RSS 66: AI Visibility for Modern Marketers: Malte Landwehr on How to Measure LLM Search and Win More Brand Mentions

0:03Does your experience with AI sound a little something like this? You've been prompting for 20 minutes. The output is polished, confident, and completely useless. That's what happens when AI doesn't know anything about your business. HubSpot's AI works with your actual customer data, every interaction, every signal your team has generated over time. So instead of prompting and hoping, you ask once and you ship it. Check out HubSpot.com, the agentic customer platform for growing businesses.

0:34when someone asks chat GPT cloud perplexity or grok a question about your brand or your category do you show up or are you completely invisible that's the question that a lot of modern day marketers need to answer when they go into the boardrooms of their CEO CMO or executive team brands across the world are looking for AI visibility but not just that they're looking for results. They're looking to see whether or not the people who are actually filling out the forms, actually buying, are connected to this rise of AI and LLM usage. My next guest, Malte, is the CMO of Peak AI. And in this conversation, we go deep, deep into what they're building, deep into the research that they have done.

1:16And we uncover some of the core fundamentals that are needed to thrive as a marketer today. This is an episode that I think every CMO needs to hear. This is an episode that I think every marketer needs to hear. If you're playing the long game, you're in the right place. Welcome to The Ross Simmons Show.

1:48Malta, thanks for joining. I am super excited for this conversation, but I want to start with a big one. And it's one that has everyone talking. What is the difference between this new world of AEO, GEO, LLM, SEO, EIEIO versus the traditional thinking of SEO? Is there a difference? Is it one and the same? What's your take? There's a difference, actually many differences. And the biggest is, of course, with traditional web search, it was all about getting visitors to my website. And then I could do all kinds of things with them. Give them a retargeting cookie, show them my newsletter sign up, count them as a visitor in my web analytics.

2:26And with AI search, the whole game is about having my brand recommended. But this might happen without a person visiting my website. This might happen without somebody reading the words that my copywriter wrote on my website, without experiencing my corporate identity that I put a lot of time and effort in. For brand marketers, it starts with the difference that people read it in their normal user interface instead of my custom font that I have on my website, right? So the way how people experience brands is very different. And that, of course, then has all kinds of impacts on primarily, of course, SEO as a marketing channel, but also, for example, on measurement of what actually is successful content.

3:08How do brands think about their customer journey? How much can they actually track the customer journey while it's happening? And if we think about the whole ecosystem, there are also some participants, especially in the affiliate marketing space or price comparison who used to be part of the user journey in traditional web search, but are not like a direct part of the user journey when it comes to AI search. But of course they are a source used by the LLMs, but how they interact with the ecosystem is very, very different. Right. Could you take those who have not been familiar with Peak, the work you guys are doing, et cetera, and take me back in the time around the early stages, how you got here, why you're solving this problem, what is the journey for Malte and the team?

3:50Yeah, the journey is that around the time that JGBT started adding their web search functionality, before that you would always just get the answer from the foundation model, the three founders got together and also talked to me via LinkedIn as like, hey, can you be an advisor, support us a little bit? And they had this idea that this will change how people search and this will require different tools for companies to measure their success. and there was very, very good product market fit basically from the beginning. So the company is now 17, 18 months old almost and we are at way over 10 million ARR.

4:30We communicated that number a few months ago and then have not given a public update. So I also can't give one here. We are at over 3 ,000 paying customers and have seen very, very rapid growth. And our idea, our vision is that AI search can be measured and it has to be measured differently than traditional web search. And many, many companies have a need to measure it and we want to enable that. And I think the measurement part we also figured out pretty well. And now we have also started adding functionality that allows brands to increase their visibility, not just measure it. So if you were to describe the way to measure this, what would you say is the answer?

5:09Is it simple? Is it straightforward? How do you measure the results of this type of effort today? Yeah, it's definitely not simple. I believe if you want to do it perfectly, there are actually four different things you have to track. One is if a user signs up, just ask them, where did you hear about us? And then make chatbots one of the options. And once somebody says, yeah, I found you an LLM, maybe also ask them which prompt they use. Can be a very nice source of information. This is self-reported attribution. It, of course, has some issues, right? Of course. A certain percentage of users lie, especially in B2C.

5:43Also, users just forget. So you don't get perfect data, but you get data for every session, every user, potentially. They might also just put ZZZZZ or something in the answer box. Then the second thing is you can check your web analytics, because even though there are very few clicks from LLMs, there are some. And especially if you're in an industry where people need to click to do a transaction, e-commerce, app downloads, font downloads, these kinds of things, there's actually a chance that a significant part of people is still clicking. Maybe 1%, maybe 5 % of your SEO traffic. You might also see that in terms of clicks coming from LLMs.

6:25So while it's a much, much lower level of traffic, you can still use it to see, is this traffic converting? Where are they landing? Et cetera, et cetera. then you can look at your log file. Because right now the LLMs are very bad at caching. So if they use an article on your website to generate the answer to a prompt, a special user agent, in the case of ChatGPT, it's a ChatGPT user bot, will come. You can see that in your log file and you can use that to understand which articles are being requested by the LLMs how often. It is especially if you verify the IP address to avoid people who spoof the user agent.

7:05This is real data. This is the truth. This is the reality. It has a couple of disadvantages, though. You only see the URL. You have no idea what the prompt was. You have no idea if your brand in the end actually showed up in the answer. You actually don't even know if your article became the source because maybe the LNM requested it and said, this is garbage. I'm not going to use it. and threw it away. And then the fourth dimension is prompt tracking. This is now synthetic data. But what you can do is, and this is also something where PKI is helping together with the other things I've already mentioned, generate a set of prompts.

7:44And then we simulate a user going in their browser to, for example, chatchip.com or gemini.google.com perplexity, typing in the prompt. And we look at the answer. We also look at the sources. We can often get the fan out queries, the searches that the LLM is doing in the background. And from that, you can actually get something like a visibility score. Am I visible or not? Right. You can get a share of voice. So you can compare yourself to your competitors. That is usually what your boss or your boss's boss will want to know. Because they don't care about average position and visibility. They need one number per marketing channel, right?

8:24And that is usually market share or share of voice, or some people call it share of model in order to have like a new and fancy term for it. Right. But in the end, it's all the same. And this also shows you the sources that are being used. So you can find URLs that are being used as a source for your prompts where you are not mentioned. So getting mentioned there is important. You can also get inspiration for what content to create by seeing which of your competitors' content is being cited. So the prompt tracking is actually this ideally daily feedback loop. That also tells you if what you are currently doing is working, it can enable your future work because it actually gives you some homework.

9:04And the others are more like the measurement after the fact to see, OK, what I'm doing, is it actually moving the needle? Right. And if you look at the combination of all four, you can actually measure this really, really well. Very cool. When you look at those four different angles, if you were to stack them in terms of priority, best, most efficient, the number one versus the least reliable, like how would you rank them? It depends where you are as a company. If you are still not sure if you should invest into LLMs, start with the self-reported attribution. And then if you suddenly see, okay, 5 % of our customers are already saying they come from LLMs, this seems important.

9:44And this is not an unrealistic number. And in the B2B space, we sometimes see 20%, 25%, even 30%. And then the next priority for me would be prompt tracking because it's the only thing that gives me something to work. Like it gives me prioritized things that I can do, things that I can do to improve because as a marketer, measurement is important, but I also need to do the work, right? I need to know what content to create, where's my gap, how am I standing versus the competition. And then the other two, I would say, are add-ons, especially the log file analysis for a very, very small website is a little bit boring.

10:24for larger websites with hundreds of thousands of pages that also have different areas, like a help center, a blog. It can be very, very interesting. Often what companies see is that their help center is actually the most important content format for LLMs, which is an area of the website that the marketing team is often ignoring. True. And so suddenly your customer support, CX team, is actually in charge of how are you perceived by LLMs. So this is then often a very, very important insight for companies. Very cool. When you see the shift where things like the help center or FAQs or even like in a SaaS company, it might be the API documentation.

11:04Like a lot of these things are being crawled. Are you seeing clients shift their strategies to place more attention there and the marketing teams getting involved? Or are we still in the early innings, early days of this entire shift? So we are still in the early days, but there are companies that have realized this and put a lot more attention there. And the most extreme example I've seen was a company where the leadership team gave 16 different teams an AI search goal around citations. Because customer support, they are in charge of citations of the help center. They have a B2B marketing team that has their own little blog.

11:41They are responsible. they have influencer teams creator relations teams these influencers create content on social media that is being cited they have a corporate communications team investor relations team all of them create content either on their own website or on partner sites right that can influence citations affiliate marketing teams and many many companies are starting to also get goals around AI citations because many affiliates, like the publishers, were actually very strong in SEO in the past and now their traffic is drying up a little bit. And now it's a new revenue stream for them to tell advertisers, hey, my list of the 10 best vacuum robots is cited by ChatGPT 500 times every day.

12:30For 1 ,000 euros per month, I put your vacuum cleaner on position one. And then they don't earn money anymore per click, but basically indirectly by number of tokens they have influenced or number of conversations they've influenced, of course, with a proxy because nobody can measure that perfectly. That's cool. Yeah. As I said, many, many different teams have an impact. So you mentioned a lot of different teams now being involved with what traditionally would have been a marketing problem. When you look at leadership teams who have executed that type of a shift where the silos break and other teams are now getting involved, what do you see as the trend amongst the leaders who are able to do this successfully to get PR, customer support, technical development relations, everyone at the same table moving towards the same goal?

13:19what's the shift and what's the difference in the cultures within those companies versus those who are still scrapping and fighting and trying to figure it out so when it comes to success they're actually very different cultures it goes from completely enabled teams and the ceo is just saying this is the new direction and then everybody wants to contribute but it can actually also be successful if it's a super strict hierarchical top-down structure and the CEO looks at their chief customer officer, chief marketing officer, chief technical officer and says, hey guys, your personal bonus next quarter, it depends on AI citations.

13:59And suddenly they all can align their teams under it. If it comes to dysfunctional, it's usually always a combination of either unclear leadership so they they give this direction together with 16 other directions right because we also need to make more money from the existing traffic that we have right and we need to prefer for a completely agentic web without humans and we need to assume our google traffic is zero tomorrow yeah like if you give teams too much to change and to adapt to it doesn't work right or it's the priority for a week and then it's the next priority there are also many different ways to be not successful.

14:38Yeah. So you're spending a lot of time talking with clients, learning more about the technology, involved with the product, advising and now involved deeply. When you fast forward, let's say three years out and you think of agentic commerce, you think of agentic buying, what shift do you think is coming that not a lot of people are yet paying attention to, but they probably should be spending more attention to because it's going to impact the way consumers buy in the future? I think most of the people I talk to are probably overestimating it. But then, of course, I live in my bubble, right? Yeah, yeah.

15:10If I think of the general public, I believe many people still think it just will not happen. Like, there will be no agentic commerce. Okay. And there are still many companies that have zero preparation for it. Right. But if we think about behavior of humans, there are purchasing decisions we want to make, right? Which clothes we buy. Yeah. We at least want to see options and click, right? There's not something where anybody's going to say, buy me six blue t-shirts and one red one i mean a few people will like i might i think i would

15:45but i agree i don't think most people would i think i think i'm an anomaly in that factor yeah yeah but there are also purchases like somebody works in the procurement department of a manufacturing company and they have 32 different kinds of screws to order between 10 000 and 12 thousand pieces per different kind of screw yeah and they only care about two things it should be cheap yeah the quality should be good actually three things it should also be shipped on time right they do not care who is the seller they do not care about anything apart from this right and this is something where right now a human is tasked with checking probably the websites of five different vendors or calling them and this could be done by an agent yeah right this is a very, very clear rules for this transaction.

16:34Also from the seller's side, this could be an agent. They could just quickly negotiate, have a done deal. And I believe in that regard, a lot of things, especially for B2B procurement, can move to some form of agentic commerce. It makes no sense that for every new employee, I have to go to Apple and order a new MacBook. And then I go to Microsoft and I need one more license. And then I click on Miro that I need one more license and Figma one more license. Right. A lot of this provisioning will probably happen in an agentic fashion. Yeah. And of course, some of it could have been automated forever with APIs, but it's difficult.

17:13It needs to be maintained. But now that we have MCP as a technology, you just tell your Claude, connect to the MCP of this company. Yeah. You write what you want to do. And magically it happens in 90 % of cases. Yeah. Which is about as good as telling your new intern to deal with it. It's true. Yeah. good enough for for the corporate world right so i think there is there's the first big shift that's cool and then when it comes to to consumer purchase behavior i think a lot of the research part will become agentic so people become get like their own little briefing like these are the three headphones you need to look at based on your requirements right but in the end many people will want to see them and feel like they are in charge right and make a decision on their own Very cool.

17:59When you think about prompts and the way that brands track them, there's good ways to do it. There's somewhat bad ways to do it. Is there anything that stands out from your perspective that you see the best brands always do when it comes to identifying certain prompts that they want to track? Like in search, we always had the commercial intent. We had transactional intent. We had informational intent. All of those types of levels of search intent. When it comes to prompting, should you follow a similar pattern? Should there be something else that we look for? What's your perspective on the right prompts to track?

18:33So I think there's a couple of different dimensions we can use to look at this topic. One is I would always start with thinking about my topics and intents within these topics. and maybe also personas. If I want to have all my marketing split by personas. If I have a very complex sales funnel, I might even have sales funnel stages. Or if I want to be hip and modern, I'm using the micro moments from Google, from the messy middle model instead of sales funnel stages. And once I have this setup, then I can start thinking about what are the prompts that I'm actually using to express this intent for this topic in this final stage by this persona.

19:16Right. And there, the good news is the exact wording doesn't matter too much. There was a very interesting analysis where Rand Fishkin asked his followers on LinkedIn, tell me how you would prompt if you want to buy noise cancellation headphones. Right, right. And I think like 144 people participated. Yeah. And it was 144 different prompts. and Rand Fishkin was very nice and shared these prompts with us and we tracked them many many times and all the LLMs and it turns out for 90 % of these prompts they are the same brands in the answer. So while people used very different words to express their need for the LLM it was kind of the same because LLMs are really good at understanding the current topic the context the goal.

20:05I mean, you can start writing a sentence in English and then you switch to Klingon in the middle of the sentence and finish it in French. Right. And you start with I want to do and switch to the user wishes to do. Right. And it still understands what you want. Right. It's very, very good at that. So the fact that people use different words is actually not that much of a problem. Right. That's very good news. Yeah. Because it means you do not need to know what is the exact word that Ross used yesterday to look for something. Also, you don't need to know that because the prompt that you used yesterday might not be repeated ever again because there are also ways to observe real prompt data and real prompts are mostly unique.

20:49Like if you try to calculate traditional keyword search volume for prompts, the only thing with the traditional volume is thank you, try again, hello, bye, like these kinds of things. Right. So we should not think about I need the exact words. We should think about a topic, an intent, maybe a persona, maybe a context and how to express that. Now, of course, I need to be careful because my target audience uses specific language. Right, right. And this language, of course, I should also use in my prompts. My favorite example is from, I think it was like 12 years ago, I did consulting for a pharmaceutical company.

21:31and big shock, they were not ranking for the keyword heart attack. Right. But they were producing one of the most successful medications in the space. Fascinating. And they had content on it. And each content piece was read by a lawyer and a doctor. So they had$1 ,000 plus cost internally to produce a content piece. Why was it not ranking? Because people were typing heart attack into Google. Right. But their article was about myocardial infarctions. Right. It's the same thing. Medical professionals, please don't cancel me for this. I understand. There are nuances. But it is one of the medical ways to describe one of the phenomena that a layman would describe as a heart attack.

22:18Right. And this is, of course, also true for prompts. Because if you ask about mycardial infarctions, you will get very, very expert content. If you ask with a couple of typos about heart attack, you will get a simplified version. So you need to know the language of your target audience. Where can you find it? Of course, traditional keyword research can still help you. If you have a search function on your website, track what people write there. If you have a chatbot on your website, track what people write there. if you have zendesk or similar for customer support ticket read what they are writing if you're using gong or similar software to transcribe your sales for sales calls look at the language people are using right depending on where you are based there might be some data privacy concerns with this many of these systems nowadays actually also have some form of chatbot and prompting function so you can literally just write a prompt like give me the last 50 questions people have asks etc etc and then use that to inform your prompting strategy right to use the exact terms people are using and then how you phrase your prompts that is less important i love it that's that's gold malte really appreciate you diving into that i think one of the pieces that gets a lot of marketers stuck is kind of overthinking how specific how unique these things need to be and i think that insight there is super valuable

23:48seven day weekend is brought to you by the hub swap podcast network the audio destination for business professionals self-proclaimed anti-workaholic laura anderson brings together this podcast and it's something that you really should check out it's a podcast where people are going to learn the ins and outs of business finance side hustle and more in this podcast laura brings to life some of the brightest guests who have started their own business from the comfort of their own home to create businesses that generally anywhere from 10 to 20 to 30k a month. This is a podcast that anybody who doesn't want to get caught up in the corporate race, the rat race, if you will, should definitely check out.

24:26If you're maybe thinking about how you can use a side hustle to escape at your nine to five, this podcast might be for you. Laura brings in guests and also delivers authentic and real conversations into your earphones based off of lived experience. If you want to live your own seven day weekend, then this is a podcast that you won't want to miss. Listen to the seven day weekend, wherever you get your podcasts.

24:54If you were to say there's a mistake that a lot of marketers are making when they approach this stuff today, what would you say? The number one thing is that you see that you want to state a message to all marketers. Lots of them listen. You need to stop doing X. This is not the right thing to do. Is there any of those that you see common amongst marketers that they need to step away from? Yes. Yes. Looking at very small amounts of data over very short timeframes and trying to interpret it. So one example, you track five prompts. Right. And today you have visibility 100. Tomorrow you have 80. Right.

25:30This is not something to worry about. Right. Because llms are not predictable right there is some randomness in their answers so there will always be some fluctuation and you can think about flipping a coin if you have a fair coin it's 50 50 heads or tails yeah so you flip it once let's say it's heads you actually know nothing about the coin right it could still be an unfair coin that only has a one percent chance to give you heads so you have to flip it a couple of times if you flip it 10 times you have five times heads five times tails you are still not sure right it could still be 30 70 in reality so you have to flip it more and more and more there's a lot of math around this and of course it's different a prompt is not a fair coin right because most brands don't have a visibility of exactly 50 you might have 90 visibility then 10 coin flips or 10 times running your prompt is actually enough to tell you you you have probably 70 to 99 % visibility, like putting it in a range.

26:36But if you have 5 % chance to appear for a prompt, you might track it 20 times that never appear. And then the next week you track it and you appear and then you jump from 0 % to 100%. So I always urge people to look at a set of prompts for the same topic or for the same persona or the same intent. and then look at that over the course of at least a week. And then you can compare week over week what is happening. But even then, you have some fluctuations. And I know agencies that always get a phone call every Monday from their client. Why did we drop 2 % visibility? We did some work. It should have increased.

27:19But then also on the other hand, a 2 % increase this week might not be something significant. and thinking a little bit about the significance is very important. And then the extreme case of this is looking at one single prompt. You would have to run a prompt six months to actually get statistically significant data to understand if a few percentages of change is really a change or if it's just like random fluctuation. And over the course of these many months, there will be a model update, there will be content updates there will be so many things for something changes so do not run around saying i'm checking this one prompt and i have 53 visibility you can say it's around 50 but it could be 40 it could be 60 you cannot know this right so it's more directional versus concrete is that fair to say that like a lot of the yes yeah what do you think is stopping so many marketers from being able to come to that conclusion like i feel like there's a lot of very firm statements being made by a lot of people that this is the way, this is the number, and there's no black and white.

28:30There's no directional caveat associated with it. Like, what's stopping marketers from having some nuance there? I think from SEO, many marketers are used to, you type something in Google, you have the sorted list. You type it again, it's roughly the same. Even if multiple people are doing it, it's roughly the same. So you have a very simple, very clear and concrete answer. Also, the results are sorted. There's no debate. What is number one? What is number two? That kind of like makes you want to have these numbers. That's fair. That makes sense. Also, a lot of internet marketing is driven by performance marketers who love to measure things very, very exactly.

29:09Right. And what I have observed is that SEOs or performance marketers actually often struggle with the concept of AI visibility or AI search visibility because either they just don't trust it at all because it's not exactly measurable or they over rely on running one prompt five times and taking that as the truth right but people who come from brand marketing who are used to work with a panel yeah they actually get it very very quickly because for them this is the best measurement they've ever had because even if you're a large company you do not get great panel data every month you get it maybe once per quarter right and i know even very large companies making billions in revenue yeah who only do their their big panel survey once per year and with llms you can get this every day for a fraction of the cost of a real panel true and you can actually filter a lot more.

30:09You can have a prompt like, I'm a 40-year-old white man in Germany. I don't do sports. Which running shoes should I buy? Or you are the young mother in Norway. You're used to do competitive running and is getting better into running. Which running shoes should I buy? You will get two different brands' recommendations likely for these. And for brand marketers, this is an unbelievably great accurate kind of panel right with an amazing refresh rate for panel data yeah so brand marketers usually get it very quickly and also embrace it very very quickly and they understand that 60 is not 60 written in stone it is 60 for this panel that i'm using right and if i change my questions or i use a different panel provider it might be 70 but it will definitely not be two percent right like right and also the relation versus my competitors is probably going to be the same right but it is not an exact and perfect science it is not a physical experiment we can fully observe it is a thing that has a lot of privacy which is good yeah it would be horrible if we could see everybody's chats all the time but we can do these different approaches can can use these approaches that i managed to make it measurable and the one that has the richest data, the prompt tracking, is only ever directionally, correct?

31:39But that is enough to take action. I have a hypothesis. We live in a time where what bleeds leads. A lot of writers, a lot of creators are pushing out their tracking reports that say, hey, Perplexity loves LinkedIn now. Everybody needs to be on LinkedIn. Hey, ChatGPT loves YouTube now. Everybody needs to be on YouTube. And they're tracking queries. They put out these data points, but they don't speak to whether or not it's specifically for this industry in this category, in this niche. It's just a macro snapshot. My belief is that you have to run your own data and use a tool like to figure out your situation and prompts with your ICP and your space.

32:20And you need to get away from the clickbait and just stay in your lane for a bit before you make these big decisions to kill everything on YouTube, because an article said then medium.com is now number one. What's your assessment? Am I out to lunch? Tell me if I am. Or is that accurate? Like, are we overreacting to some of the stuff where people are saying Reddit no longer matters? Well, it probably doesn't in this niche for luxury goods, but it probably does in software. Like, am I off? What are your thoughts there? You are absolutely right. There are huge differences between industries. So if we look at total sources, TikTok is really not that big.

33:02But if you are marketing beauty products for Gen Z, Gen Alpha, in some LLMs, TikTok is the top five source together with Instagram. So it's super, super relevant. It's one of the most important channels for you. Or another example is B2B. TechRadar and G2 are super important. For sure. But for many consumer brands, they don't matter. Right. Or if you go into local, then Yelp is super important. BBB, the Better Business Bureau, super important. In the UK, there is Yelco UK. I think it's for yellow pages. Super, super relevant. And then if you go into a niche like supermarkets, tattoos, handymen, there are also these niche directories that are super, super popular.

33:52And there are big differences also between the LLMs. and also over time there are huge differences so for example perplexity and youtube right or perplexity and linkedin or perplexity and reddit right for all of these there were times when their usage dropped by 90 percent right and then a few weeks later it it went to a new high like i think with with youtube and ai overviews there was a time when the average ai overview already was citing YouTube more than once. And you would think this is the maximum. And then over the course of a couple of weeks, it doubled. And then the average AI overview was citing YouTube twice.

34:34And on perplexity, this went to extremes of like citing Reddit like eight times in one prompt. But then it also goes down again. Right, right, right. My recommendation is look at over a longer timeframe, over multiple LLMs, look at the sources that are there to stay. It's very likely YouTube. It's very likely Reddit. If you're using the Google LLMs, so Gemini, AI mode, AI overview, it's actually very likely Facebook, surprisingly. Fascinating. If it's something for younger people, there's going to be Instagram in there. There's going to be TikTok in there. If it's B2B, there's going to be LinkedIn in there, TechRadar, maybe G2 and stuff there.

35:16And then work on these and treat them as relevant surfaces. But don't put all the effort on one of them because in the last seven days, it was cited most often. Because these are trends and there are changes. And it's the wrong idea, in my opinion, always, actually, to just obsess over the current trend. Try to understand in general, where do LLMs want to pull information from and be present on all of these services. I love that. So let's say you've pulled the data. You understand your category. You understand your niche. You're seeing some of these sites be sourced. You're starting to see this happen.

35:52What's the next logical step for a marketing leader after their team, an agency, or platform gives them this data? What do they do next? I would look at the sources used by LLMs for these chats that are relevant for me. And I can do the very data-driven approach of saying, let's look at sources where multiple competitors are mentioned and I'm not mentioned. And let's sort them by how often they were retrieved or cited by the LLMs. and that's where I want to be mentioned. That's one way. The other way to think about this is just get a look at a list of domains that are the top sources and think about, am I currently well represented on these?

36:33And can I do anything to be better represented? And if it's Reddit or YouTube or Instagram, it's like social media, UGC, you can either be present on your own or you can work with influencers and creators to have content on there. If it's editorial news websites, you can work with a digital PR agency or you can try to see if they have commercial content teams, which are often separate and might also list you for money. You can see if there are options to buy an advertorial. If it's a lot of corporate websites, check if some of these are maybe not direct competitors, but potential partners. and then say, hey, you have an integration for our system.

37:20Can we write a blog article about this? And can we use these keywords? Many marketers will actually do that. Right. And if it's your competitors, then look at the content that is being cited from your competitor's website and think about, can I create similar content? So this is what I would do with the data. I like it. And there's one thing I would do regardless of the data. And that is come up with my positioning if I don't have, I mean, ideally you have it, but let's be realistic. Many marketing teams don't. And then describe yourself consistently on the internet. So don't say on your own website, I'm a podcast agency.

38:00And then in your press releases, say I'm a video interview and clipping service company. And then have on the founders LinkedIn that you are offering digital reach for founders. right pick one messaging and use it consistently everywhere right especially for small brands this can work within 24 hours right to change the llm answer from i don't have no idea what this company is right to you're talking about this company because llms are looking for consensus so you need to have the same information on your website on your g2 on your google maps on your Yelp, wherever you are present, whatever is relevant for you.

38:44So this is usually the first thing I recommend to almost everyone. I mean, if you are McDonald's or Coca-Cola, I might say the Spence, you don't need to do this. The LLMs know who you are, even without grounding. But if you are not, if you are a smaller brand, especially a new brand, describe yourself consistently everywhere and use these, I mean, not own surfaces, but managed surfaces where you can add it on another website how you are perceived right like your google maps like your yelp like your g2 and describe yourself everywhere to help llms understand you i love it that's awesome is there anything that you have seen or recommend people use to do this at scale like i'm thinking of some brands that might be bigger and they're like that's a lot of work we have to research all these places where we're talking about and mentioned like is there anything that they can use to do this at scale there's no perfect version of course if you're in the local space there's software like Yext, et cetera, that can help you.

39:39Right. But many of these platforms require you to log in and do the work. Right. What you can do is you can use a software like PKI to track a couple of prompts and then look at the top sources. And that already gives you the target list. Cool. And then you can put that into Cloud, for example, and just ask it, like check if I am described like this there. and if not create a linear ticket or Jira ticket or Trello ticket or whatever for me to change the text there. Do I remember this correctly? Peak now has an MCP connector directly into Cloud and you can use the product through Cloud. Is that correct?

40:18How does that work? Yes, so we have an MCP and an API so you can use us anywhere with this MCP. We do not limit it. And with Cloud, we are also an official plugin connector approved by Anthropic. And in a couple of days, we will probably also be that in Codex. Nice. Very cool. So what's the thinking around those platforms being the first interface for a marketer and a user? Do you folks have a perspective on where software is going around giving this type of connector, giving this dashboard? What's the thinking there around how marketers want to interact with PKI and what you're building? Yeah, I think for most software, very advanced users that are AI-pilled now prefer to have it in their cloud.

41:05I have team members who do not lock into Notion or Linear anymore. They do it all via cloud. And I also rarely use these tools now directly. I use them as a system of record. But all my interactions are in my cloud code interface. Is there an example of how you're using cloud in how you would view the best use case for Cloud and Peak at the same time? Like how are people using the connector? Oh, so many different things. One example is when people register to Peak AI, we ask them this question that I mentioned earlier, how did you find us? If they say LLM, we ask what prompt? Okay. This is written in post hoc.

41:46I have a routine in my Cloud code that once per week, it connects to the post hoc MCP and downloads the data for the last seven days. It filters out the ones that are duplicates and the ones where somebody just types something random. It then uses the Peak AI MCP to write these prompts in a project I have in Peak AI called Peak AI Real User Prompts. So of course, there's bias in the data because these are the prompts where people found us. So we are the most visible brand in there. It's boring. But still, it's interesting for me to see that over time, I now have this bigger and bigger and bigger project with real user prompts that I'm tracking.

42:26That's cool. So that is one way how I'm using it. Another thing would be at Peak AI, we allow users to select the brands that they want to track. So you're not limited to a few brands we identified. You can add hundreds of brands to your project. And I'm a lazy person and I create a lot of demo projects. Right, right. So sometimes, and we actually now turned this into a skill that also lives inside Peak AI. but in the past I just told my Claude go through all the brands that exist remove the duplicates add the most common typos as aliases and set the color of each brand to the official brand color and also add all their addition their international domains smart and of course I could do this manually but it's very boring yeah and then I can also tell it and now go through the last 1000 chats and look for any brands we might have missed cool and this way i can get a project set up with all of the brands all of the aliases which is especially relevant when i do it in a language i don't speak like recently i set up a demo project since we started tracking also quen and deep seek in china right i wanted a demo project in chinese about chinese electric vehicles right I do not speak Mandarin at all.

43:45I cannot read it, can speak it, not a single word. So I also have big difficulty understanding what is a brand and what is maybe a sub-brand of another company because you sometimes, especially in the car space, there are so many joint ventures and one brand owns another. And it was very, very nice to just let Claude figure all of that out for me and then immediately write it all into Peak, like pulling the data from Peak, doing the work, writing it back into Peak. That's cool. So apart from creating your projects and inviting your co-workers, you can do anything in the MCP. Or we also have now a chat inside Peak where you basically can also do it.

44:27And we have users that have very much embraced the MCP. We have some agencies where 10 plus people are using the MCP every single week. And I believe actually this will increase over time. I believe like independently of peak AI, there's a trend that people use software in MCPs and do not log into user faces anymore. Of course, the user interface is still amazing for a new user to find out, okay, what actually is going on. It is also good to have some default visualizations. And right now it's an edge case. Somebody clicks on chat and said, I'm a social media manager. Create me a report what happened in my PKI project the last 30 days.

45:11You can do this and you get a custom dashboard, but it's rare that people do it. I think in two years, this might be just as common as looking at our dashboards that we created for you. Right. I love it. That's awesome. I agree. I think in the future, we're going to get way less emails from executives saying, hey, what's going on with this? And instead they will ask an LLM that's actually connected directly to the data. They'll get their answer. They'll maybe take a screenshot, throw it in their presentation and share it during their town halls, but they don't even interact with the team to get the data.

45:44They're gonna trust the LLM has the source and be able to provide it. So it's cool that you folks are being so open with the data and even the customizations within Peak has always been one of the coolest features that I've seen. Question for you. So when you think of your role as a marketer, when you think of AI, Talk me through like your own practical use cases. Like how are you using AI to do more in today's market? Like, are you leveraging it to lead? Are you leveraging it to execute just for insights? What is your workflows look like today to execute and be like a marketer? So I have to admit, I do very little operational marketing work myself.

46:23But one thing, for example, that Claude is doing for me. Every morning it checks my calendar and the Notion page where I list all of the talks and podcasts I'm doing. And then it checks if I have a podcast or webinar in my calendar. Is there a blocker that starts half an hour early and goes half an hour later? And did I book the content room in the Peak Air office at this time? Which is the room that I'm in right now. And if that room is not available, it tries to book one specific content room that at least has a nice looking wall that I can use as a background, that I don't have to use a wide wall as my background.

47:02And only if that is also not available, does it tell me, hey Malte, for this webinar, you need to figure something out. So this is, for example, something that makes my life very, very easy. That's awesome. And on this level, I have many, many things. Another thing I'm doing also with PKI, every morning, Claude uses Firecrawl to look at the Google News homepage and then these topical Google News homepage like Business Sports. It extracts the top stories. It creates prompts for these top stories. It goes to one specific project I have, deletes all of the prompts from yesterday, adds these prompts.

47:41so basically every single day i have the current news topics of the day as prompts and of course visibility is not something you can look at in that project because the prompts change every day but you can look at the sources and i can see over time and for certain topics how often is reuters being cited how often is cnn being cited so it's kind of a citation index for news and this has been running for a couple of months now i can run nice nice analysis with it right and these are the ways how i am using that's awesome claude like doing things where in the past i might have told a junior employee or to run it i mean i never had personal assistant but maybe i would have told a personal assistant and to do these things yeah these are things that claude is doing for me that's awesome i love it when you look ahead there's going to be more change in the marketing industry, there's going to be more shifts.

48:35There's going to be things that happen that we can't even expect or predict. Do you have suggestions for folks who might be listening to this and they're trying to get started with their career? They might have had one or two years working in-house. They want to have a successful career, but they don't know what's going to happen like none of us do. What's your advice to early marketers trying to make it in this wild, wild west? work in a fast-paced environment where you are exposed to a lot of different things so that is either a growing company or an agency that has different kinds of clients absorb as much knowledge as possible and try to be ai native in anything like your first instinct should be trying to solve it with ai before you even think about talking to a human or hiring a person or asking someone in another team if you have notion you can chat with notion you can answer so many questions about your company with that right if you are using linear to manage tasks you can now chat with linear do not be afraid to use ai always question what is this doing and be aware what it can do and what it cannot do and be aware of hallucinations do not blindly present something that that an llm created for you, but embrace it as much as possible.

49:57So maybe another example. Recently, we changed the layout of our business company presentation. And I do a lot of presentations. And I threw all of my presentations together. And there were something like 1 ,200 slides with a lot of duplicates. I told Claude to delete the duplicates. And then I told it to update them all to the latest design. Cool. And there were two or three slides that it messed up. Wow. But everything else, it did better than I could have done. It's unreal. And I think this is how someone who is very new in their career should right now think about anything. Yeah. Because you can really do things that previously you couldn't do.

50:40Right. Like, let's say your boss tells you, hey, we have this presentation with the client. Can you visualize the workflow that we discussed? You should not start PowerPoint and then moving some boxes. Right. You should speak to like a whisper flow or something to have like a long detailed description of your thoughts. Put that into Claude and tell it in Figma, in PowerPoint, whatever, create 10 different versions of this. Make them as different as possible. And then you just let it work for 20 minutes and you do something else. And then you come back and you will have 10 different versions. And these will be super creative.

51:18I mean, maybe I'm just not a very creative person, but at least for me there's always stuff in there that i didn't think about like oh yeah i could do it like this like this yeah and then you probably delete eight of them and then you let it iterate on the two that you like right and in the past this was only possible for someone like steve jobs like he he could tell 10 different designers to make this and then pick the nicest version yeah but no sane person would do this also you would feel horrible to do this to your designers right Like design this whole deck. And then from nine people, you throw away the work.

51:54But now you can. And it only costs a couple of euros. So it's included in a 200 euros per month subscription. So don't be afraid to burn tokens is the best advice. And no one's feelings get hurt. That's the other part. It's like you're talking to an LLM and you can say, I don't like that. That's ugly. And they'll just fix it for you. There's no issues. Yeah, 100%. Malte, if folks want to learn more about you, more about Peak, where should they go? How can they learn more? Follow me on LinkedIn. I think that's the best way to learn from me. And if you want to learn about Peak AI, go to peak.ai, P-E-E-C.ai.

52:27We have a free trial. We have a button to talk to our sales team. We are very open and transparent in what we do. Awesome. I'll include some links in the show notes. Malte, thank you so much for coming on. I appreciate you sharing with the teams. Thank you again for making the time. Really appreciate it. Thanks for having me. I've been a big fan of you for many years. Thank you.