field and so I think that it's actually I would recommend everyone panic and run out and start optimizing this like I actually do think that that's true in this case but again it's additive to search. So have a single holistic strategy but apply your SEO knowledge to these new things like mentions, help center, pricing page, stuff like that.

48:57 >> You reminded me of something something I think also is very important. So again back to AI visibility tracking. You can track it in Chad GPT, you can track it in Perplexity, you can track it in Gemini, you can track it in Copilot. But the thing is all of those systems have different audience sizes, right? So, Chad GPD is pretty big. We know Gemini would be big. Didn't they recently sign some deal with Apple or something that Gemini is going to power Apple? So, the the answers by Gemini is going to expose. Wait, let me let me finish. Copilot, Microsoft would probably be pushing it everywhere in their products as well. So, they would try to integrate it into Windows and whatnot. So everyone would be battling for visibility. I think perplexity would be the smallest player. Now the thing is the thing is when we're talking about conventional SEO Google being duck go no one cares about optimizing for duck go and typically no one cares about optimizing for Bing because it's just so small compared to Google. And the strategies for optimizing Google and Bing are I don't know 95% the same. I don't even know what unique things you need to do to to show up in Google that wouldn't help you to show up to to show up in Bing that wouldn't help you to show up in chat GPT but AI chatbots are different because there were a lot of studies in terms of citations and we know that where chat GPT pulls citations from is different from when Gemini pulls citations from it's different from perplexity is different from copilot so once you start tracking your visibility in Chad GPT Gemini Copilot. Do you really want to spend resources individually? Oh, like I I'm showing up well in Chad GPT, but now I want to show up more in perplexity. Do you really want to spend more resources on that specific AI assistant or maybe you want like a broader strategy that would just generally help you target all of them at once? So, what are your thoughts on this?

51:04 Yeah. So, I have data on this and I started SEO in 2007 and I did Bing optimization and Yahoo optimization in addition to Google optimization and it was great. they're similar but they but but but they're different and that was really fun. I have data on let me see if I can do this. Yep. So the this is the market share for search. So this is worldwide. Google has 94% market share and I think 90% in the US but you know >> just worry about Google >> and AI it's basically chat GPT and and Gemini have like 95% plus or 90% plus I think so it's clearly those two but we don't know who's going to win because I think Yahoo was bigger than Google for a while and and AOL you know like AOL and Excite and Dog Pile like who knows is who knows who's going to be dogpile and is going to be Google. We don't know actually. There's all these LMS. There's Grock, Claude that the largest ones are Chachd Gemini, Grock, no, no, sorry. Perplexity, Grock, Claude, but any of them might win. To your point, Google just signed this massive deal with Apple. Is somebody going to sign a deal with Microsoft?

52:18 Probably. Meta has Meta AI. They have a massive surface. So, are they going to be like they could easily be be huge as well. So, to your point, yeah, who who's to know who's going to win? I'll say yeah but yeah you should track your presence on all these the other thing is you need to ask the question many times because you get different answers and so you need to do multiple runs so you do multiple run like 10 runs of the same question on each surface across all the surfaces and then you get this probability distribution distributions of distributions and you have this really complex probability density function of your rank m much more complicated than I'm position three but that that's basically how you need to do it.

52:56 >> Yeah. So back to the question, should you care about optimizing for perplexity or because my point is you need to optimize for the biggest. So whoever is the biggest one, you need to show up that your AI visibility is best there and then you need to do a little bit of extra effort just things that make sense. Like I said, you need to go to to a tool like content explorer. You need to you need to see all the mentions of your competitors. You need to see all the websites that are covering your topics and make sure you have visibility there. And without specifically targeting perplexity by doing this thing that just makes general sense, your visibility and perplexity should improve as a result. So this is my point.

53:39 Optimize for the biggest and do a bit of extra work that just makes sense for all of all of additional chatbots. >> I think there's two answers. There's the infinite resources corporation answer and there's the startup answer. >> So the startup the corporation answer is well there's all these trends and we need to think about these trends and perplexity and blah blah blah and like and I have all the all the resources that I could ever want. I can do whatever I want. Then there's the startup answer which is I'm dying in 18 months and I need to make it to the next round or I'm dead. And for that just focus on the biggest channel which is what I would do. So if it's me, I would just focus on chat GBT maybe Gemini and I wouldn't worry about anything else.

54:22 Wait until those are important and then you can worry about them. But there's so much wasted time optimizing for some sort of imaginary hypothetical thing that might happen at some point in the future. And I I I haven't met many companies that have infinite resources and can sort of just spend resources on theoretical hypothetical things. Almost everyone that I know has finite resources. So if it's me, I would just optimize for the thing that's large right now. and but but it is large enough to like it the channel is big enough to worry about right now and to spend time on but yeah just do chat GPT only or if you have some extra time do Gemini Ethan a lot of what we discussed today especially all the new data that you have shared is going to be published soon where people can find it >> so people can find it on our blog so graphite.io io5% which stands for actually 5%.

55:15 >> We'll put all the links in the description as well. I just want to make sure like it's it's going to pop be published on graphite blog obviously on your LinkedIn. So people should go and follow you on LinkedIn and look for mentions of that. will you syndicate it because the the previous study you you syndicated it with what? Axios or >> Yeah, we'll see. We'll see if somebody wants to pick it up. okay. We we may or may not. well, we'll do our best to see if anybody else will pick it up, but we'll definitely put it on our blog. We'll we'll send it to our newsletter and LinkedIn and, you know, hopefully somebody thinks it's interesting to pick up pick up. I think it is.

55:50 >> Okay. So, if if our listeners want to see all the fancy graphs and charts and numbers and the the studies and the data, they can find it on your blog, they can find it on your LinkedIn, Ethan. amazing. Yeah. Oh, and we included the raw data. So, we included the source and the raw data. So, everything is reproducible. So, you don't need to trust us. You can click and see the actual raw data all in our blog.

56:15 >> Ethan, thanks a lot for this episode. I really appreciate how deep we went and how much data you use to support your claims. And like I said, you're you're definitely the the top expert in AEO and SEO. And yeah, if people need to trust a voice with what what is happening right now to SEO and how it expands into AEO, I would totally choose you as the source of truth. Thank you. Thank you so much.