Episode Transcript
[00:00:00] Speaker A: Hello, I'm Jeremy Rivera, your unscripted SEO podcast host. I'm here with Malte Lanver who's going to introduce a little bit of his history first and tell us why we should trust him as a voice in SEO.
[00:00:15] Speaker B: Hi, I'm Malte and I'm surprised by the question, but I will try to answer it.
I have been doing SEO for the last 20 plus years. I started as a teenager randomly seeing people landing on my website.
Went to university university to study computer science. Started a PhD, never finished it. Did a lot of stuff in social media analytics, web scraping, social network analysis.
Literally wrote my bachelor thesis about page rank and applying it to find good and bad scientists or successful and unsuccessful scientists to be more precise.
Co founded an SEO agency many years ago.
Worked at Such metrics, a P2P software for SEO and content marketing.
I was leading the product team there at some point.
Then I did in house SEO for five years for the largest price comparison website in Europe, potentially the largest one in the world. There's some debate who can actually claim that and now I'm at pki. It's a software to measure and increase your visibility in LLM based answer engines and I'm running primarily the product team here, but also our research team.
And why should you trust me? Because I've been an SEO practitioner for many, many years and I have always focused on things that actually work and don't like to be bullshitted by fluffy statistics, but actually see experiments and test what's working, what's not working.
[00:01:48] Speaker A: I'm curious if you, you know, having done your thesis on PageRank, do you, do we know to what extent that algorithm is still used behind the scenes in and where they might have folded it in and how you see its echo or its effects to this day.
[00:02:14] Speaker B: So we cannot know if the exact algorithm from this random surfer paper that Google published a long time ago is being used, but some variation of it is definitely used still today. It is such a useful algorithm you can apply it to virtually anything. Like I mentioned, scientists.
You can also look at a graph of animals and then map which animal eats another animal. And then you can use PageRank to calculate which animal would starve first, like what would be extinct first and which other animals starve. There are so many great applications of PageRank.
Probably Google made some changes, like probably there are some websites that are not counted at all anymore. There are potentially some links, like links that are in white color on white background somewhere in the footer. These are probably not counted anymore as strongly as in the past. But overall, the core principle of PageRank, which is creating a graph of the whole Internet and using links between websites as the edges on your graph and then calculating which nodes so which URLs or pages are the most prominent one, most important one that is surely still being used by Google and probably also by every other web search engine in some form or another.
[00:03:37] Speaker A: That's the reasonable surfer paper, right. Of evaluating placement.
But I had an interesting challenge come up in my last interview where the guest pointed out that for, well, 50% of my traffic I just checked SEO Arcade and compared GA for data and a whole bunch of it fell into direct. But Microsoft clarity through its API MCP gave me numbers and showed me that half of those numbers and almost all of my direct was actually bot traffic of bots crawling and scraping and digesting and eating up the page from different various different sources.
So does reasonable surfer apply to robots?
[00:04:29] Speaker B: I mean, in theory, if a robot acted completely random or reasonable for a task, it could. But of course the reality is that a lot of crawling on the Internet is done for very specific purposes and is just not the same as using pagering. Like for example, if you have a crawler that checks every five minutes if your competitor changed the price of their main 10 products, you're always just going to refresh those URLs, right? You're not going to care about links, you're not going to follow any links.
So I would say no.
If we ask the question which pages are crawled the most on the Internet by bots, there's probably going to be some correlation where pages that have very high page rank are also typically crawled a lot. But for many parts of it it's just not working.
Many crawlers now log into websites, for example.
Of course that is not part of PageRank, right? Because these are not crawled for the original patron calculation.
So I would say it is more an approximation of user behavior.
And of course if we assume that bots just mimic human behavior, you could still argue that patron is also for bots. But I think artificial crawling by bots has very different characteristics from human crawling.
[00:05:57] Speaker A: An extension of that I.
I probably will have to correct this in the show notes so if you read this later it'll have the correct example. I believe it was Time magazine started serving up markdown site wide.
Do you think they're jumping the gun or do you think they're seeing the writing on the wall that with.
I think we passed the threshold again linking the article here.
You know, we.
More than half of the traffic on the Internet is now bought.
So Google wants to hold on, hold the line and have, you know, not have to double dip because that's their resources. They don't want to have to process markdown and HTML, they want both at once because they've gotten really good at it. But LLMs would happily take down a, take up a markdown file with the same information. That's what, 45% faster for them to process something in that range.
So our technical SEO is going to have to grapple with best processes. We're also styling in markdown. Does that mean just an extra step of hey, I'll make a PDF downloadable version and a markdown version of this?
And do markdown files even have a canonical system?
[00:07:31] Speaker B: You could probably put a canonical in the header. Otherwise the markdown is just a text document, right? There's no such thing as an HTML header. You can only use the HTTP header already server side, I think creating a markdown copy of my website. So for example, every page that's available as example.comtest and then I create example.comtest MD.
I would not do that as an SEO. I don't want the same content on two different URLs. It wastes crawl resources. If humans land on the MD version, there are no links to click. There's nothing for them to do. It's a horrible experience.
What can make sense is that my server makes a decision.
When a human comes, I serve them the HTML version and when a LLM crawler comes, I serve them the markdown file under the same URL. Like it's not a redirect, it's not a separate URL. It's a form of. You could call this cloaking, like serving different content based on the user agent that is requesting. And I would not do that for Google because cloaking and it's really different from what humans are seeing, right? So I would be very careful with it. But for a ChatGPT crawler, I mean, why not?
They can process it, it can work.
And it's especially if I have a JavaScript heavy website with a lot of, maybe even client side render JavaScript. I can make it much easier for the LLMs.
So I can definitely understand websites testing this for LLMs. There's one thing about the story from Times, they also inject ads specifically for the LLM and that is again going very much in direction of cloaking because this cannot be in the interest of ChatGPT, right? That you have one version for the human and then when the LLM comes, there's a different version and it says, oh by the way, this is the best insurance company or like whatever the article is about.
So I have not looked at examples of what Time is doing, but if they are doing what I think they are doing, it's definitely a huge risk. And at some point there will be a product manager at anthropic@OpenAI who will say, wait a minute, this is not okay that they serve us an ad that humans are not seeing. We want this out. Or the ad has to be marked like ad start, then comes the ad, ad end, something like this.
So the LLMs can ignore it. But I think it's not a top priority for the LLMs. And for now it's a great way for Time magazine to monetize the traffic that they have because this is as far as I know, the only way to monetize bot traffic right now to having advertisers pay for ads in your bot retrieved content.
[00:10:32] Speaker A: It is a brave new world and certainly cloaking is a big no no from the early days, but we have practiced some form of it or another of giving googlebot what it wants.
A fascinating, I'll take it in another direction that is it me or are these LLMs just bloody reckless when it comes to the protocols and decisions and just the careful nature and the battle, the back and forth battle that we as SEOs have experienced against to gain any ground at any time and get anything that doesn't immediately get swatted down.
Do you think that they're just relying on the resiliency of the distributed LLM intelligence, you know, semi intelligence based system to suss out and just kind of inherently do it? Or are they just rushing beta so quick to market that you know what, it's broken, but we got the users and just keep going. We're not going waste time trying to prevent people from poisoning or whatever. That's just going to happen and we'll deal with it tomorrow.
[00:11:58] Speaker B: So reckless is such a negative word. I would frame it a little bit more positive.
The LLMs are living this Silicon Valley mantra of move fast and break things.
And they are definitely moving very fast, but they also definitely breaking things.
And I mean the most public example I think was Google in the AI overviews recommending to put glue on your pizza.
Or the more funny one being horses have five legs and many other animals have the wrong number of legs. But of course there are. I mean with LLMs it's not like this predictable algorithm, right?
It's unpredictable what it does and Unless you harness it super super strictly. You can always have these weird outcomes and it's more like human than an algorithm in that regard. In the sense that if you train a thousand humans to support a job or to perform a job, one of them will do it wrong, right? There will be one person who works in customer support but writes a rude email when they shouldn't. And LLMs are the same. You will have very hard time to get them to zero percent errors. Zero percent weird answers.
So I can understand why the LLM labs move so quickly. But of course that I also understand the perspective why you would say it's reckless. Because yeah, there are many things that can go wrong and there are many things how people can influence LLMs.
So for example, a few days ago ChatGPT switched all users to GPT 5.6 versus 5.5, even the non paying users. And GPT 5.6 is doing more fan out queries and it's also doing more fan out queries where they perform a site search. So site example.com and then some keyword when they look for content from a specific domain. And this contains hallucinated domains. And there are some prompts I'm tracking where it's in the range of 2 or 3% of the prompts have a hallucinated domain.
In these cases the domain is just parked. But if I was an evil person, I would now register this domain or buy it and put up some completely negative content about the brand that supposedly owns it.
And I could very much influence a significant percentage of chats where people ask questions about this brand. I could show a wrong pricing information for example, or another example where LLMs are maybe a bit reckless.
I can buy advertorials on many media websites, right, where I basically dictate what kind of article they write. And these are marked as advertorials. So as a human I see, ah, this is an advertorial, I ignore it.
LLMs use these as sources and sometimes they say based on an advertorial on, I don't know, New York Times or whatever this and this was set but I monitoring a set of prompt in the insurance space where regularly 2% of the sources used by the LLMs are these advertorials. So you can basically, you can take some money, you can buy these advertorials and you have 2% of the grounding influence immediately on your site.
And that is of course questionable if LLMs should use clear advertorials for their web grounding.
[00:15:26] Speaker A: That kind of leads me to Kind of the fundamental question and you're the perfect person to answer it because you're on the side of giving the data, trying to give data for the visibility side of LLMs.
Of course I'm on the other side and I need action items. I talked to Matt Brooks of SEO Tarek and he says, hey, treat these platforms as uneducated customer support representatives and do your best to publish as much education material for your wayward customer support person as part of your, you know, natural SEO campaign, as part of your content strategy.
What are the. Is is that accurate? And then are what other are tactical things should people be looking to do once they see these query? Okay, great. There's 75 query fan outs for podcast production, you know that I would be looking at or podcast producer or something like that for my agency.
What do I do with that?
How is it, how is. How do we take that data point and turn it into an actionable point in our report to the client are handed as a action item for the SEO manager or the content team to actually do something with?
[00:16:55] Speaker B: Yeah, great question and a topic we are actually also thinking about a lot and have been thinking about for the last couple of months.
I think the first part of the question, thinking about the LLM as like customer support agents that just need a lot of training. I think it's a good metaphor. I like it. I will steal this. It's very good when we talk about branded prompts. So question about my brand and that's actually true, you probably don't need to publish as much as humanly possible because at some point you will publish low quality, it will be duplicate. But the LLMs are looking for consensus.
So if you only talk about your pricing on your pricing website and then you change your pricing and then there are five Reddit threads and two reviews on random blogs that still talk about your old pricing. ChatGPT will answer with your old pricing if a user asks about it.
So it's actually important that you also think about do I have a profile on G2? Or if I'm a different kind of business, do I have a Yelp profile? And then if you change something about yourself, publish it there. If you have a key message, put it maybe in the footer of your website, put it on all your social profiles that you have for your brand or for your company, put it in the footer of your press releases. Like make sure that your core message is out there on the Internet and the LLMs can quickly find consensus. Or if you publish a feature you can do the blog post, you can do the product details page on your website, you can write in your help center, you can write in your product docs, you can have a couple of different content assets on your website that all tell the same story from a different perspective for a different use case. And the LLMs will likely look at multiple of them, then come to consensus that the information from the first one was already correct and then say that. Now that helps you with branded prompts where people ask about your brand. But when somebody asks about what is the best software to record a podcast, of course the LLM doesn't start by reading the content from these softwares, right? They start by having fan out queries like 10 Best Podcast Reporting Softwares 2026 or Best Podcast Recording software for a solo freelancer or for someone who's traveling or for an enterprise company.
And there it can be your own content that is being found, but it can also be third party content.
And then you need to see can I be part of these third party articles that are being cited?
And then you mentioned the fan out queries and how I think about it is this.
Let's say I found a set of prompts that I want my brand to be known for and let's say it's Best podcast producer, best podcast agency.
With podcast agency, should I use these kinds of prompts?
I would run them multiple times and I would look at multiple things. One is I would look at the brands that are currently winning to get inspiration from them. The second is I would look at the sources both on the URL and on the domain level and see for these existing URLs. Can I get my brand mentioned there if I'm not mentioned yet?
And the trick is often to look for the ones that have multiple of your competitors already mentioned because then it's often reasonable to contact and say, hey, can I also be added if it's an interview with the CEO of your main competitor, there's no way that you will ever be mentioned on that page, right? So you can mainly forget about that. And the second thing is to look at the domains and think can I create new content about my brand there? So if it's, if it's. A news and media publisher, can I work with digital PR with press releases or can I just contact their commercial content team and just pay for having an article there?
Can I buy an advertorial? Can I do something with affiliate marketing? There are many different ways. And if it's social media website, I might just become part of the community.
Maybe somebody wants to do an AMA with me, something like that.
And then the last step I do is I look at the fanout queries and maybe for those listeners who are not familiar, when I ask ChatGPT a question like which podcast recording software should I use, it will not just search in various search engines for what is the best podcast software, but it will also take my context and maybe add because I'm based in Europe it might say that is GDPR compliant. Or because my LLM knows I work at pki, it will say podcast recording software for a startup.
And these are the so called fan out queries and it doesn't really make sense to take all of them and then optimize for each individual one. But what I like to do do is I like to look at common concepts in these queries, especially terms that the LLMs are adding that were not part of the prompt.
So imagine I search for best podcast recording software or that's what I'm prompting and then half the fan out queries contain the word review.
Then if I'm a software recording a podcast recording software reviews is obviously an interesting topic for me. And there was a time when ChatGPT always added Reddit to the fanout queries. So Reddit content was everywhere.
A few days ago ChatGPT started adding the term official to a lot of fan out queries. So right now I recommend everybody to put the word official in the footer of their website just that they have this word out there a little bit more often and get found for it.
Yearly numbers are very very common. So just adding in 2026 or just in brackets 2026 to the titles of your articles can actually work.
And in general LLMs are looking for fresh content.
So refreshing content, updating it and then also making sure there's a machine readable last updated date that that is updated and changed can can really work and can increase the likelihood that that specific page is being retrieved and then of course that leading to your brand being
[00:22:57] Speaker A: mentioned anything on you know, because you can either republish it with a new date or you can keep the original publish date and then have the updated and then serve both numbers. Any any information have you seen a preference or bias or both work equally.
[00:23:18] Speaker B: So there are two two people sitting on my shoulder and one is saying be transparent, be nice to your users, have the publish date and the update date and the other one is the hardcore pure SEO who says Malte, you know exactly what will happen in the go snippet. They will still use the old publication date, delete it, override it. In reality, what I would do is I would want to display both, but maybe have the original publishing date in JavaScript or put a line break in the middle of the date and then with CSS remove the line break so it's not like purely visible. Like do something to mess up the publication date, that the updated date is the one that every crawler sees. It's a bit of a tricky topic where the maximum short term SEO impact is different from the long term trust and brand impact. Unfortunately,
[00:24:22] Speaker A: with the Query Fan out, there's a famous line from Lord of the Rings where meat's back on the menu, boys.
And in SEO there was a fad, there was a moment of focus of cannibalism. If there's two articles for the same keyword, this cannot be. You can never write for the same keyword twice. Ever.
With Query Fanout, it seems like maybe there's an argument for, you know, going native and Papua New guinea and getting a little bit cannibalistic, a little bit repetitious.
I personally never fully bought into the anti cannibalism arguments as hard as they wanted to go. So where's your position on that?
[00:25:13] Speaker B: So as an SEO, I always was very, very strongly in the anti cannibalism camp, but I have to say I've often worked on websites with a million plus URLs, sometimes hundreds of millions of URLs, so their cannibalization can become really, really big issue. With 10,000 landing pages essentially for the same topic, I think now it absolutely makes sense to have multiple pages on the same topic to make sure LLMs find this consensus that I mentioned earlier. But I would still try to have a slightly different intent. Like you can have one article about the 10 best health insurance providers and then you can have another one about the health insurance providers that won the best awards and another one with the best health insurance providers for every single state. Or the best health insurance providers for people earning above or below a certain amount of money. Like things that are overlapping but there's like some distinction at least already on the title level.
I'm still not a fan of having three pages with the same title and almost same content.
I would want something unique everywhere
[00:26:33] Speaker A: for all these query fan out things. You know, one of the temptations is just to push out AI slop everywhere for it. What are some of the guardrails or best practices that you're recommending that you use yourself to hit the note of volume and scale that is table stakes now when it comes to content marketing, but also not, you know, put out slop or you know.
[00:27:04] Speaker B: Yeah, I have actually three different answers for this. Okay, if you only care about short term success and you only care about AI visibility and you don't care about losing your Google rankings, push out a lot of AI slop. Like at least a thousand articles per day. Don't be shy. You can publish 200 listicles about the best content management systems.
It works short term, it works very, very well, unfortunately, especially if you're older, well established domain. I hate that it works, but it does if you are super, super careful. Just don't do any AI content because it's all bad. It can all be detected.
But for most people I think there's a middle ground.
And even within this middle ground there are two options.
I would not publish AI content for what I consider editorial text. So if the main content piece is a written article in full sentences, I would not use AI to produce it.
I have like, basically if you can create it with a prompt, why would ChatGPT or OpenAI or Google crawl index and rank it and use it right? They could just use that prompt on their own.
If you add something unique, like your own unique data. So for example, let's say you have structured data on every single college basketball game in the United states in the 1980s and 1990s.
Like this minute this happened, that minute that happened. And there's no article about these games.
Of course you can use AI to now produce an article about each single game. Like talking about was it close game, was it not? Like you have like, because you have a unique input for the data creation.
Also if you have a, you are an online shop and you have product detail pages and for each product you have a handful of user reviews. Of course you can use AI to write a summary of the user reviews. Like people say this fridge uses very little energy and is quiet. It's perfect for the home office. Or this fridge is an absolute monster. It's super loud, but it's also very reliable. It's the best fridge ever for the garage. Like, of course you can use this level of summarization with AI.
And if you want to publish a lot of AI content in a previous job I might have also done that for a few hundred thousand URLs.
I recommend to look at the following things. There's something called perplexity, and I don't mean the search engine perplexity.
You can measure the perplexity of a text. It is basically Based on the previous words, how predictable is the next word?
If you have a set of human written texts, calculate the perplexity, then use AI. And if the perplexity is a lot higher, you probably need to adapt your prompt and humanize the whole thing a little bit.
The next thing you can look at is something called compression rate. It is something that's actually used to combat email spam because email spammers started rewriting their emails already many years ago and often by adding filler words. So compression ratio more or less tells you how many words can you remove from this text without losing information.
Very, very rough explanation. I would also do this for your human written text and for your AI written text. And you will often see the AI written text, if you use a very generic prompt, can have a lot of this. Especially if you say write 600 words about this topic and then you provide no data, then you often have very high compression rate.
And then you have two measurements of similarity of text. One is Jacquard and the other is Cosine. And if you have 100 human written text about, let's say it can be product descriptions about the same kind of product, it can be news articles about similar topics, it can be biographies of 100 actors, whatever. Like there should be some, like the topic should be similar and the content format should be similar. And then you can again, for these human written texts, look at the Jacquard and Cosine similarity and then do it with AI. And you will also often see that the AI text has more similarities.
And Jacquard is really good, or Jacquard similarity is really good at identifying text where it's basically the same sentence in every text, just you switch a couple of words. Like Berlin is the capital of Germany, it had this many inhabitants, this is the main industry, Washington D.C.
and you do this for every, like it's always the same, basically. And cosine similarity on the other hand, is very good at finding sentences or paragraphs or whole text texts that every word is unique, but the information contained in them is actually the same.
And if you use these four measures, so perplexity, compression rate, cosine similarity, Jacquard similarity, you will often find that there are many, many very easy to detect footprints in AI written content.
And this can, by the way, you can tell like Claude, to implement this for you, it can write you a Python script that checks for all of these things. It's not rocket science. You don't really need to understand what's going on below. It but then you have a very, very quick AI content detection that you can just use in house on your own content.
What also works very well is if you don't tell AI, write me 600 words on this topic. But rather you first have a briefing and then you have a second prompt that checks. Does this briefing make sense?
And then you write a briefing for each paragraph and you have with one prompt, write about one paragraph. And then you have another prompt that is a fact checker, like check if all of these facts are true. And then you have some guidance, like for example, check for empty paragraphs, repetitive concepts. You go over it again and again and you can of course chain all of this together in a somewhat automated fashion.
People then call that an agentic workflow, even though it's actually just a workflow. With LLMs, there's not a lot agentic going on there.
And then you often end up with five to seven and a half euros of just token costs to create the content.
But then it's usually really, really, really good content.
[00:33:53] Speaker A: Right.
[00:33:54] Speaker B: It might still be detectable as AI content, but if you write these prompts well, if you like, think about how you would do briefings for a new author who is very smart but doesn't know anything about your business and will try to bullshit you unless you give them very explicit instructions. And then also have still someone double check it. If you're good at with that mindset, you can do with reasonable risk, produce AI content. And I know very successful SEOs who do this at scale and it works well.
I know very successful SEOs who say they would never do this. Everything has to be handwritten.
I'm somewhat in between.
What I do know is if you just publish a lot of AI content without quality measures, you often lose your Google rankings and then regularly also your chatgpt citations as a consequence of that.
[00:34:47] Speaker A: Yeah, previous guests said that we were thinking about. We put, I think it was Nick Eubanks from DigiStore24.
He said that we were injecting AI into the wrong step and that there were so many tools out there. They're like, oh, hey, you have AI write your brief. No, no, no, you write the brief. Humans write the brief, bring the data, bring the quotes. Like, that's the answer to the old challenge of a B2B, you know, oh, I make agricultural equipment and, and oxygen monitoring not sexy. But if you know your stuff and you've got the quotes from the CEO about why it matters, if you've got the stats from the regulatory side that businesses have to check those are the hard facts. And then you, you work in and have a source from a transcript.
I did that with Permacast down in Florida. Interviewed the CEO and the installation manager to get the words from their mouth about what are the challenges, what do they ask when they're on the ground? What keeps this from being installed? You know, how can we clear the friction? All those came out. It wasn't all directly usable. You have to sort and filter your human transcripts because we do a lot of nod and wink, we do a lot of alighting and that doesn't process well out of transcripts, but raw sourcing of like input that goes into the brief and then you're going to, you know, come out with an article, then have a human add some flavor, you know, human spice that's going to be a million times better than, oh, I did keyword research through a tool, fed it to this AI prompt gen deliverable engine and they gave me a spec for the article and then I ran it and then just had an editor move four sentences around. That's not, that's not human editing at all. That's just rubber stamping.
[00:36:52] Speaker B: Yeah. If you can get these like expert quotes, especially internally, and then the whole transcript of them talking about that topic, that is of course the best input to create a brief. And then also an article. Yeah.
[00:37:05] Speaker A: Kind of coming around the corner. Tell me what's some interesting things. I think I saw that you had added an MCP to your own tool.
What are you, what mindset do you suggest that people adopt in the digital marketing space now that it's almost like you can't have a SaaS unless you add an API and an MCP?
Where's your headspace at for the future of your business model as a SaaS? Are you just, you know, is the internal of the SaaS now just redundant to what you can provide through the API or is it a hybridized approach and you provide both sides of the SaaS experience?
Where's your mindset when it comes to the product management?
[00:38:02] Speaker B: So I think for a couple of years we will still have the majority of users using user interfaces of SaaS software.
But we like on our, in our case, we already have customers who primarily use us via the mcp.
And also a lot of tools that I'm using, I'm using them via the mcp. Like it's more likely that I tell Claude to create a linear ticket or tell me the status of a linear ticket than me actually logging into linear and I'm also more likely to ask Claude to search through notion than me opening notion in my browser. And I think all systems that have this character of being a system of record, like a cim, task management, knowledge management, I think these are becoming basically databases for an mcp.
Because why would I log in like there's nothing there for me that I need. There are also software that is showing me unique data, like pkai for example.
Of course I can pull that via an API, via MCP into my own systems.
But I think in our case, for example, we have a lot of proprietary data.
Don't want to go into all details there about our product strategy and what we see for the future. But we know that we have a lot of parts that you, you cannot wipe code them. It doesn't matter how many tokens you give fable, it just will not have the data that is needed. It could write the code and it could start, excuse me, maybe start gathering the data from that day.
But we believe that we have a lot uniqueness there that cannot easily be replaced.
And I think in a few years the MCP or a CLI tool or the API might be our primary user interface.
And another thing we're experimenting with is putting a agent chat directly in the software and then users can chat with it. And for example, you don't need to use our dashboards, you can just say, I'm a social media manager, I only care about the social media aspect of AI search and I need to do my monthly reporting. Please create me a dashboard.
And then it will just create you a couple of charts, it will pull the data, it will give explanation for it and it's completely customized. And of course you could also do that in an external chat via the mcp, like in Claude and Codex, wherever. But we believe there's also these agentic interfaces are becoming part of the user experience.
But for the uneducated user, I think there are still many people who want to log in and they want to understand what data is actually there to form a mental model of the existing data in their mind.
And I think what many SaaS companies need to realize, and I just use the term uneducated user, an uneducated user is also someone who has to log into 30 different tools every day. Of course that person is not an expert in every tool. Right? They probably want two or three things.
And with just an MCP or just a chat, you sometimes do not know what to ask because you do not know what's in there. So it's also important to still have this, have a navigation, have a user interface, have nice charts like this is still needed for people to understand what is there. But if you asked me a year ago I would have still said hey, it's 100% a user interface and tiny, tiny use cases for the API and MCP us a few early adopters. But now I would say it's probably something like 90% user interface, 9% the MCP, 1% the API.
This is not actually true because there are some API power users that they alone have a lot more than 1%.
But in general I believe that there's a shift coming from user interfaces to MCP and other agentic interfaces. And this of course also an interesting thing for us.
Not just how do we make our product available there, but also how do we support our customers in becoming more visible in these systems.
[00:42:23] Speaker A: Absolutely love it. Give a final shout out for your app, your software. If you've got any conferences that you're going to, if you're on any particular platform, open to questions and follow up shout out.
[00:42:36] Speaker B: Yeah, so I'm Personally I'm on LinkedIn. I think it's the best platform to follow me. You can also ask questions on X or write me an email. It's my first first name @Peek AI Peec AI.
And yeah, as I said, Peekai is a software that helps to measure and improve visibility in LLM based search and answer engines.
Started with prompt tracking. Now we also do brand perception, we do log file analysis, we can ingest your web analytics data, doing a lot of things and then also providing actions for people.
What are potential next steps to increase their visibility.
[00:43:13] Speaker A: Thank you so much for stopping by and giving us your wisdom.
[00:43:18] Speaker B: Thanks for having me.
[00:43:20] Speaker A: For everybody that's listening, you can find a full recap of this conversation@unscripted SEO.com along with dozens of others with other SEO experts. Stay tuned in. Thanks.
