Transcription
All right. Well, we're a few minutes past the hour. We got a fully packed room and we have Ethan here to give us the latest and greatest on how to win AEO. And with that in mind, I will mute pass the mic over to Ethan. And if y'all have any questions, Ethan has said that he may answer some of them live. So definitely drop them into the Q&A box and we will get to them either live or at the end of this webinar presentation. With that in mind, Ethan, the floor is yours.
Awesome. Thanks, Bernard. Long time no see. So I've [clears throat] given this present I've I've t done about 12 different webinars and I want to make sure that I'm not repeating the same webinar. So my plan here is fewer slides, more live demos. it seems what is the most useful and then as people have questions uh the more specific I can be the better. So if you have questions just put them in the chat and Bernard will interrupt me and ask me but um that's the plan and uh I have a new case study that I'll share with legit security as of this morning so uh accept and shut up.
Okay so I'm Ethan uh our company is called Graphite. We are a digital marketing agency. We specialize in SEO and now AEO and we do that with a bunch of companies. Here are some uh rippling, notion, web flow. Uh and we do so we build out entire strategies. We don't just do audits and like slide decks, but we're actually building stuff. So we actually see the outcomes of our work. Why that matters is all of the stuff that I'm suggesting are things that we've are building or actually building and some of which we've seen outcomes for. uh we're early in answer engine optimization and I don't think anyone actually knows uh what works like we do for SEO where you know Bernard and I have been doing that for 18 years so we're early so I'll share with you either what I do know and I have two case studies web flow and legit security and then what I think is probably true based on analysis so that's the structure of this.
Okay so uh AI is growing I think we [clears throat] all know this but it's especially been growing since January where there's been a big surge in this is from similar web a big surge in uh usage of LMS and then also the answers are starting to get a bit more clickable the uh chatbt still has most market share but uh Gemini and complexity are large and Gemini is actually growing I think faster than any other LLM uh right now uh like number of downloads and usage Gemini is blowing up this is actually not this is actually one month out of date But the the the market is heating up. The the second thing is last year, you'll see last year for Web Flow, no conversions. So, Chad GBT launched in 2022, but we didn't start seeing a lot of actual business value until the last few months starting in uh January again, but we're starting to see it's an actually substantial channel. And uh and therefore, it's not like a theoretical thing. Should we focus on this? It's it's uh meaningful right now. So, I think now is the time to focus on. Usually, when there's a new channel, I say don't worry about it until it's large enough to worry about and it's now large enough to worry about.
Okay. So, when LLM's first launch, it was based on next word prediction, which is I asked what the what the capital Sacramento is and then chat GPT. It had a model trained where it looked at billions of web pages and then it could predict the next word which is Sacramento more and more the alms are based on search and here's an example. So what is application security posture management and here it'll be doing a search. Yep. Searching the web. It's finding a bunch of stuff. It's giving an answer. And then you will see at the bottom here that the search results are not shown. So it's not 10 blue links. I'm going to see almost done. Okay. So here are the sources. So it started with a search. So it did not start by asking the model predict the next word. It did a search and then the model summarized these. So there's two pieces to the answer. There's the core model and there's the rag response. The rag response is this set of search results. rag is just a fancy way to say search results, retrieval, augmented generation. So the R the uh the R in rag is this just search results and then the core model then summarizes what it sees here. But it's not the case that the core model is this mysterious thing that you you know you don't know how it works and it's unpredictable. The core model frequently is perhaps mysterious but the rag response is the majority of the answer which is less mysterious. So you can see the search results here and so the answer is a summarization of this and that's how the model works.
So then uh how do we actually show up like you know how are these relevant and what you'll see here I'll jump over to this other one here. So I asked what is what is this and then it had the search results and then a company we work with called legit security appears. So, why did they appear? They appeared with an SEO article. So, one thing I'm hearing a lot is SEO and search engine optimization are completely different and you should just stop doing SEO completely. And actually, what you'll see is if you just do traditional SEO, it works for free really well with AEO. So, here's an an article. It's the same exact same thing that we would do for for SEO. What is application security management? We have this article here. This showed up because we used we probably use something like clear scope where it says you know include things like how does it integrate and what are the benefits and what is this thing. So we optimize the content uh and you know we optimize the content with a with a like like a clear scope. The other thing that we did here was look at all the uh look at all the keywords that this ranks for. So it's a combination of a clear scope like tool and then and then looking at the uh keywords that we acquire on which we'll see here. So we're ranking for 18 different keywords and you'll see some sub themes like posture management tools security and uh yeah applications. So we want a list of tools and that's the general relevance for how this URL shows up. The second interesting thing that we did here is we're starting to play around with uh we're starting to play around with videos that are not entertaining. And why I mention that is because YouTube is filled with entertaining videos like food, travel, beauty, and um and so those categories are crowded on YouTube. YouTube is not crowded with videos about boring things like application security, posture management. Not many people spend their weekends watching videos about this, which means that it's not crowded. And so it's but but you know, it's a very high LTV. So it's it's the most interesting uh products to do video on are the least entertaining ones because they're not crowded. So here you'll see we have this piece of content. There's this YouTube video that's embedded. I will open this. This boosts engagement. This is also not a super produced. Yeah, it's not like we spent thousands of dollars on this. Uh, this is easy to make and it's increasing the engagement on the page and it's starting to show up in it's starting to show up in AI overview. See, so how to secure generated code and you actually see the video here. So this relatively inexpensive video is helping the URL show up in chat GBT. It's boosting in in uh AI overviews. I believe I saw it. Yeah. So, similar with Gemini, the same article with the video shows up here. So, this one strategy is showing up all over the place. Uh here we also see what is this here and this also has a video. So, basically doing just straight SEO like you would 5 years ago. uh informing the structure of the page with clear scope with looking at all the different keywords and then embedding videos and the video then boosts this one page across all search engines. It boosts this video within YouTube so it gets more views. We just launched this so it's uh no wait sorry we just launched this one uh or we're starting to yeah we just embedded it. So it's basically creating this loop where the URL shows up in chat GBT uh you go to the page you view it it boosts in YouTube then that shows up in Gemini it shows up in video search it shows up in AI overviews and so creates this loop this is especially interesting for uh non-entertaining B2B categories.
The second thing we see is okay. So that's earned that's uh or sorry that's that's a that's a owned URL on your site. So the second thing that we're seeing I asked this question here which is what is this? Are there any good tools to do this? Cuz like did we get value out of this URL? Maybe maybe not. So are there any good tools that do this? And you will see that it will include among others legit security and it's mentioning legit security because we have this page and an interesting thing here is uh yeah this is a list of the types of features that this particular tool allows you to do and so this page on legit security now means that legit is a specific product that is recommended within the answer. So these are two different strategies. One is where we added value to the answer but we're not necessarily a tool mentioned and then the second thing is where we are part of the answer which you could see here. The other thing to show here is uh that I want to show you is um Reddit and so how does how do you do secure AI generated code and uh you know there's a few different themes of the kinds of citations. So there's UGC which is Reddit, Corora, Stack Overflow. There's video which is YouTube, Vimeo, Tik Tok, uh, Instagram. There's your site, which is owned. There's affiliates. And so to win, you want all of these things. Uh, if you just optimize for one, you it's not enough. So here's an example of a Reddit thread with uh, Web Flow. So for web flow we do a bunch of stuff and um we've [clears throat] we've played around with Reddit and it's hard. So basically it's it's common for your comments to get down weighted or your thread to get removed because the the community is very intense. So we've been playing around with how do we do web flow. the the the uh strategies that we've seen work is to just say who you are and give say who you are, give useful information and uh like this is my name, this is where I work and here's useful information. So, this sounds obvious. Back to your question, Bernard, how do you optimize for Reddit? You create hundreds of fake Reddit accounts and then you post things and then you upvote each other's stuff so that you get more karma points and then you start spamming all of Reddit by saying you should really consider using clear scope. And the way that I liken that is it's kind of like if Bernard showed up at a dinner party that he wasn't invited to and sat next to someone and like have you considered using Clear Scope? You know, that would be kind of weird. But if you said, hey, you have a question about optimizing content. My name's Bernard. I work at Clear Scope and why don't you consider filling in some content gaps? That would be perfectly fine. So, I've seen that work. I've also seen uh I've also seen this is a fun example. So, this is a this is an example where there's a thread about is Ethan Smith another SEO influencer? What am I missing? And so, then I uh commented on it by saying this is my name and this is who I am and here's a thoughtful response to whether or not I'm overhyped. and this did really well. Uh so that's my general suggestion about how to how to optimize for Reddit is just to be an actual authentic person. Say who you are, where you work, build karma over time. The other uh interesting thing is just Reddit ads I think are a very under underutilized category. So So that's Reddit.
Uh, let me jump over to a another example and Bernard, let me know if there any questions. We got we got some some questions for you. So, specifically around the video concept, it works for non-entertaining topics like you said, but how do you decide what topics are worth creating videos around? Asks Molly.
Yeah. Okay. So, what's what's the intense like what what are the five steps to go through for this? So, let's say that I'm a brand new I'm a brand new uh Ripling competitor. So or let's say a Gusto competitor. So I'm I'm I I I want to go after gusto. So what I would do is the following. First thing that I would do is what questions should I care about in the first place? And how I would do that is there's two different ways. There's there's uh looking at search data and then there's looking at some of the panel database tools. I personally like the uh search data just because I have clarity about where the search data comes from. So I would go to Austoto and I would start by looking at what are the top paid keywords that they're going after and those will be my keywords. Like let's take direct deposit form. That's step one. Step two is I would ask uh please make this keyword into a question and you'll get a pretty good response back. So now you don't know for sure the volume for this but you do know the volume for this and you know that probably people are asking questions about this. So this is a known this is a known data set whereas handle data is unknown. So what I do is I just take top paid and and organic keywords and then I get questions out of them like this. The second thing I do is I put them in a tool. So I put them in a tool. We have our own tool and then there's also a bunch of these other tools. I'll just show you peakai. So peakai is a tool and then you put a bunch of questions in and then you start getting citations back. And so then when you start getting citations back, you can then see what channel should I care about in the first place. So best website builder for designers. This might be very different from uh you know best television like the affiliates or sorry the citations that are appearing for for website builders are different from from televisions which are going to be different from travel. So you you don't want to just optimize generally for any type of citation. You want to know the specific citations you want to go after. So I have my I I took my 100 keywords then I transformed them into a 100 questions. Then I put them into a tool. Then I look to see what cit what themes of citation they're showing up here. You'll see it's affiliates like Techraar. It's video like YouTube. It's Reddit threads and it's Yeah. And then it's owned sites like a web flow. So these are my four categories. Video affiliates, Reddit owned site. Then I want to come up with a strategy. There's on-site, there's off-site earned, and there's off-site owned. So, on-site earned, sorry, on-site is just a page on my site like uh like this page. [clears throat] Off-site owned is my YouTube account, like Web Flow's YouTube account. Off-site earned would be somebody else talked about me. Uh, does that happen to your question?
I believe so. Yeah.
Yes. Yes. Molly says yes.
Great. Okay. So, so then uh I I'll talk about web flow a little bit. Again, I have two real examples cuz uh I have lots of examples of causing outcomes in SEO. I have two in AEO, which is better than zero. So, I want to make sure I talk about the actual causing of outcomes. So, web flow is the second uh causing of outcomes example. And here I mentioned the Vivian uh at Webflow comment here. Let me see if we have that here. We don't. Um, but YouTube has a big uh a big sorry, Webflow has a big YouTube account and they're basically creating tons of videos about how to do X. It's not just how to use my product, but how to build a site, how to do X. It's essentially SEO topics. And what that means is that Web Flow, when you say, "What's the best website builder for designers?" They can show up with a page on their site, uh, a YouTube video, a Reddit thread, an affiliate mention on Techraar, and you can show up multiple times, and then you can, you know, basically own that answer by showing up over and over and over again. Uh, so that's uh, YouTube. The the next case study, so the next case study is with Ripling where we did SEO and it just worked for free in answer optimization. So, it wasn't necessarily a focus specific AEO strategy. It's just that the SEO work worked for free and chat is built for so let's do what is the best payroll management software software for HR and you'll see here hopefully it'll say rippling somewhere and for for Google you would probably type this in and you would get an answer then you would leave but chat is based on chat is based on conversations with ongoing questions. So you'll see here we have this page for rippling with all these features that they have. So employee database management and time and attendance tracking. So which these uh employee uh database management and attendance tracking. This is what chat's built for. And the average number of words for a question and chat is I think 25 or something pretty substantial versus a unfortunately rippling's not showing up. part of part of the part of the joy of doing a demo. But why is why is Gusto appearing? And you'll see Gusto's appearing because they have a landing page specifically about the feature that I asked about which is employee database management uh employee management software and then on pay shows up for their help center. So I I mentioned that SEO works for free. What is the difference between answer engine optimization and and SEO? And the answer is everything you do in SEO works for free and AEO. The differences are the head and the tail. So the head is offsite and then the tail is very specific follow-up questions like which of which products do have this feature, have this integration, have this language. And one of the best places to optimize that is the is your help center cuz your help center describes how your product functions. And usually in essay you wouldn't optimize you wouldn't spend a lot of time optimizing your help center. But since chat has questions about how your product functions, it's actually a great place to to optimize and it's usually not optimized. So you'll see here Bernard will notice that we have a subdomain and um you know my recommendation for optimizing your help center it would be three things. The first thing is just never use subdomains because they tend to not perform as well as subdirectories. I don't know why but that's just how Google built their algorithm. Uh so number one move this to a subdirectory. Number two is typically internal links are not optimized. So you'll see here there's only five internal links and um we know that more internal links are better. And then the third is to I would say optimize your title tags like you'll see here using your employee directory help center but it doesn't say payroll management software. So you can do a lot more to optimize this title tag. And the fourth thing is that a key decision with your help center is whether or not to ask allow anyone to ask any question and have it be publicly indexed or not. So, if we go back to this chart here, the tail is larger in chat. And I mentioned that you could use search data to inform what the tail looks like. But if you've if the tail is larger, you don't you necessarily do not have search data to inform what the tail looks like. And there's some question tracking tools based on panel data, but panel data is also a subset of people and real data. And then you're extrapolating from that. So I think that there's probably a fair amount of mystery around what the tail looks like. There's probably very little information. So how do you know what the tail looks like? And you know goes back to uh uh when uh Bernard and I first met and discussed programmatic SEO. Programmatic the premise of programmatic SEO is to find what the tail looks like for search. And if you and if other people don't know what the tail looks like and you're the only one that does, then you can just eat up the tail. This is essentially a tail for questions without a really good data set. So how do you find out what the tail looks like? One way is to allow like what are all the what are all the very specific questions. One way is to allow people to ask any question they want within your help center that has concerns about whether or not you'll have junk there. But that is a potential strategy which is to basically have an open-ended forum for your help center saying does your product do this thing? The uh the other thing that you can do is to look at Reddit threads. So, Reddit threads are people literally asking questions, and these can help inform the uh specific follow-up questions that people might have and what the tale looks like.
Ethan, got a got a quick question for you. So, I mean, this idea of adding a lot of context around your brand is a very fascinating concept, right? I think I've heard people who you know have written say product use case pages for their B2B software and they mention that their product is a good solution for X or Y and it's a very unique you know outcome. Let's say, you know, you do HR software and then you just happen to throw manufacturers into a, you know, answer or question that you put on some page. What is to pre prevent someone from just spinning up a bunch of misinformation about how a brand works, right? and imagining, you know, there's this like modernday uh like keyword stuffing, if you will, where I could say, you know, like product X is good for, you know, industry Y and then I just repeat that over and over and over again on like a specific page, like does that work? Should that work? And what are your thoughts around that?
Yes. So, let's say I wanted to spread information like Clear Scope is an email marketing tool and uh get and it costs $5,000, so you shouldn't use it. So, how would I do that? I would do it [clears throat] exactly what you just said and I don't think that there's a way to prevent that. Um, let me let me show you this one quick example. So, how do you spread misin how does misinformation get spread and how do you cause hallucinations? Essentially, hallucination optimization actually a brand, you know, a derivative of online disreputation management. So, um, this is an example of an article that we did with Axios where we wrote this, uh, AI content study and Axios wrote about it. And [clears throat] so, here's our study and we we it's four of us and none of us attended Oxford, unfortunately. Uh, but, you know, we went to pretty good schools, but we didn't go to Oxford. So then Perplexity reposted an AI summary saying Oxford researchers just confirmed what we already knew. And um and the reason why it did that is because Perplexity summarized our article and as part of the RAG response of the summary, they looked at this model collapse paper which I did not write nor was it affiliated with our study and these people went to Oxford and um and then the summary said that we were Oxford researchers and then that posted okay that post oh and then uh Marcos on our team said hey perplexity That's not actually what happened. It was us. And then Perplexity said, "Oh, thanks so much. Attribution updated, but not updated." And then that got reposted across LinkedIn with Oxford Researchers. And then it went viral on inst uh on yeah on Instagram, Oxford Researchers. Okay. So why this matters is this is a real example of misinformation going absolutely viral. In fact, it the the misinformation version of our post did better than the posts that mentioned us as the as the author. So, why does that all matter? I do think that there's no protection at all to do um to do uh online disreputation management uh optimization. Like I could say that clear if I get Clear Scope is a email marketing tool that costs $5,000 like I could do that. There's no protection.
Yeah. Yeah. That's that's what I feared. And from a lot of research looking underneath how different algorithms specifically Gemini and GPT perform research, what I've been finding is that I think GPT tends to value official sources. This would be then looking at help centers, FAQs, product pages from the website itself. Whereas it looks like Gemini likes to value broader, you know, more generic resources and they deviate away from looking at official sources probably to prevent, you know, both the massaging of sentiment for a specific brand whether that's a a positive outcome or a negative outcome. What are your thoughts about this reputation and you know owned assets and how are you supposed to then position you know which pages like make sense to really optimize for right you've mentioned help centers you've mentioned blog posts and you've mentioned uh you know product like use case pages do each of them have different weights and values or are you finding that a specific page type matters more.
I think the question is which appears more frequently [clears throat] and I think any of them can appear frequently like it's just well part of it which appears more frequently which I think goes back to volume. So the question that you're asking is are people asking more about integrations, features, use cases and and we don't actually know but I think just from you know working in search what I do see a lot is use cases and templates number one and then I see more specific things like integrations maybe third. So that that's what I see the most.
Got it. All right. Going back to some questions that have popped up. This one's interesting. So when you say normal SEO is still applicable to AEO right a lot of that's the keyword research the building of the you know content pages the optimizing of that content but how do you think about core web vitals or like site speed and technical SEO does that matter as much
no nor has it for a long time and [clears throat] um this is actually an interesting uh that's my short answer now a a question that you didn't ask that I that will also an uh ask or answer is um [clears throat] much of SEO best practices are things that are true and not impactful. So SEO like much of SEO is stuff that's a waste of time. Things that are true and a waste of time. Core web vitals is a perfect example. So core web vitals I've never seen it drive impact within within the last 15 years. Uh however, if you ask people, they'll say it's like I'll have job I'll have interviews and people like we optimized their core web vitals and we saw traffic go up by 20%. And there's no way that that's true. And what's h what happens is there's bad either bad analysis or just misinformation that gets repeated and that's happening in answer optimization at an even faster rate. So I'll I'll frequent I'll briefly discuss core web vitals. So core web vitals comes from when a page was when you saw a page in 2000, it might take 2 minutes to load and that would be bad and you would want a faster page and it's a lot better to have a page that loads in 2 seconds than 2 minutes. So it did matter then there are not pages that take 2 minutes to load. And so if your page takes 2 seconds versus 1.8 seconds, there's no difference. So it is a signal. It's true. It is a signal. If your page loads in 2 minutes, that would be bad. But nobody's page does. So in effect, core web vitals is a thing that is true but zero impact. The same I am not seeing any any uh any impact in LMS. I haven't rigorously tested slow pages versus fast pages and LMS but I doubt it. Um, but that's what I see. So then there's technical. So technical SEO again [clears throat] in 2005 technical SEO basically meant fixing Google's shortcomings or exploiting them. Uh so now what that means is fixing chatbt's uh uh shortcomings or exploiting them. So what what uh what shortcomings might chatt have? The page can't be read. So I have this article with text and images and CSS. The bot can't read that. It's a $500 billion company with the smartest engineers in the world, but somehow they haven't figured out how to par parse HTML and CSS and words and images. So, I'm going to transform this article into this other markdown structured data weird thing so that that bot can support that even though there's no evidence that they look at that or you know any any evidence at all. But I'm just going to sort of make up a new protocol because because uh chat GBD hasn't figured out how to read text and images. So I I I see I mean it's true that you know the LLM uses markdown but you can make an article and transform into markdown like it's not that hard. So, you know, there are cases perhaps where uh LMS can't read JavaScript properly. If you did a really terrible job of implementing your JavaScript with yes, that that could be, but I think for all intents and purpose, there's these things that pop up about, you know, the LLM can't understand your page. I haven't seen any of those to be true. And the uh misinformation of it is true is uh, you know, I hear this all the time. Just the magnitude of the misinformation and the acceleration is is pretty substantial, which is uh interesting to observe. So, it's kind of like the misinformation of SEO that took, you know, over 10 years collapsed into one year.
All right. Along that same line, Alex asks similar stuff. Should we be focusing on HTML structure, schema markup? What is your perspective on the new LLM.txt file concept? Does this fall under misinformation or is this actual like best practices in your opinion?
Yes. So, two answers. The answer one is that John Mueller g see if I can pull it up explicitly said uh yeah as far as I know I'm I don't know anyone who uses this. So this is an interesting example where someone said this might be a good idea and it is a good it's not a bad idea like send data in a structure to make a algorithm understand like that's a good idea. It's just that it doesn't nobody uses it. And so it's this idea that then has gotten repeated so many times that if you hear the thing like, "Hey, Bernard, have you heard about this? Have you heard about Bernard.txt?" And like, you know, I've heard 10 people talk about Bernard.txt, that must be a thing. And so that that's what happened that that's what happened here, which is that an idea got repeated so many times that it became a fact. So there's no evidence that this uh has any effect at all. Schema for sure has an effect, and I can show you that. So if we do what is the best TV for my apartment apartment uh you will see schema and schema I I see that often especially with commerce commerce and ratings and aggregate reviews that for sure I'm seeing up here and when we do this we'll probably see a grid at the bottom top eight and let's see if Yep. So you'll see this grid and then when I click on this you'll see multiple places to buy it and the places to buy it are appearing because they have schema. So here this page has schema and therefore this this TV can be uh dduped and we know that there's multiple sellers for this one TV with this one model and that is why it's because of schema. The second thing that we're seeing around technical that I do think will matter and is new is the uh MCP stuff that OpenAI is seeing with Shopify and Stripe and all this. And so uh and there's also Chat GBT Alexa. So chat or sorry Atlas. So Chat GBT Atlas is where you say uh plan my vacation. So, plan my vacation and you're going to have a hotel booked, a restaurant booked, flight booked, and you may not have even needed to go to those websites. Like, you got a flight on United and you, you know, uh, reserve something through Open Table and Marriott, you you booked your hotel and you don't need to go to those sites. So, what we just saw was uh chat GBT giving guidance about give information about your buttons and your forms which I think probably I haven't tested but I think probably that's because they want to be able to execute conversions within your website. So I think that what we will see is a new field of technical stuff that does matter which is you want the agents to be able to perform actions on your website. That is new that is probably kind of messy and probably will need to be fixed. So that area of technical AEO uh I I do think that will will come. We're not seeing people do that yet, but that's what I think is coming.
Yeah, I think that that makes a ton of sense. I cannot pronounce this name. I think it's some sort of a Greek name, but how do you think about new ways of forming keyword clusters? Right. keyword clustering, obviously the concept of wrapping a bunch of identical, if not similar sounding keywords into one particular bucket. And how does that apply to, you know, this taxonomy that, you know, this language that we've found ourselves in around the fan out of queries that AI is performing for the end user, right? If I ask what are the best TV apartments and GBT decides to perform a web search, right? What is your thoughts around building clusters to target that particular prompt that the user is going after?
Yep. Let me see if I can recreate this. So, let's see. Okay. So, if when I click this, I think it'll give you some fan outs. What are the best TV or sorry, what are the best TVs for apartments? So, you'll see here I think it'll do a fan out. Yep. Searching here top. Yeah, like top 43 in TVs. So, that's clearly a fan out from just TVs and oh yeah, assistant steps. So, you'll see the fan out right here. So, how I would think about this would be two steps. Step one is just my keywords that I know I want. And let's do this here, robust TVs. So, how I would do that is I would take Whoops. TVs for apartments. So, I would do your your typical SEO clustering best TVs for apartments here where you'll see a bunch of keywords in this particular URL, 4,000 different keywords, and you'll see some some themes like the year and best and for a particular use case. So, that you can start with your search cluster there. And then you could take your fan out. And how you could fan out would be to put them in perplex probably put them in perplexity because they're readily available here. And you'll see some interesting things like the 43in example or compact. So I would basically start with your searcher cluster and then fan out uh using a perplexity to say tell me what the fan out is. And so would you in your opinion create separate pages to target best apartments or best TVs for apartments, compact smart TVs and then you know a separate one for top 43in TVs or would you try to bundle them all together on one specific page?
Yeah, for me there you know you could debate about the definition of a topic but for me the definition of a topic is there's one page that should target this cluster of things questions or uh keywords.
[clears throat]
So for me you have one page and you know why would be this one page ranks for these 2500. So given that it's ranking you probably don't need to separate it out. So you start with these and then the exact same thing should be true. Ideally you have a single page that is fulfilling the intent of these. If it's impossible, perhaps you have a like perhaps you do a 43 inch, but presumably this question, you know, uh the URLs that are appearing for this question are fulfilling the intent of 43 and of compact. So my general suggestion is uh to to have a single page or topic that is targeting all of the search keywords and all of the fan out keywords or questions.
Got it. All right. Ulie asks given that there's so much what is informational content on you know the what is concepts how would you spend your resources and budget in percentage for each of the top ofunnel middle ofunnel and bottom of funnel content when thinking about that as you know part of your content strategy.
Yeah, I would, this is a little bit different. I would spend no time at all really on glossery definitions and low intense stuff. I would focus on the end of the journey and probably the middle of the journey, but the very beginning of the journey and glossery definitions, there's no clickability within the answers. And the intent is low. Even if you asked me 5 years ago, should you do glossery terms, I would still probably say no because there's no intent to know the definition of payroll management software. The intent is buying it or researching options or understanding use cases but not the definition of the word. So the same answer that I would give you 5 years ago is the answer I would give you today which is don't spend time on on definitional uh terms like that but do spend time on the end of the journey. The non-obvious thing is also spend time on the middle of the journey. So a mistake that I would say is ranking for payroll management software but ignoring the use cases and the integrations cuz that is your persona and your buyer. they're going to be ready to buy eventually. Even if they're not ready to buy today, they will eventually be ready to buy. So, I would still focus on the on the middle and I don't think that that'll get uh uh you know owned by AI and not be impactful.
All right. I personally have a a follow-up question to this. How important do you think topical authority is in relation to large language model or AEO? And does it make sense to create top offunnel content targeting the what is you know why X matters type stuff simply to win an AI citation to then further enhance your brand's ability to perform across the middle of funnel and bottom of funnel prompts.
Yes, the answer is yes. Okay, so let me break it down into a couple different things. So the first thing is how does the how does topical authority work. So how topical authority works is I think it's a few different uh features. So how authority work was just back links and anchor text. So if I link to clear scope and I say SEO tool that means you're an SEO tool. The more the better. Then the authority algorithm diversified and it included things like branded search SEO tool clear scope or and then it also expanded into if I search for best SEO tool and I go to clear scope and I stay then the engagement for the search was good which is what you're actually asking about. Or if you had chicken recipes and there were great chicken recipes and then everyone was searching for that and claimed came to clear scope, you will now get a topical authority for chicken recipes even without backlink text. So searching for something and clicking and getting good engagement means topical authority for that and you will rank better for that. So if [clears throat] you have content if you have more and more content about chicken recipes you will get do do better and better and better and so uh yes to your to answer your question uh if you create content in a particular topical area you want to create for you want to rank for even if it's not valuable that like let's say that this is back to the definition what is the definition of payroll zero conversion rate ever but if you get good clicks you're buy payroll software payroll management software you will rank better for that so that works for Google. I think that that also works for LMS because LMS are using search engines and search engines are an evolved technology. There's not, you know, they're all kind of the same thing based on similar u similar uh infrastructure. So I think probably that the exact same thing would be true for all the elements.
Got it. Yeah, I think that that makes a lot of sense. I know we have some other onpage stuff before we talk about off- page things like Reddit, so I figured I would, you know, link those together. Julia has an interesting question. Winter has a dedicated page to increase their chances of getting cited by LLMs. If you look at the chat, there's a specific AI info page that they've created to communicate to large language models what they are, you know, who started it, what it's ideal for. Is that something that you would recommend or is that just part of the noise that you know we've been seeing happen as discussions within AEO continue to happen?
Let's see. So how I can answer that is ask the chat tell me what you know about this company. So this company winter is trying to tell the bot trying to tell the AI this is uh this is this is information about my company like where we're from and where we headquartered. So let's let's take a look at this. So funding how it works speed and agility. So let's do something pricing 10,000 and let's say where was it founded? Where was this company founded? founded and uh and then who is the founder and uh what what is the uh what is the category and what the category so you could basically answer this question by just asking the bot tell me about this company and let's see what it comes back with I do think that you know back to the uh example that I gave with rippling and with gusto where it says we have these features we have these HR features. It's clear that those things are pulled. So, this does seem like the sort of thing that would get pulled. So, let's see here. Winter and it's pulling from the about page and it is not pulling from the AI. So, the AI page/ AI info page. So, it's pulling from there and it's pulling from the about page. And once again, and at no point is it pulling from the AI page. So, I mean, the interesting thing here is it's just pulling from the about page because you already have the information here. And so you didn't need to you didn't need to give it a name like hey a llm just so you know you're you're the the door for you is over here. You had the information here and you didn't need more information. That's what I'm seeing here is uh it needs to exist somewhere but it doesn't need to specifically be called AI page.
Yeah, that makes a ton of sense. So I could see to summarize I could see a world where if you're a brand new company or product or service having this information whether it be in an AI specific page like what winter has done or just an about page or your product use case or service pages would be the same outcome as what we're seeing happen here with winter.
Yes. And again, you know, this highlights having the information exist is what you need. It's not calling it something. It's not structuring it differently. These are just using search engines and search engines can find pages without being called search engine page. Just list the features that you have or list the company information uh that you have and the bot will find it on its own. You don't need to do anything special. All you need to know is what questions people have and to answer them.
Totally. All right, one more question. Daniel asks, "Have you seen a difference in the difficulty to getting visibility and citations based on traditional SEO metrics like keyword search volume, KD, keyword difficulty, or cost per click? Or does the previous keyword analysis not relate to LLM
visib LLM visibility at all? And does it take the same time in SEO, right, for results to start showing, or have you seen it happen faster or slower?
>> Anecdotally, I've seen the same correlation with difficulty and and authority and all that. I have not rigorously tested that, but anecdotally, I have. We flow is a good example. I spent a bunch of time looking at their stuff, and I'm I I typically see high authority sites. I typically don't see low authority sites. And then um I will say that the speed is different. So the speed specifically for new companies is different because for a brand new company if I'm if I'm competing with uh with a a web flow and I'm a new noode tool I could win by winning the citations. So what is the best website builder? So back to this example.
>> You're not very yet.
>> Yep. It would take me a really long time. Like let's do best website builder. These are going to be crowded. Lovable Web Flow, Reddit, Web Flow. Yeah, Canva. Like this is a really hard page to win. If I launched my site tomorrow, this is this is not going to happen. Essentially, I wouldn't even I wouldn't even try. And then when you're here, then you can win quickly if you paying the Tom's guide and say, "Hey, would you please mention me for we may earn possibly earn an affiliate commission?" Like, if I give you $500, will I become the best website builder in the probably for the right price? And that could happen tomorrow. Like, as soon as Tom's guidance says you're the best, you become the best. So, if you go through here, you get mentioned on Reddit and Techraar and Tom's Guide and Yep. Tom's Tech Radar, like we might earn affiliate commission. I'm not really sure. So, that [clears throat] can happen very quickly, way faster than my brand new website, you know, slogging through trying to win a position in in Google search.
>> Makes a lot of sense. Let's let's talk about off- page because you just basically mentioned how getting brand mentions within top performing content for the competitive keywords is important. So talking about Reddit on Reddit should we pay attention to should we pay attention to old threads that get a lot of traction and rank in the SER and how old is too old for targeting those? asks Fernando.
>> I think that they shut them down. Uh I believe that they closed commenting for for old popular ones, which I think is the question. I I I anecdotally I think that you need to look at fresh ones because it's hard to go back from a thread from a year ago and then suddenly have an answer there. So I would focus on new threads. That's anecdotal. That's not rigorous analysis, but um anecdotally that's what I'm seeing.
>> Got it. And then back to one of your original early conversations about Reddit with the commenting of, you know, hey, my name's Ethan, I'm at Graphite, so on and so forth. Should the account itself look generic or company branded? Does that matter?
>> It does seem to matter. And I I I Reddit has also given me specific suggest like the company Reddit's given me specific suggestions. So part of their suggestions are to do exactly what I showed you, which is I I'll summarize some of their suggestions, but it's to say who you are. It is to what web builders I'll just I'll just literally show you my account. So say who you are. Don't use a picture [clears throat] of yourself. Do a little icon. So this is me, Ethan Graphite, little friendly avatar. And um I have my bio, so everything's transparent. And then uh this is generally what does well better than company accounts.
>> Yeah.
>> And that is that is Ethan describing what Reddit said to Ethan versus Ethan is a master but but Reddit is a master at Reddit marketing. So that's what they suggest to me.
>> All right, I'm going to ask a follow-up question to that. when you are commenting say on Reddit or if you're trying to win branded mentions on these top performing sites for the topical areas that you care about, how important is the A mention of your brand name like say graphite, b a link specifically to graphite and c the lexicon surrounding the discussion of where that graphite brand mention or link exists.
>> I will show you. I I think I know some and I think others. So, how I would answer that would be if we go here and we find Let me see if I can find this. What I want to do is I want to find Reddit. Yeah, ideally I ideally we would find a specific mention. So what here's what I think. I think that it's just getting mentioned at all. I don't have data to suggest that a link matters and I [clears throat] don't have data to suggest uh yeah sorry here's what I think. Mention number one and semantic relatedness number two. So semantic relatedness like the the lexical stuff that you mentioned will increase the probability will increase the relevance of the rea thread which will then increase the likelihood of that red thread appearing for a particular question. I think that that's not controversial. We know that. So if this best website builder that has a bunch of content about you know the examples and benefits and all that sort of stuff it's more likely to show up. Uh it'll have a high relevance score more likely to show up. Number one. Number two is a mention. And it's clear that mentions matter. It's not clear that links matter. So I don't know about links, but the first two I do know the answer to, which is a mention is good and that the uh semantic uh relevance will increase the probability of the URL appearing in the citations.
>> All right, one more follow-up question to that. In SEO, there was or is a theory on the above the fold scroll depth of content mattering. Do you think that the scroll depth or mention placement higher or lower has an impact on AEO?
>> I don't know. But I will tell you what I do know. What I do know is if we do something like uh like this, what I've seen in Google is a discrete b or a binary thing meaning above fold matters. Above fold is bucket one with greater weight. All below fold is bucket two. Whatever the position in bucket two doesn't matter. So this is equal to Whoops. That's equal to the thing all the way down here. So I played around with moving stuff around and the it's just two binary buckets below fold and above fold. The second test that I've done is you'll see here let's say that this was all you saw above fold. This is not great. there's not there's not a lot of relevant text and so you're you're putting you're you're leaving money on the table. If instead you have some of this appearing above fold, this is probably what you would see above fold. This is good. If this didn't show up above fold, this page is not optimized. But whatever the ordering is below here doesn't matter. So that's how Google works. And I've tested that a fair amount. Now do I think that
>> Yeah. Uh do I think that chat GBT is built the same way? I think my general thinking on chatbt is it is a new search engine and they haven't had 20 years to build their search engine and I think that there's a bunch of that like the core search engine probably function similar to Google. I know that there's all these like tons and tons of things that Google has added on to to like refine stuff that I don't think that they've had time to build. So it wouldn't surprise me if they didn't do above fold. Now, you know, one other theme I've generally seen is that there I think that their authority algorithm is not as evolved. An example is I see Techrider, which nothing against Techrider, but that's like showing up all over the place and you don't see that in Google. So, you're seeing these strange themes where the relevance is generally very good, but the authority is seems off. Um, so I think that some of these more nuanced refinements that Google has built have not been built by chat GPT like the thing that you mentioned around scroll depth.
>> Totally. All right, we're coming up on time. I'm going to ask one more question because I think this one's very interesting and comes from Sai. They ask, does having personal experiences or expert tips help AI recognize the content and display it in general? Um, you know I'll tack on the concept of you know information gain helpful content EAT, you know what is your opinion of the role that personal experience plays within getting your content having your content influence or be part of the answer.
>> I think that information gain matters and I think that personal experience matters. I think uh well sorry I know information matters information gain I have decent data to suggest that I also have decent data suggest personal experience matters how I think what the what how I think the algorithm is implemented is I think that the algorithm is looking for two things the first thing is the uh uh comprehensiveness so do you have gaps are you missing something that the other 20 results have second piece is are you saying something that they are not saying that is still relevant And so information gain would be you you gave an example that many people have not mentioned. You took a contrarian opinion. You've done stuff around this contrarian opinions like the downsides of this and that'll get boosted because it's the downsides. I think that they also are looking at personal experience. I don't wouldn't necessarily say it helps them understand something. Rather I think that the algorithm is specifically looking for uh I think uh HSM Rush did a correlation around like uh personal pronouns or something like that. mention of that information gain and um so that's what I think that they're looking at.
>> Awesome. Well, we're right up on the hour and this has been a fantastic session. I usually like to end by asking you Ethan, is there something that you would like to share with our audience especially as we head into 2026 and you know the world of search continues to evolve. Any last notes, comments for our wonderful audience that has tuned in?
>> I would say two things. Thing one is that the majority of the information that is uh people share about this category is not true. Uh and the magnitude is high. So I would suggest to test things and uh set up experiments and validate whether or not these things are true and not take at face value what you've heard. Uh and then the second so set up tests and do experiments uh and please share your experiments uh if you have findings please share them because I don't have all the answers and then the second thing is I think that a very interesting category is going to be this autonomous agent plan my vacation and make decisions without my involvement. That's very interesting and I think that that'll be a big change that we don't have the answers for yet.
Yeah, that one that one's pretty pretty crazy when agents have your your credit card information and purchasing power and can simply just book things and operate autonomously on your behalf. At that point, you know, AEO wouldn't even necessarily be the name of what we're implementing here, but just a very fascinating world that we find ourselves in.
All right, >> well, that's a wrap. Thank you all for tuning in. We will upload this recording to YouTube and we hope to catch you on another webinar. Thanks so much Ethan. Catch you.
>> Thanks. Cheers.