Name: Marketing to Machines: GEO & the Rise of Agentic Buyers
Uploaded: 2026-04-28T18:16:13.836Z
Duration: 52 min 48 s
Description: Marketing to Machines: GEO & the Rise of Agentic Buyers

Transcript for "Marketing to Machines: GEO & the Rise of Agentic Buyers": Hello, everyone, and welcome. Really excited to have you all here. Before I hand it over, a quick word on who we are for anyone that's new here. I'm Alexander Bleeker, head of operations at the AI Marketing Alliance, which is a community built for senior marketing leaders navigating the shift to AI first strategies. If you're here today, you're exactly who this is built for. And speaking of reasons to stay engaged, we've got a $150 gift card for one lucky person in the audience today who has also happened to like our LinkedIn page. So if you haven't yet, check out AI Marketing Alliance on LinkedIn and give it a like, and we will email the winner at the end of the day. I also wanted to note that we have, a take home document, which is what we usually do, with these AMO workshops. So you can scroll over. Just check the docs tab there. I believe it's at the bottom, is the one for today, and it's gonna be, a document that you can share with your team and outlines everything you need to know from today's session. Now our speaker, Malte Lambert, has spent twenty plus years at the intersection of search, product, and marketing. And right now, that background is incredibly relevant. He is CPO and CMO of Peak AI, a platform helping over 1,700 brands measure and improve their visibility in AI search and is one of Europe's fastest growing companies. Before that, VP of SEO at Idealo, VP of product at Searchmetrics, and worth noting, he built and ran one of Germany's 40 largest blogs. Definitely a practitioner to his core. He's been recognized as a leader, by eConsultancy, a visionary by Adelo, and is one of the sharpest voices on GEO and the agentic shift that is happening in search right now. So it's exactly what we're here to talk about today. Malta, welcome. And without further ado, take it away. Hi. Thanks for having me. So since you already introduced me, there is nothing else I have to say about myself and can jump right in. I'm gonna talk today about the topic of GEO, generative engine optimization. Some people also call it AEO, answer engine optimization, AI SEO, many different terms out there. I don't really care about the term. I wanna talk about what it means for us and how we can measure it and how we can potentially, optimize our websites and brands to be more visible in that thing of, generative AI chats or a LLM based answer engines. I wanna start with a question. How's your your Google traffic going? Maybe you can you can write in the chat if it's, if you have more traffic now or if it's easier to get traffic now than, let's say, five years ago, most people think it's it's more difficult and, many websites see a decline in traffic. And and why why is that? For a very, very long time, there was a data number, like, floating around that roughly 60% of searches in Google are so called zero click searches. So meaning, people type a question, a search term, and do not click anything. And roughly 30% actually clicked on a website, and roughly 30% clicked on some kind of Google property like switching to a different search, doing a different search, etcetera, etcetera. And with the Google AI overviews, these these little summaries on top of the Google search results that are a form of of AI content. This number seems to have now increased to 65%, maybe 70%, which is also something that, for example, publishers are feeling in their traffic. And then there's also the Google AI mode where you don't just have an AI summary above the search results, but where you are basically having a chat experience similar to chat GPT where you just have a text answer. And there, the zero click rate is based on a couple of different studies, 95%. Meaning, for a 100 prompts, there are only five clicks. That is, of course, significantly less. And there was even one study that said the zero click rate is a 100%, but I think they used, they did it in a lab, and I believe they just used tasks that didn't require a click. And then if we look at JetGPT, here it's much, much, much more difficult to get reliable data, but it seems the zero click rate is more in the 99% range. So much, much, much less traffic than we are used to from Google even though we are already complaining about Google as SEOs, like, how low the clicks rates are and how difficult it has become to get clicks. Yeah. Here's some data. It's already a bit older. But like many, many publishers, especially the ones publishing informational content, are already feeling this a lot. There are fewer and fewer clicks coming out of Google. Yeah. Then what about chatty p d traffic? And for most brands, the realistic answer is probably what chat CPD traffic. It's it's maybe half a percent, 1%, 2% of traffic. On some websites, it can be 5%. But usually, traffic wise, it's a really, really, really low number. And, here's some data aggregated around across, I believe, thousands of websites, even more. And you see, chat g p t, if we measure it in clicks, it really doesn't play a big role. It is a tiny, tiny, tiny amount of traffic that arrives on websites. But if you think about it, I mean, you ask a question to CheckTpT, and and there are some brand mentions, but that's it. Right? There it's just a brand mention. I can't click it. There's no traffic. Even if I now want to click on one of the things mentioned here or recommended here, I often cannot click. And there are, of course, links for the sources, for the citations, but these are often not the entities that are recommended in the answer, but third party sources. So why would I even want to click that? The good news is the little traffic that comes out of these systems, it usually converts very, very well. I think I've now read 18 studies that said the traffic converts much, much better than organic search traffic. I've also read two studies that say it doesn't convert good, but one of them only looked at very small, very unknown ecommerce websites, and the other one could be an outlier because most studies say the traffic converts very, very well. Where the studies don't agree is how good it converts. Right? Some say it's four times higher conversion rate than Google. Others say it's 20 times higher conversion rate. It kind of makes sense that the conversion rate is higher because people are already very primed by by the answer they are reading, by the follow-up questions they can have before they actually click. But what the actual uplift is is a little bit unknown. So let's quickly recap. Organic click through rate is going down in Google. Click through rate in LLMs is much, much, much, much lower. LLMs send very little traffic to the open web. The traffic that is coming converts well, and decisions are often made in the chat. They are no longer made on our websites. So what does that mean for us? Is the chat traffic maybe just does it maybe just not matter at all? There is an interesting dataset. I think it was Accenture and and eMarketer, if I remember correctly. It's it's somewhere in the footer if you care. And it's about consumers and trust. And consumers who purchase something like a physical good, they have the biggest trust in the actual physical store where there's a salesperson that that recommends something to them. But then the second thing is already generative AI. It's ahead of social media. It's ahead of search engines. It's ahead of brand websites. Of course, this is a statement from people who might not know that you can also manipulate the answers from LLMs, but they are aware you can manipulate the answers from search engines. But it is the status quo right now. People have a lot of trust in what LLMs recommend them. And there's a second dataset. It's a study. It's from last year already from g two. And g two is a software review platform. So they made a study where do people get advice when buying business software. And g two has every incentive in the world to conduct a study where the result is that software review websites are the most important thing. But even in their own study, the outcome was generative AI is the most important thing for people to create a shortlist. And there's also a ton of data from, Google that recently came out on this topic where you can see that almost every b to b software purchasing decision is now influenced by LLMs in one form or another. It's a little bit different impact across the different stages of the buying journey, but LLM touches basically every b to b SaaS purchase nowadays. Then I have a client example, and I I really like this one because we talked about how there are so few clicks. Right? And this client, it's, an agency. They looked at their web analytics system. How many leads are coming from CheckTpT based on click based attribution? And in this case, their web analytics system was HubSpot, but this is not it is to HubSpot. This would be the same for for every click based web analytics system. The answer was 0%. 0% of the leads are coming from CheckTpT. And they said this cannot be true. They must based on conversations we are having with new customers, we know this to be not true. So they started asking everybody who clicked this button of please contact me or or, like, book a meeting. They asked them, hey. Where did you hear about us? And 2% of people said perplexity, and 21% said JetGPT. Now with self reported attribution, there's always there's always some noise. Right? Like, because some people click on a random thing, or they they click on whatever they know the most. But still, signal wise, is this marketing channel responsible for 0% of my leads or for 23% of my leads? It's a huge difference. Right? So the the the important thing here is that click based attribution makes you blind to cases where a decision happens in the chat, but the click is not happening. And those of you who are a bit older like me, you might remember, like, more than ten years ago, there was this concept of dark social. That was when you got clicks from the Facebook app to your website, and the Facebook app would not send a HTTP referral. So you couldn't know where the traffic is coming from based on typical, click based attribution. And then people were doing fancy things like looking if the preview bot from Facebook visited them earlier or what user agent. Was it maybe the in app browser of of a mobile device without a referral? Then it could be the Facebook app. And this was called dark social, and I think now we have this era of dark chat where we are just just not sure anymore, where some of the traffic is coming from. And, yeah, the younger ones here in the call might think I'm crazy talking about getting clicks out of Facebook, but it was actually possible back then. It it really did happen. Please believe me. Yeah. Then there's another interesting study about who's reading AI summaries. I don't wanna go totally into the details. We can very roughly summarize it by young people are much more likely to read AI summaries versus older people. And just with any trend in society where young people do adopt a new technology earlier than older people, it usually doesn't mean that they will stop using that technology, once they age and get older. It would rather mean that as a cohort, they will continue using it. So at some even if it's now primary young people, at some point, it will be middle aged people, old people using this. So, AI summaries are here to stay in in one form or another. So what you can take away here is, yeah, purchase recommendations. Like, people trust the purchasing recommendations. B to b buyers are using GenAI to create their their shortlist. And even though ChatGPT doesn't send, clicks, it it does influence, your your prospects. And there's this dark search or dark chat or whatever we want to call it that is a bit of an attribution channel. And there's also one learning on top that's a copy and paste error from a later chapter. So you got a little bit of a teaser there already. Now, you hopefully now believe me that this is an important topic. Let's talk about what to do. And one question is always I get, isn't it just SEO? Is it enough to do just SEO? And while SEO is the the foundation, there are some some things we need to talk about. This is a dataset of a couple of 100 prompts for the same topic. And I'm looking at the the brands that are mentioned in the answers of LLMs. And if a brand is mentioned, that's counted towards visibility. And if the website of that brand is counted, cited, that is counted as the source. So we see here there's a bit of a correlation. Right? Website like brands that are often recommended, they also have their website cited a lot. Like, typical correlation, We we are not we don't know at this point what is causing what. We can just observe there's a correlation. There's also a clear winner who is mentioned in almost every chat and is also the most cited website. And there there's even web even brands that are recommended in more than 10% of the chats, but their own website is never ever cited. Never. Not once. And, when I see this, there's one condition that must be true. There must be positive content about this brand somewhere on the Internet. Otherwise, they would not be recommended. And then there's a second set of things, and there one of them must be true. Either they are blocking LLMs from visiting their website, which is probably a stupid thing to do nowadays, or they make it technically impossible because they have maybe very slow website or a lot of JavaScript to for the LLMs to find their content, or they just don't have any content that is relevant. All of these things are easily fixable. And if this brand were to optimize their website a little bit, they would also increase in brand mentions significantly if their website were to become a source at least every once in a while. But, technically, they are successful without their website. This is the opposite example. This is a brand that is never ever recommended by AI for this topic in any single answer, but their website is a source in it's it's actually the second most cited website in this whole dataset. And this is because this brand writes about the topic, but doesn't offer a solution themselves. So while SEO wise, they optimized for it to have their website ranked for it, and Google ranked them for all these keywords, AI doesn't use doesn't recommend them because it recognizes this is just an information source. It's not actually the solution. But, of course, this brand has a lot of impact about what other brands are recommended as the solution. And then this is this, different example now. This is a customer of ours from the financial space in The US, and they offer a very specific thing in a couple of states in The US, and they are dominating Bing and Google, usually. They are doing very, very well in SEO. But for one specific topic, one specific set of prompts, they were invisible in perplexity. They had 0% visibility. And what was happening was this. A typical answer from perplexity looked like this. It was a list of 10 competitors, and then our client's website was the most cited source. And it actually was the source for multiple of the competitor recommendations. So what's happening here is SEO wise, this brand did everything correct. So when perplexity tried to write its answer, it found the most relevant documents on the Internet. Our client's homepage was in there, and then it tried to write an answer based on these documents. The bad thing is on our client's website, there was a little table or a list, And it said, look, we are not the right solution for everyone. If you wanna work with someone else, here are five competitors we can trust and whom we can recommend. And since it was a list, it was structured, it mentioned concrete entities, perplexity set. This is the text passage I'm using it I'm using. I'm citing it, and that's what my answer is based on. And then what this brand did is they changed the headline of that list a tiny bit and then put themselves on number one. And the next day, they became the most visible brand in perplexity for their topic. So the fact that they were already very, very good in SEO, it was the foundation, but it was not enough. Like, they did everything right in SEO already, but there was a tiny switch they had to do to get the maximum benefit out of it for LLMs. And I think this also describes the situation very well. SEO is the foundation, but it's not enough. So what we can learn here, having my brand mentions and having my website cited is highly highly correlated, but not perfectly correlated. And you can do everything right in SEO and still be invisible in AI search, which is why you should probably track it and see if you can optimize for it. So what are now the concrete things we can do? Let's start with a little case study. We we didn't do the work here. We just do the we just did the the success measurements, so it's not not advertising for what we do. Customer was momentum, and what they did is they published a lot of content, expert content. And when we started tracking, a set of prompts for them, they were only mentioned they were only used as a source in 3% of the chats. And after they published the content within one and a half, two months, they were at 60%. So in 60% of chats, they were suddenly used as a source. And in the same time frame, also their brand mentions, increased accordingly. They were originally mentioned in just 3% of the chats. Once their website became a major source in the space, they were mentioned for 34% of they were mentioned in 34% of the chats. So, thousand percent increase, and they could also tie this to traffic and revenue gains, from LLMs. So very, very successful campaign that they did there. K. What what content actually gets cited? With a little type typo from me on the slide. What LLMs love to cite are summaries. And this is not because LLMs are unable to read the whole page. Actually, LLMs are amazingly good at reading the whole page and summarizing it themselves, finding the important pieces of content. But right now, when you type a question to CheckGPT, it tries to give you an answer quickly. Right? So what it's doing is it is looking for a set of documents, maybe via Google, maybe via Bing. There's a bit of a debate going on there. They might also be using services like Exa or Tavely. Doesn't matter. Also, other LLMs might use other grounding other search engines for grounding. What's important is they use some approximation of web search to get relevant documents. But then they don't get the content from every single document in their context window to write the answer, but rather certain text passages, so called chunks. And you might accidentally have written a page where there's a perfect chunk to be cited. But if you take the time to write a summary of your page, put it on top, you make it much more likely that an LLM can find then the summary and realizes, hey. This is actually a perfect citable text. And these summaries work very well if you summarize the whole page, the whole article. I have also seen websites starting to summarize each longer paragraph. I haven't seen data that it works, and I think you need to be very careful with your reading flow when you are doing it. But what is often overlooked are charts and tables. As a human, it's very easy to understand in a table, okay, the the entry on top is the best, the entry on the bottom is the worst, or in a chart, what are the outliers, etcetera etcetera. If you use a reasoning LLM, a deep research agent, of course, it can do the same. But, again, when a user just types a question to check GPT, wants a quick answer, the LLM is not gonna do that. It's not gonna render the page, do OCR, try to understand the image of the chart. Rather, what is happening is it's looking for a text chunk that is relevant and citing that one. So if you have a table, it can be really, really impactful to write a one or two sentence summary, not explaining everything, but putting in the takeaway. And very important, you should not describe the table. If you write something like, this shape table shows the development of population for, US state capitals. That is correct. That's factually correct, but it's probably not something the LLM would cite. Right? You should rather write about what would a human take away from this. You could write based on, I don't know, data from the US Census Bureau from 2023. Is the fastest growing state capital and is the fastest declining state capital in terms of population growth. You can probably formulate that a bit better, but what I mean is the the core essence should be in there, the key information that an LLM might want to cite. Then something that also works very, very well are questions. The numbers that I'm showing here are from a growth memo from Kevin Indic. I don't know if these are the correct numbers. I don't know if it's 10% versus 19%. I'm sure other studies have slightly different numbers. But what I know is that this is directionally correct because I've also observed it over many, many, many different brands and websites. When you have question and, of course, also answer content, I think a similar thing happens like with the summaries. You are forcing yourself to write this very dense chunk of text that just makes it very easy for the LLMs to cite. Then what also works well are definitive text chunks. And, also, if you think about this logical, if you were a human looking for a source to cite, what would you rather cite? A text passage that says, the Statue Of Liberty is located in New York, full stop, fact, spoken with authority, or something like, yeah, I I think the Statue Of Liberty is in New York, if I remember correctly. Yeah. So if you want a certain text passage cited by LLMs and increase the likelihood that it gets cited, it should be a definitive statement, not vaguely spoken. Of course, other passages of the document can have some vagueness where it's appropriate. Right? You shouldn't just make up things and suddenly start talking with with authority about topics where people are just guessing. And then the the next part is the entity density. And if you think back to the examples we use, like, I don't know, the Statue Of Liberties in New York that actually has two entities, and then also a relationship between these entities because entity one is in entity two. And this is also something that LLMs seem to prefer when they are citing. This also makes sense because if something is dense in entities and a statement is dense in entities and even has relationships between the entities, it has it must have a high high degree of information embedded with within there. K. So this tells us a little bit about, like, how to write, what to write, but what what is the general content format. And with this part, I'm always a little bit embarrassed because to be fair, it's embarrassing. We see here a question for the best CRM. And then in the answer, they are HubSpot and Zoho. And in the sources, they are also HubSpot and Zoho. And if you think about this, if you ask other humans for what's the best CRM, you would probably want to ask from know from other users, from other practitioners, from independent third party, say, what what is the best one? What worked for you? You wouldn't want to know from these vendors because, obviously, everybody says, yeah. It's me. It's me. And this is not specific to HubSpot and Zoho. Right? There's also what was this? SuperOffice, has this listicle that that they are the best CRM. NightWatch has an article about the best, SEO tools, and the best SEO tool is NightWatch. And the second best SEO tool is the free Chrome plugin from NightWatch. And then we have Monday, article about the best CRMs, and surprise surprise, Monday is the best CRM. And we have, FlatLogic, with an article about, 10 best pipe coating tools. And surprisingly, Flat Logic is the best pipe coating tool according to Flat Logic. And, many, many large brands also like Rippling, Blaze, many others are doing this where they write a ton of listicles. And then if you look at the column where it says mentioned, you see that they always put themselves on number one. There there is some variation who is number two and number three, but where they are always sure is they are number one themselves. Yeah. So why why are they doing it? Why are so many large brands writing these articles where they say, hey. I'm the best? Because it works right now. LLMs, when you ask them for what software should I buy, what software should I use, what is the best software, they love listicles. They eat them up. And they even eat up these self preferential articles where you just write it yourself on your own website. There are some risks onto it that I will mention a second, but just factual observation right now it works. Yeah. We did a study where we looked at different industries to understand if a listicle is the source. Is it an independent listicle, like a newspaper, for example, that has actually tested a couple of different things, or is it vendors fighting about themselves? And in the SaaS space and in the professional services space, we observe the highest ratio of these self promotional, listicles. Important professional services includes agencies, and agencies includes SEO agencies. And they were, I think, the first ones to notice this, that it has an impact and then started doing it. But it does also work in other industries. And I believe in the SaaS space, the the ratio is so high because the SaaS buying journey is very complex, and it was one of the first way I started having major impact. So multiple players noticed, hey. We need to do something. Now these listicles that you just write on your own website, they work amazingly well in Google AI overviews. It's they are actually not included here in the charts, but the value is even higher. They work still okay in Google AI mode and perplexity. They don't work that well anymore in ChatGPT. So I believe ChatGPT, at some point last year, did something where they recognized this and started putting a little bit less trust into a list of the 10 best CRMs that is published on the Salesforce website, for example, with Salesforce as number one. By the way, these listicles even work for made up brands. This is, something that a German SEO agency did, Claneo. They invented a matcha powder, and they wrote a couple of listicles about the best matcha powders in Germany, and they included their made up one, matcha teos. And now LLMs from time to time are still recommending the matcha teos, matcha powder even though it doesn't exist. It was just mentioned in a handful of listicles that they specifically created to trick LLMs. Now are listicles the only form of content? No. There are there are many other forms. One that can also work very, very well are comparisons. I unfortunately had to anonymize this a lot because, the brand that is doing this is is our client. I searched in Google AI mode for red brand versus green brand. And, of course, then there's this comparison article. Right? Red like, this comparison written by AI mode, red versus green. And in the sources, there's an article, red versus green, and the source is the red brand. Of course, this gives the red brand a huge advantage about the green brand. Right? Because whatever they use as criteria for the comparison is likely showing up in a in the AI mode answer. But now it gets a little bit crazy and scary. This red brand also wrote articles, green competitor versus blue competitor, green competitor versus teal competitor, green competitor versus purple competitor, and green competitor versus yellow competitor. And all of these were pulled in as sources by AI mode. This is, of course, a very, very extreme example. It normally doesn't work that well, but it can work. There there's also an example here from Rippling. They compared SRP and Salesforce on their own website. And then only on the very bottom mentioned by the way, Rippling can be an alternative to both. And they do this for a lot of brands, where they just compare everyone. And they didn't even stop at two competitors. They also started something like Brex versus Mercury versus Rippling, so comparing, three competitors. And the the reason I picked rippling here is I usually try not to publicly point out things that are a little bit sketchy and on the line, but a couple of SEO influencers have written about written about this publicly. So in certain circles, this is already well known that they are very, very aggressive. So that's why I thought here it's okay for me to use them as a as a concrete example, because I think it's always more interesting for you to see concrete examples. But usually, I try to avoid calling out people. Now I can create content on my own website. I can do listicles. I can do comparisons. Of course, I can also do regular articles. Is there also stuff I can do outside of my site? Yes. There is. If I take a set of prompts and I type them into my l l l m of my choice, and I look at the sources that I can do this manually or with the tool, and when I then notice most of the sources are my competitors, then I should, of course, create content on my own. Right? Because I can never ask my competitors to please include me. But if LLMs are willing to cite my competitors, they are probably also willing to cite me. If I see primarily editorial websites, so like news publishers, there there are three different things I can do. I can just do traditional digital PR. Have my my in house team. I could call my agency. I can tell them, hey. I always see these 10 big publishers as sources for our prompts. Please get us on there as much as possible. But that's not everything I can do. I can also do advertorials, where I just pay to have my article placed. And on certain publishers, I can even pay an additional fee, and then they remove the notice that it's an advertorial. This might be legal or not legal in certain jurisdiction, blah blah blah. Some laws apply. Again, I'm just reporting what works, not giving a concrete recommendation to do it, but I can tell you that it does work. Or I can go the affiliate route Because many large news publishers today, they don't just have the journalistic team. They often also have a commercial content team. And if their commercial content team wrote an article, the 10 best, vacuum cleaners, and you have a vacuum cleaner. You can give them a call, and you can try to give them a very, very good affiliate commission. And maybe in their next independent test, your product will be number one on their listicle, or you can sometimes even outright buy your placement on listicles. Just pay a certain amount of money, and then you go to the number one spot. And you are not doing this to reach the readers that read that article. You also don't care about a couple of affiliate clicks. What you care about is this listicle is a major source for LLM citations, so I wanna be included in that listicle for purely that reason. And on the on the advertorials, please don't cite this number. This is not a gen generically true generally true number. I monitor a set of prompts where I know that a lot of advertorials are used. It's, a specific niche in the insurance space. And since October, over two slightly above 2% of sources used by the LLMs are advertorials. So paid placements on newspapers, the exact value switches a little bit over time. This chart actually ends a few months ago, but I have, this morning asked the the the I asked Claude to connect via MCP to my dataset and PKI and and do the refresh, and it's still slightly above 2%. So this is still true. LLMs are citing advertorials where it's very, very obviously to a human and should also be obvious to the LLM that it is an advertorial, a paid for article, but they seem to, accept that. If you see primarily user generated content, social media sites, if you are the SEO team, please don't think that you suddenly need to own this channel and create content there. Please talk to your social media team, your brand team, your influencer marketing team. Encourage them to create more content on these platforms, maybe by maybe with including certain keywords. And I would always recommend to think about very hard, do I want to create my own channel, or do I want to work with creators? On on YouTube, Instagram, TikTok, you might even consider doing both. On Reddit, it's a little bit tricky because most subreddits really don't want brand accounts. There are also some brand accounts that work very well. There are also some agencies that offer completely anonymous, like, services where they use fake accounts or a large influencer network to to publish content that speaks positively about brands. Again, there might be some regulation against that, but I'm also seeing it work. What is very important? In best practices, you always see be active on YouTube and Reddit. And this is often true, but not always. Especially if you are looking at non English speaking countries, Usually, these user generated content platforms that are very US centric, they go down a lot in terms of how often they are cited, how important they are. And for some topics, it could also be that TikTok and Instagram are the most important sources for LLMs. If you look at, I don't know, jewelry for Gen Alpha, these these sources are more used just because there's more content on them. And what is important? You would sometimes read news like Reddit doubled in LLM citations or AI overviews are no longer citing YouTube shots. And, yes, this always sounds very exciting, but the reality is there is volatility in the market, and things sometimes go down, sometimes go up. Like, here is, Reddit in perplexity. It went down a lot. So often it's used as a source, then it went up to higher than it was before. YouTube went down, and the same time LinkedIn went up. Here we have YouTube and AI overviews. Please keep in mind, both AI overviews and YouTube are Google products, where AI overviews already were the most, sorry. YouTube already was the most cited source in AI overviews. It now it it doubled again, I think. So these changes are happening, and there's always, like, this big excitement on LinkedIn about it, like, from from from influencers and creators. Please be a little bit calm about it. Do not shift your content production strategy just because YouTube had a big increase or slight decrease. Look at what are the top sources across LLMs over larger time frames time frames and try to have a sustainable approach to how you are handling that. So what can we take away here? More citations need to prevent mentions, content works, summaries and questions work. Try to speak in a different definitive entity than its way in the chunks you want to be cited, listicles works, say works so well they even work for nonexistent brands, comparison pages work, get inspired by your competitors if they get cited, and so on and so on. I think we will also send this out, and you can read it for yourself. And I think I've spoken for forty minutes, and I think Alex is coming on stage to tell me that there are questions. That was perfect timing. Wow. That is insane. First of all, incredibly incredibly informative as a content marketer myself. I was taking notes, So thank you for all of that. And we have a lot of questions. We have a lot of questions, actually. The first one I wanted to share and, actually, Tori, if you don't mind helping me out and sharing Eric Smith's about, the question is, how should we discuss a decrease in traffic with c suite, that will be convincing? Because I I I know I've I've had this question myself, and and it got the most amount of upvotes. Yeah. What what I always try to tell people is that it's just a reality that now fewer clicks are happening, but people are still spending the same amount of money to buy microphones for their laptops or or or or soda or whatever. And it's just switching where they make that decision. And in the past, they clicked on 10 websites and consumed content there. And now they consume content in a LLM summarized form, but the LLMs are still pulling that content from somewhere. And once a decision has been made, the money is still being spent. So I would just make c suite aware that this world of traffic is not coming back. And there's a lot of material also from Google, some statistics you can cite. Rand Fishkin is always publishing on this topic. Many, many other authoritative people in the space. It's just the reality that there's less traffic. Now for brands that are selling a product or service and just use the website as a marketing instrument, this is actually not a terrible thing. Like, Starbucks can sell as much coffee as before. Mercedes can sell as many cars as before. They just need to understand that their marketing spend, their attribution needs to change a little bit. The problem are, of course, publishers that are ad financed and affiliates that make money with clicks, like price comparisons. For them, it's very problematic. And, I mean, I spent five years working at a price comparison website before I joined TKI, roughly a year ago, so I know exactly, how that is. And for people working on this business model, I would recommend them to think very, very hard about how to adapt the business model because there will be less traffic, not more, in the future. And there are ways to pivot to deal with it, but I think that's a case by case, decision. There's no no overall playbook, that you can apply. For companies that don't actually need to click, that actually need a buying decision, I think for them, it's it's an easier easier process because the core business model stays the same. They just have these nice free clicks going away. Interesting. And you you touched on this earlier, and Eric has a a similar question. What are your recommendations to positively influence dark search? Do do you have, is, there a smoking gun? I don't know. of time that I can repeat my presentation? No. I mean, what you want to influence in the end is that your brand is recommended, right, or your product. That is usually what you recommend what you want to achieve. And, on the last slide that I showed, there there there were a lot of things you can do to get your brand mentioned more often. And then metrics wise, like, how to make it somewhat measurable, I I recommend a three three front approach. One is ask people where did you hear about us. So self reported attribution. It's the it's easy. I mean, for b to b business, it's easy. For consumer, you need to think about a little bit about sampling. Every question brings down conversion rate, etcetera, etcetera. The second thing is if you have access to your log file. You can actually see how often are the bots from the LLMs coming to your website. For some LLMs like ChatGPT, they even reveal exactly, is it a GPT user bot that's only triggered when there's an interaction happening in a chat and your URL is requested. That can tell you which of your pages are used for grounding and how often that is real data. But it has the problem that you cannot benchmark it against competitors, and you know nothing about the prompts. You only know this page was requested. You don't even know if it ended up influencing the answer because maybe it was requested. And said, I actually can't use any of this text. So it's it's real data. Right? And it's absolute. There there are no gaps, but it's not that insightful for smaller websites. And then the third thing is prompt tracking where you define a set of topics and then other criteria like intent, persona, sales funnel stage. And then for these combinations of your dimensions, define a set of prompts and track those prompts to to then deduct things like what is my approximation of my market share of is my brand cited, my website cited? And all of this always benchmarked against competitors, and you can drill down by the dimension that you thought about in the beginning if you checked your prompt set properly and whatever solution you're using to to track it. And with these three things, I believe you can get quite a good picture of what's happening, but it will never be as good as performance marketing. Right? You need to think about this more as a brand marketing exercise where you cannot attribute every single click, and you just do not know for every single click. But it did come from And you you will have some uncertainty because if JetGPT recommends me Goldcast, for example, Yeah. and there's no link to click, I might type it into my Chrome address bar. Yeah. I might type it into Google. So there is no perfect attribution where my specific user session, my specific journey is attributed. So there is some uncertainty, and then this brand marketing mindset can often help better than the performance marketing mindset. That makes a lot of sense. I think we have time for for just a a couple of questions. One here, and for those who are, not aware, Melta's in Sweden right now, and he's gonna be doing a, a webinar on, like, international geo. And there's a question here, basically, all that, so we're maybe helping you out prep for the webinar tomorrow. The question is, for a company selling internationally, is AI search visibility basically a separate effort per market per language, or does strong English language content carry over? So it's definitely a local effort, but less than an SEO. If you because what's happening is even if I'm now in Sweden with a Swedish IP address and I write my prompts in Swedish, JetGPT, Gemini, they will pull into some degree English sources. And English sources usually means US source. Right? Because the English speaking Internet is very much US dominated. And that gives, of course, US companies a huge advantage. There are, of course, also local sources that are being pulled in. And if I want to be visible at Sweden, I need also my content in Swedish, and I need to write specifically about Sweden. This actually depends a lot on the content. Like, I just did the study that I will talk about tomorrow, so quick sneak peek. If you have a topic like project management software, that is the same everywhere. Right? Like, Trello is not different in Sweden than in The US, Right. and and people use the same things. The the one big exception is always China. So let's let's cut that out for a second. But then you have stuff like employee tax management. That is, of course, super different per country, so there it's much more localized. And there's also for some international companies a big challenge because we have, for example, of car companies as clients. And in their, first Spanish prompts written by potential consumers in Spain, sometimes South American sources are used. But the problem is the car, it has the same brand. It has the same model name. But what is sold in South America is very different from what is sold in Spain because it's, like, 25 to 30% cheaper, and it has much fewer security features. So it might say something like, no. This car doesn't have a side airbag. Even though, like, it's legally required in Spain to have a side airbag. I don't know if that's true, but I think most of cars now have the side airbag in Europe. And, this caused a lot of problems. And, in this this can even go across, like like, for for Spanish, it's especially problematic because it's spoken in so many different countries. Also, for English speaking countries, this is a problem if The UK version is different from The US version. And sometimes the LLMs even pull from completely different languages. Like, I have a set of prompts that I'm monitoring in The US and The UK in the car space. And I think from 50 different countries in total, I have seen here and there a source pop up in the LLM that is, of course, often just wrong because it's it's a different model, different standard. So it's a huge pain for for multinational companies. That's super interesting. We have, I think, probably time for one more. I think we're just a bit over time. But, Kelly has asked, and and we've had a few upvotes here, how do landing pages from tools like Marketo work with LLMs? I would have to look at Marketo specifically, so I can't answer that exactly. In general, as long as there's a unique URL that is ideally part of your domain, maybe via subdomain or something, and the HTML like, the content is rendered in the initial HTML, then it's totally fine. If it's completely client side rendered JavaScript, it might be that it's not working. There are some tools. Like, Lily Ray, for example, just published a tool that that tells you a little bit about how are LLMs most likely perceiving your website. But you will have to test, to be honest. I I can't tell you about Marketo. I also could like, the way I remember Marketo from when I used it, I'm also pretty sure there are 15 different ways how different companies have implemented this. So I I can't give a general statement about Marketo. What I know is that until very recently, Lovable, for example, produced websites with client side rendered JavaScript That was very, very difficult for LLMs to comprehend. They recently published an update that should fix that. I haven't tested it. But, yeah, that's usually the biggest thing, the the JavaScript topic. Okay. Amazing. Malte, thank you again so much, for your time. This has been incredibly informative. I know the chat has been going like crazy. For everyone who has asked, I know there's been a lot of questions. This is recorded. It will be sent out within, I believe, the next twenty four hours. And, again, if you go into the docs tab, there is the take home document which summarizes everything that you've learned here today so you can share that with your teams and really get going with this right away. I know I definitely will. So thank you again, Malte. We're so grateful, to have you and and spend your time with us, and, we will see you all at the next webinar. Take care. Thanks for having me.