AI Search Optimization (GEO, AEO): What is it, Common Errors & How to Optimize AI for Success

The AI Search Roadmap Series. Welcome to the first episode in which we are going to go through what is AI search optimization, what are the top mistakes, what are the most common obstacles, and how to establish a winning process despite all of the uncertainties that we are going through. For this first conversation of many to come, to introduce the topic, I have invited some of the leading voices sharing insights about reasonable ways, cost-effective ways, SEO-friendly ways to tackle AI search optimization. And with me, I have Lily Rae, Chris Green, and Roxana Estingu. Welcome, welcome everybody. Thank you very much for joining me today.

Hello.

And before starting our conversation, I want to thank Similarweb, who's our today's episode sponsor and the whole series sponsor. Thank you very much for your sponsorship. It wouldn't be possible without you. And if you haven't taken a look at all of the features that they have for AI search optimization, go and take a look. You will be surprised. And to start already the conversation, my first question, what is it and what is not? And it's still conversation about, oh, how it should be called, who should own it. I actually asked a couple of days ago in LinkedIn about who already owns or leads or is in charge of AI search optimization across teams. And in 95% of answers, people were saying it is whether the SEO team or dedicated AI search optimization team with SEOs, because it's, yeah, the reasonable thing to do. But what are the most common mistakes that you still see today that you would like to clarify? And in general, how you communicate what is AI search optimization with your clients and decision makers?

A lot of my work, I sort of, as a senior consultant, I kind of help manage like global programming. quite a few businesses. And I've actually seen different things where other teams taking a hold of this. So dedicated, like geo AEO type teams, names vary, but also plays from ad tech and paid media teams also into this. That's more in the e-com space. But I, yeah, a lot of the last year has been spent kind of saying, look, we've got about 15 or so years experience with this kind of shenanigans. Don't overlook the SEOs in the room and In most instances, they're coming around to that way of thinking. So I think there's the degree of, right, okay, the listicle bubble is popping or the like scaled content bubble is popping or have you hopefully. And it's, yeah, we're kind of coming back. I mean, it's like, let's not make the same mistakes we made earlier in our careers and just spending a lot more funds. So I think for me, it's just the big tension I've seen in one of the biggest areas is trying to split that resource and assuming that you can act, you can have a geo or an AO team act differently. Because you can't. And then the scales fall from people's eyes when you say, look, when ChatGPT runs a search query, it's querying a search database that, that, and I've run some correlations and some scores on my bot data and other tools. And quelle surprise, like the most visible URLs in AI chatbots are the ones that are indexable. And it's kind of like, if you're not indexable, the chances of you appearing in a chatbot result are virtually zero. It's like, okay, there's something in this, you know, like our 15 years worth of experience might be worth listening to after all. So. Seems fundamental, seems like absolute, like zero point for everyone to go into, but it's not. You cannot take that for granted that Google being able to find and index your page is actually a prerequisite for AI search, which it obviously is.

AI search is still search. So SEOs are in the best place to understand that, right? At Alamy, I don't just do SEO, I actually oversee our internal search engine based on the same principle. That search is search. Once you understand the search engine, you kind of understand them all. And of course, there will be differences and complexities, and I'm never going to compare our search engine with Google. I wish, but not even close. But once you understand the information retrieval in general, you can see it across platforms, even if you do it by keyword or you start doing it by prompt, even if you get a list of results or you start getting like a the synthesized response back, it doesn't matter. You look behind it and you still need to have your information somewhere, so you still need to be discoverable and indexable and crawlable to show up in the first place. And SEOs are the best place to understand that. We already have the tech, we have the tools, we have the knowledge, we have everything to be able to analyze it, look at it, and come back to the business and say, look, This is what's going on. This is what we need to do. But I think that's the key point. You go back to the business and you can't do it alone as an SEO. It is just, you know, it's only so far you can take it without a marketing team, a digital PR team, um, people to write some content around the things you want to show up for and so on.

Because of that, also easier to understand by companies that were used, are used to have teams to optimize for other platforms, SEO for YouTube, SEO for TikTok as well, and didn't necessarily correlate search to Google SERPs and in that particular format, but they were more open-minded about, yes, we're search optimizers. We can optimize your discoverability independently of the platform, right? That is what we are here for indeed.

Yeah, I totally agree with everything everyone has said so far. And, uh, I'm a solo consultant. I have a new consultancy called Algorithmic, and I've also worked At an agency called Amson for over 8 years now. So I kind of see both sides. With my clients, basically the way we're thinking about it is AI search is— SEO is part of it, right? It's kind of like an evolved version of SEO, like Chris mentioned. Obviously the language models are using search retrieval to get a lot of information. Um, but beyond that, you know, we have to think about what information is informing the training data as well, right? So I think that's where branding comes in. That's where digital PR comes in. That's where your social presence comes in. And like you said, Aleyda, The language models are not just looking at one thing, they're looking across the whole ecosystem. So I think, you know, if we were more myopic before, kind of only focusing on maybe Google or Bing, we should have already been focusing on different search engines, right? I think YouTube and Amazon and everything have already been very big players in the space for a long time now. But I think now we have, we have to absolutely be focusing across the entire ecosystem of how people search. And I think what language models and AI search is trying to do is trying to get an understanding of brands and facts and information across that broader ecosystem. So I think SEO is a core part of it. We have to make sure everything's optimized, retrievable, easy for agents and language models to understand. But we also need to make sure that we're doing the right things so that our brands and our clients are well represented and that the information is accurate and consistent across the ecosystem.

And should be a win-win in the first place also with our more traditional SEO type of strategy because I have seen this argument around, it's like, oh, but now for AI search, they need to align with digital PR and community management because a lot of the citations are coming from Reddit, YouTube, and all of this major publication depending on the vertical, right? But I was like, we should have been already doing this for SEO in the first place, right? If you were running an actually sophisticated, well-funded SEO process in any case. The problem is that a lot of people were not, unfortunately, because one reason or another, right? Because sometimes because of the resources availability, other times because of big organizations silos and the lack of capacity to align. And now it's more evident that this needs to happen. Then another thing that it reminds me a lot, right, is unfortunately how in SEO we, we have the tendency to be very Tactical driven, tactical focus. So when something moves, all of a sudden it's like, oh, I cannot do this anymore. But it's not like that. I would love to hear regarding how you have moved forward effectively and successfully so far. What are the best ways that you have identified, the steps that you have taken, the criteria that you have set in the organizations that you work with?

First of all, technical SEO still matters. It will always matter. Don't ignore it. Um, the same old things you were doing before that were working for Google and Bing and whoever else will work for these LLM-based search engines. Still make sure that your content can be discovered, as it can be crawled and everything. And do pay attention that most LLMs won't render JavaScript in the way Google does, in the way Bing does. So you want to make sure that your server-side response, your raw code, does have all the information you need in it. Because if you rely on JavaScript to render a client-side You might find out that LLMs can't actually read that. Having said that, some of them, when they go to check your website out, so for instance, when you give it a website URL and you ask something, they will go and generate a screenshot. Again, that screenshot, that's not rendering JavaScript. So if your content is hidden in dropdowns and things that you need to click on to open it, in that screenshot, they will not be able to see it and read it. And it's absolutely ironic because Google always said, if it's important content, don't hide it in accordions and dropdowns and things, just make it visible on the page. And now with LLMs, the same thing applies. So your technical SEO, the same things you were doing for Google now matter for AI search as well. Sometimes for different reasons, but it still matters. So do not say, you know, I'm just going to focus on citations and grab those up and then forget about your technical SEO. I'd say that that's kind of the most important for me, at least.

It's the base indeed. Then you build from that 100%.

Add on to that one. There's a couple of technical things that I've seen that are really super interesting. I think firstly, just because you can see like an AI chatbot can do a functional call, a tool is not assuming that that is going to be part of the pipeline that retrieves and surfaces your data. One of the more recent ones I had to wade into was around images. It's like, oh, they can read images so we can put valuable text in images. It's like, If you upload an image, it can read it, but beyond that, it's kind of missing. So again, it's that we know fundamentally just all of these pipelines are gonna fetch the HTML at some point. And it's kind of, that's the lowest bar. It's like we want to guarantee we're in it. Like that's, that's the kind of the critical piece. And I did some more research with agents and I was trying to trick agents to interact with elements of the page differently to a human to see could we create honeypots for them? And the short answer is they're super lazy by design. And just because they can see something's available that they could interact with or a link that's hidden that's referenced somewhere else, they don't even, unless you explicitly say, go search for the hidden stuff. Most of the time they're just gonna be looking at what's like indicative of a user's experience on that page with the least effort possible. And then there is like a, there's some kind of time budget attached to that as well. So when we move beyond just AI search and we go into like agentic and more letting Like agents do the whole process for us. Again, similar principles, like everything baseline in the raw code, make life easy for yourself. And all of these hacky workarounds, if it feels really smart and it feels like you're going to all of these lengths to make it better for a bot, you're probably introducing like fallibility and weakness in this kind of brittle layer. So if it's in the HTML, it's good. And the final technical thing, which is also really interesting, doing some correlations with like the chatbot. access logs. It correlates super well with PageRank in terms of like the pages that were hit most for chatbots had a like a 0.7 or 0.8 coefficient on the sites I've tested. Obviously, that's isolated. It's not like broad research, it's on a couple of sites I work on. But that just felt really interesting in terms of how these bots visited, what was getting queued up or not. I mean, it's logical because PageRank broadly reflects the importance of a site. So it might be based on the same ones. But Again, kind of drawing these parallels between just technical best practices from an SEO point of view is linking to, well, they still seem to be very much what they are.

Thank you very much for bringing that up because I see a lot of buzz regarding markdowns, for example, and I am like, why would you prioritize or invest resources on creating a replica of your information to make it easier to reach rather than investing on your own content, actual content of your own pages that the users, the actual humans need to see anyway, and try to invest on that. There might be some very specific edge cases in which this is non-doable at all, in which maybe that is an option, but by all means, this is the way to go, right? And, and there are more and more tools. In fact, Lily, you actually have a tool. to validate the dependency on JavaScript, indeed, on the main contents and elements, right? There are quite a few nice tools and resources in order to validate this.

I'm really glad that Roxana and Craig both touched on the technical elements because I think a lot of people are talking about AI search without doing these kind of technical diagnoses to see whether or not the crawlers can actually even see the content. And that, that's where technical SEO really comes into play. So it's important that you're doing that, but there's There's a lot more that goes into it as well. So I think what we're trying to do, like on my team and, you know, with our clients, they need to figure out what— first of all, AI search is becoming highly personalized. So we need to think about the different personas that might be searching for these different things, which platforms they're using, right? There's different tools that we can use to understand which language models our audience might be using the most. For example, SparkToro has a great tool you can use for that. And then really understanding Because the way AI search works, the way I think about it is like there's infinite possibilities for how somebody might say something or type something or prompt something. So it's going to be impossible to track all of them. But we do want to get kind of like a proxy for what are the most important topics for our business, what are the most important keywords or aspects of our business or things that different personas and different audiences might be searching for that we want to show up for, right? And there's all kinds of new tools that allow you to kind of get a directional understanding of how visible your brand is compared to your competitors' brands, how much you're being cited, which are the companies, which are the websites that are being most frequently cited for different topics and different categories that you care about. So I think it's kind of like traditional SEO keyword research in the sense that you're getting that data of like, what are people really searching for? Although with, of course, with AI search, it's much more difficult because we don't have that really like prompt-level data. So there's all kinds of new ways that you can kind of like estimate where, how are people searching for us? What's valuable to the business? You know, you can cross-reference that with paid search data, for example, to get like conversion data and what's actually driving revenue for the business. And then create kind of like an understanding and use the tracking tool of your choice to see how visible you are. Obviously tie that back into your analytics as much as you can to understand how much people are really coming to your site from AI search platforms and everything as well. Um, but it's kind of setting up that whole process of like, what are the prompts we want to track? What are the keywords that are revenue driving and important to the business? Who are our competitors? Which pages are, are adding, you know, influencing the answers and where we need to be mentioned the most? Um, so we're kind of building out that process. That's what we call AEO on our team.

I would like to highlight a few things, and it's true that we are going to have like different episodes for each one of these topics as well later, but some of the most common mistakes and misconceptions and worthy to highlight is like the importance of setting a good prompt library that are actually representative through the whole customer journey for real user scenarios rather than made up, completely made up. And I understand that one of the issues that I see people seeing is because of they just preload whatever the AI tracker they use provide. And no, you should look into your real user search data, the conversational, for example, keywords are displaying AI overviews right now in Google, the ones provided from Bing Webmaster Tools already, but rather than copy-pasting them is to identify the patterns, the characteristics, to run understanding also asking internally the patterns across those behaviors and match those with your different product lines, service lines, different stages of the customer journeys, different personas, and then rather than trying to mix them all together, it's like isolate. And you can have 35, 40 prompts when they are actually meaningful, a good blend of non-branded and branded through the different customer journeys for your personas, for each like product line, service line. If these are actually well thought, well defined, and representative of what your unique selling propositions are about, how you want to come across and I believe that the understanding of your audience is particularly important here. And then secondly, I love how you mentioned regarding, okay, we start tracking and identifying where we already getting surfaced consistently, right? Rather than obsessing about individual prompts, rankings, and such like for the different topics and organizing the prompts per topics, per areas to have like a consistent basis. Are we consistently being shown or not shown here rather than obsessing like, One time we are shown, another time we are not. It's like, these are pretty dynamic. We cannot obsess over that. So it's consistency and patterns matter here. Then finally, I love how you mentioned developing on one hand proper actions based on the gaps and the opportunities are based on, okay, we need to create more information that can be easily citable in our pages about these and these and these topics. Or we need to get more visibility, very likely in these third parties where we see that our audiences are hanging out much more and somehow we don't have yet presence. Let's talk with the community team or digital PR team to see if they have something planned in there. Otherwise, let's align to see if they can prioritize, if it makes sense also business-wise. And if not, like, why is the reason, right? Again, this is very strategical. I see many times like, okay, we see a lot of Reddit coming up, like Let's start posting in Reddit all of a sudden, or we see that our content regarding comparisons of our characteristics versus competitors is not getting surfaced. Why is that? Or I'm going to obsess about chunking now and optimize, right? So I would love to hear your thoughts regarding all of these new tactics that we hear about with AI search optimization as well, rather than core principles and critical principles that we should follow and then aligning them in a way that actually makes sense, not only with AI search optimization, but also SEO and without destroying your SEO efforts as well.

I got one that's a bit of a pet peeve, I think, since I, um, since I started working for Alamy, because we rank for pretty much anything you search for. I kind of gave up on the idea of monitoring keywords because it's impossible. So you go in, uh, with like themes and things like that. You group them a lot to topics and things. But when I look at how people create their prompts to kind of keep an eye on their visibility in LLMs, their prompts are just an extension of keywords, which doesn't really work now anymore because if you take a step back to think, how did we even get to look at keywords in the first place? It would be that a user had a problem they were trying to solve, and then they would go to an engine and think, okay, this is my problem. What can I search for that could potentially bring answers to my problem? And that's how they came up with keywords. But now they go to ChatGPT and they don't even do any thinking behind the problem. They just type the problem itself. So if we just extend the keywords, those might be the wrong keywords in the first place. We need to think about what the problem my user is trying to solve when they come to ChatGPT, and that's what I want to show up for because I'm a solution to that. So I think one of the bad things we're doing is creating prompts that are just expanded keywords versus creating prompts from actual customer problems that could show up in conversations with LLMs.

I think the sort of slight extension of that is in terms of the content and how we put it on there. I think there's For me, there's 2 key elements that often get missed. Firstly, a lot of the areas that we might have the opportunity to be most visible, probably our websites aren't the best to cater for. So kind of these comparative level comparison searches, like best of qualitative areas, like most people don't get away with ranking top in that space, especially if it's highly commoditized. And it's because of that, people aren't monitoring that comparative level. They're not seeing who gets compared, what details are being pulled into that comparison, what details are kind of salient to that.

Yeah.

And I think it's really easy to look at your competitors and see, well, what details do they have on their pages versus what details do we have and what's available? And this leads technical support pages to be missed or press areas. These are all like really rich sources of information that LMs will use to generate these kinds of comparisons. It's kind of, it's a rich seam. It might not be marketing copy. It might not be copy you put on your main site, but it's all of that content that's used to help AI understand, well, how retrievable or eligible should you be? And then even one step beyond that is there's a lot of overemphasis maybe on the construction of language. You mentioned chunking, and I think we all got in our own heads a little bit early on, like looking to kind of feel comforted by something. But the key part in that is still explicit facts and details are far, far easier to to be cited than vagaries or hedging or bits like that. And just really concentrating on what have we actually made these facts explicit? It doesn't matter if it feels like it's almost obvious or it should go without saying, or it's implied, or we say it somewhere else. Like, I think a lot of the optimization work without even necessarily worrying about the prompts we're tracking could be catered for, well, what details are important? Like, what do people ask for when they phone or where they get in touch via live chat or the contact forms, those kinds of core pieces of information. Well, if you're surfacing those across the board, you don't necessarily always need to do like prompt-by-prompt comparisons because you're just building that kind of richness, especially whilst we're kind of ironing out the kinks of like, well, this pipeline versus that pipeline, one LLM versus another surface, like that's all different, but the facts are still the facts, right?

Yeah.

Make it easier to read and to trust. EAT, anyone?

Lily, this—

I imagine that you have many thoughts, especially with EAT that you talk about so much in the past. Indeed.

For sure. Yeah. I mean, I think one thing that's kind of interesting about AI search, like we talked about in the beginning, is the fact that the language models are all pulling from search engine databases, right? So they're pulling from primarily Google and Bing. At the end of the day, that's where they're getting their information from. And Google and Bing have really strict policies that they've built over 25+ years of existence. For how to create content and how not to create content. And what's interesting is the timing, right? Because in 2022, 2023, 2024, especially Google introduced all types of new policies that basically demote overly optimized, overly like SEO kind of spammy content, which a lot of people in the SEO industry were actually creating that type of content. And what I've seen in the past year is there's been a lot of new interest in AI search, a lot of new people building content that maybe don't have that context or that experience, right? So they're building a lot of the pages that Google literally just penalized 2 years ago, right? And then they're using AI to create a lot of them because it's easier than ever to scale those types of pages. So what happens is for 2 or 3 months, you might see a really big surge in traffic and you also might even see a surge in AI citations if that's something that you're tracking, right? But then when Google and Bing and these systems can detect what's happening, it usually takes 3 months, you might lose all of it. And what I've been presenting about over the summer at various different conferences is When you lose your organic traffic and you lose your organic visibility, your ChatGPT frankincense and the other language models follow that same pattern. So it's kind of this interesting moment where I think, once again, you should talk to your SEO team about whether building that content for ChatGPT or building that content for AI search is really a good idea, because you have to think about how search engines are going to treat that content long term.

100%. At the end of the day, you shouldn't be building, and in SEO we know about this, very well for the current weaknesses of the systems. You should not— that, that is very myopic, unfortunately. But something that is consistently reliable for growth in the long term, because indeed, unless it is your little experiment or test, your own website or real businesses and livelihoods behind of what we do. So we need to be responsible accordingly.

Indeed.

I would like to ask you regarding Setting the right and reasonable goals for AI search optimization processes, because there's something that we cannot deny, that the way that we assess across many scenarios, traditional SEO processes with AI search change, right? Not only because unfortunately we shouldn't have in many occasions unless specific circumstances, but We used to over-rely a lot on traffic, right? Like easy to understand KPI. You rank here, your rankings increase, your traffic increase, your conversions, revenue increase accordingly. So it was very easy to show the correlation of this. Not necessarily anymore. We know that the outcome of AI search visibility doesn't always translate in direct traffic from AI platforms on one hand. That is the base. We know, and actually SimilarWeb a couple of weeks ago, they released along with Rand Fishkin a really interesting study tracking the behavior of users for a week, and they identify how there was a good share of brand recalls that were generated after finding a brand in an AI answer that ended up causing a direct search for the brand, showing again the importance of aligning your AI and SEO strategies accordingly, but also how it's tricky right now to show direct value with the current measurement framework. Something that I'm recommending my clients whenever it's doable, not always unfortunately, but whenever it is, is to ask at the end of the user journey, the conversion journey, how did you find us? And to try to identify those percentages of users that say, yes, AI platforms played a major role for me to convert and see what is analytics showing instead to potentially take that percentage of attribution to AI platform and to show the value that is not shown by default. It's tricky also to set the right expectations. It requires a shift in goal setting for sure. So I would love to hear what you are doing. What are your tips and how you communicate this with your clients?

This is the tricky one. Um, because I think all of the, the activity towards optimizing in this direction has come way before we can provide a compelling picture that it's working or not. And I'm not saying that there aren't methods, there are many different ways. But the thing I found, especially like communicating this stuff upwards, is the more complex and convoluted your explanation is, the less kind of ultimately compelling it is. And the least satisfying answer for this is visibility or share of voice. Well, we can't— I'd argue you can't really do share of voice, to be fair. But visibility, but relative to competitive set, relative to the people that matter is still probably the best way. And you kind of, you have to repivot back towards the days of PR and where we used to measure like column inches or eyes in a publication or, and not because I think that's a better way of doing things. Fundamentally, I dislike that, like the offline measurement because frankly, you just, people got away with doing really bad jobs at marketing because they couldn't measure it super well. I'm not saying we go back there, but relative visibility is the best we have. How you then link that to outcomes beyond what you've said then later about just asking people is incredibly challenging. I'd say for people who haven't already, bot access log data is a really strong proxy. Like which pages are popular from AI chatbots, which aren't. And also for things like ChatGPT or OpenAI, where you get some level of granularity as to which bot is hitting, you can kind of start to understand, well, where are we kind of factoring in the process there? And you will be able to establish, well, my top performing pages, they look like this, my bottom performing pages look like that. And you can kind of give some directionality, at least towards the optimization efforts. And you can start to extrapolate things like conversion rate and then with average conversion revenue values and you can start to get there. But it's all very homebrewed at the moment, I would say from that side of things, we can do it. I'd say people that are over-relying on directly attributable revenue are going to really struggle because I promise you, you'll never justify the costs relative to the outcome because we saw about a year ago, we sort of got explosive growth as like ChatGPT adoption raised. We saw similar when like Um, other chatbots started to see growth in separate places and that kind of, the revenue didn't follow. And it's kind of like, we, we know at an intuitive level, we're being seen, studies from like when we cited from Roland Spottoro, we know that recall and marketing is valuable. We just can't stitch the two. I think we're waiting for a time where data providers like Similarweb, you've got Semrush and Datos, are going to be the ones doing that sort of slightly more predictive analytics kind of piece. Um, I'm not convinced the datasets are big enough now. And if you want to go down any niche, that data becomes very unhelpful very quickly. And that's not a criticism. That's just a fact of panel data is never as large as you think it is until you go down a rabbit hole. But for now, it's just visibility. We know that's key. We know that it aligns with SEO metrics. So technical SEO metrics, crawlability, indexability are all good proxies. And that's what I'm using, but I won't lie. I'm not happy with it yet.

We need better, better data, more platforms, better frameworks.

Indeed. I agree with Chris and I agree with you, Aleyda. I think we're in really early days of reporting. Like, for example, Google is just starting to roll out some new AI features where we can see a little bit of not so helpful, but a little bit of impression data on the URL level in Search Console, right? And Bing is giving us grounding queries. So we're in very, very early days of being able to kind of like triangulate what's happening. But then we have a lot of challenges like Chris mentioned. So for example, Not all of the refer traffic coming from language models gets grouped into, you know, the AI bucket. Some of it's going in direct, which is a challenge we've always had on the analytics side. So I think it is very hard to kind of like quantify what's really happening. So I think you do kind of need to use a variety of third-party metrics to kind of understand like brand visibility. For me, I think in terms of some of the new tracking tools that are coming out that try to directionally understand how language models are talking about your brand, Number one, you want your brand to be recommended more frequently over time. You want there to be more demand for your brand, more interest in your brand over time. So you can use things like impression data or trend data over time. And then obviously revenue. Don't just look at the organic channel. You might have to look at various different channels to understand the user journey. They might've talked to a chatbot and then 2 weeks later they searched for the brand and then came in through that. So it's like attribution gets really murky with AI search. But I think overall, if you're trending in the right direction in terms of brand demand, Brand recommendations, brand visibility, you know, leading in your category and driving revenue from various different channels, then you're moving in the right direction.

I never worked for a business where you just go and say, hey, trust me, I'm an SEO. And they would just be like, yeah, take my money, go do stuff. That never happened. It's always been hard with branding in general to measure and be able to come back and say, you know, these efforts generated exactly this much money. And I think it's the same here now with MLM, but There are things that people could do to kind of at least hint that, look, we're going in the right direction. The one would just be LLM traffic growing, that, you know, we've been doing nothing, it was quite flat, we're doing something now, it's not flat anymore. That's simple. But also, if you look at the pages that get LLM traffic and then you look at direct traffic for the same pages, if you see correlation between the 2 sources of traffic for those pages, you know it's probably LLM sending direct as well. Especially if you're not doing anything else to influence direct in any way. So that's another way to be able to produce some sort of chart to show your team of execs and say, look, we did this work on LLMs. We can see the traffic here, but also an indirect correlation, right? But one thing I think that most of the tools that track any kind of visibility in LLMs, they tell you, does your brand appear? Yes, no. They don't tell you in which context. Because if it's negative and it actually can harm your reputation and it can drive away people, you're not measuring that. I think that's something you need to have in your reporting to say, look, LLM had the wrong impression about us on XYZ subject. We need to do something to fix it because I can't measure how many people come to the website, but I can guess how many are not coming and will never be coming if they're reading this in an LLM.

Yeah.

And in an accurate way, I have to say, because there are quite a few AI trackers that already have sentiment in them.

And that's great.

Yeah.

But it is generic sentiment. A premium brand is not cheap and they don't want to be cheap. But a few of the prompts that were identified to be negative is because in the way that they were communicated in the answer is like, oh, and these are expensive. That's okay. They don't want to be cheap. They need to take into account your own context. To start wrapping up, I would like to ask you a last tip. It can be a resource, it can be an action to take, it can be a warning, anything that you would like to leave the audience with. Might be something that you haven't mentioned yet and you think it's worthy to highlight.

So for me, the biggest tip I've got is kind of coming off the back of lack of sure reporting is if we get into the habit of warehousing as much of our data as possible, like connecting directly into Search Console, into Analytics, Google Analytics, or whatever analytics package is warehousing our access log data. I know for a lot of SEOs that presents something new that presents— I don't know SQL, I don't know databases, I don't know how to do this kind of stuff. But the thing I found is we kind of right now we're in a world where we have to build our own dashboards and our own data from lots of different sources. And we need to kind of try and find proxies or directional values or correlations that you can't do without extracting and making your data portable. And if you've got large clients, or if you're working with lots of clients, there's not a single place that you can kind of do that. So for me, it's like taking real ownership of that data and learning how can we interrogate it? And how can we go beyond just what we're used to, you know, clicks, conversion, average order value, visibility, it's like, they're nice, but they're not enough categorically. So until you have that, until you start like correlating the price delta in Google Merchant Feed with your conversion rate in GA4, for example, until you start really getting into that level of depth, You're really going to struggle to know what's actually going on. And I promise you can't wait for data science teams or data engineering teams to do it because they're already too busy doing work for, frankly, teams that get more precedence than SEO or GEO. So I think get close to that data, get your access logs.

I'd say from a content point of view, do not optimize for LLMs, optimize for users. And I'm going to sound like Google, right? Because that's a Google tip. Telling us, but there's a reason for it because if you write content that's good for users, it's good for everything. It has all the right information. Maybe you're using visuals and videos and things that make it easier to read. Maybe you're structuring it in a way that's easier to read. Maybe you're using the journalistic way of giving out the answer from the beginning, which we know works well in LLMs as well. But also, I don't like reading 3 pages to find out one answer. So it, it, you do it for the user and it works in LLMs as well. And just like it, you know, it supposedly works for Google as well from a content point of view. From a technical point of view, optimized for LLMs and reinforced synthetic LLMs, they don't work in the same way. So you can't assume if I tested with ChatGPT and it worked, then it works for everybody, right? Lilia, I think you mentioned that, you know, LLMs use different other sources for search, like Google and Bing, and that we know Google and Bing have different results because they different algorithms, so you need to test with the specific LLMs that you want to appear in, first of all. And also test your website. Is it actually accessible to LLMs? I'm now testing a little tool that somebody shared with me that they're building. It sends an agent on your website and you can, you can give it certain tasks. So you can say, find my pricing page or find a discount code or whatever you would assume users are doing. And you send an agent through your website and then you get a report back on what they did, what they found and they thought might be related and what they decided to go with and so on. So literally mimic humans' interactions, but through an agent, because this is the future. People will use agents to complete tasks on websites. So you don't need a tool for that. You can just go in Perplexity and ask Perplexity to do stuff for you. Like you can just test it yourself, but try, try to mimic a user using those LLMs and ask Go do this on my website. And then if what comes back is not what you expect, somewhere on our website, there's a technical reason why that is not happening. So that's what you need to find out where it's broken. So content-wise, optimize for humans. Technical-wise, do optimize for LLM.

Totally agreed. Just from like a strategic and content perspective, it's kind of similar to things I've been saying for a very long time, but I think it's even more true now than ever, which is to focus on the real people that work at the brand that are real experts with real experience and promote them as much as possible. Because I think if you're in a business that's able to do that, because I think the way that things are trending, number one, language models love to look at user-generated content platforms where real people are having real conversations. They like to see real advice, real expertise, you know, people with real following. I think Google's taking certain steps to highlight individual creators who have big following. You know, being on YouTube, being on LinkedIn, being on those sources where language models are pulling a lot of that content, the best way to be visible on those platforms is to build a strong personal brand that people trust, right? So I think if you are able to represent your business or have a variety of different people that represent your business who are real experts, this is like the EEAT method, right? Like showcase those experts, have those people engaging in those communities because I think a lot of that expert-driven content gets pulled into the language model, and it's a reflection of your brand. It's kind of like a vote of confidence for your brand.

I will say that potentially one of the biggest real shifts has been that seeing AI search not only as a performance only, but also a branding channel. So how brands are being displayed and in what context, what it speaks, and so on. This is something that we haven't been very used to track, and now it's also fundamental. So it's critical to not only track, but also ensure that whatever message we put out there on our site and offsite is aligned to that accordingly as well. Even starting with schema markup, right? And, and the descriptions in it and the logos and, and the presence of our brand everywhere on the web indeed. And there is a really nice new tool created by fellow SEO John Will and Bobbins that is called Glippy. It has a free extension in Chrome and it makes it so easy to validate many of the structural issues that we have discussed here and understand if there's a gap. Again, many of these characteristics should also be checked for traditional SEO in any case, but yes, the explanations are the wordings there align very well with what actually matters for AI search and it's free and easy to validate. So go and take a look at it. I want to thank you all very much, Lily, Chris, Roxana, for joining me today. It has been a wonderful first episode to kickstart the series, and hopefully it has helped to introduce or reintroduce the topic in a way that is easier to understand, much more reasonable to tackle with resources and criteria that are actionable and easy to also share with clients and decision makers to clarify doubts and mystify it a little bit as well. Thank you very much for joining. If you want to learn more about AI search optimization, please follow Lily, follow Chris, follow Oksana. As you can see, they are pretty knowledgeable. And make sure to subscribe to the channel if you haven't yet, and stay tuned for all of the episodes of the series. See you again soon. Expect the rest of the episodes in the next following days and week. Have a good one. Thank you.

Bye-bye.

Bye.