Andreas Voniatis, Christian Klepp

Andreas Voniatis  00:00

A topic on content strategy. Well if you’re looking at opinions on content strategy, you’re probably more interested in opinions of either marketers or business leaders, as opposed to everybody. Not everybody even thinks about content strategy, so there needs to be a bit of audience profiling because AI is persona driven. It’s no longer just keyword driven. So you want a constant strategy that reflects the needs of your target buyer rather than the internet as a whole.

Christian Klepp  00:32

It’s a topic that everyone is talking about and trying to figure out these days. AI search. Many are keen to get in on the action, but these same people are the ones who are still using outdated SEO (Search Engine Optimization) playbooks, much to their detriment. The landscape has changed considerably, and those who don’t adapt will get left behind in a world where being found online matters now more than ever. So, how does knowing more about the science increase your visibility in AI search? Welcome to this episode of the B2B Marketers on Mission podcast, and I’m your host Christian Klepp. Today, I’ll be talking to returning guest , who will be answering this question. He’s the founder at Artios and a published author who helps to guarantee that you’re a winner in AI search. Let’s dive in. And away we go, Mr. Andreas Voniatis. Welcome back to the show.

Andreas Voniatis  01:19

Thank you for having me, Christian, it’s a pleasure.

Christian Klepp  01:22

It’s it’s great to have you back, Andreas. I feel like it’s been a million years since the last interview, but you know we’ve been keeping in touch, and much has happened since then. And I’m really looking forward to this conversation because, wow, when we got on the pre-interview call, we unpacked a lot, and I was thinking, there’s a lot, there’s a lot for people to learn here, but we gotta, we, we gotta spin it in a way that it’s, it’s digestible for the layman, and I think that’s the challenge of today’s conversation.

Andreas Voniatis  01:55

Absolutely, yes, it’s easy to be bamboozled by all the terminology that’s out there.

Christian Klepp  02:02

Absolutely, absolutely. So let’s dive in. So Andreas, once again, it’s wonderful to have you back on the show for this conversation. I would say let’s focus on the following topic and unpack it from here. And that topic is how knowing more about the science will increase your visibility in AI search. And before anybody says, “Oh my gosh, not another podcast interview about AI search, I would say, “Hang on a second, this one’s different because we’re talking about the science here. So I’m going to kick off the conversation with this question. So in our previous conversation, you mentioned something to the effect of most responses to AI-driven search disruption, they tend to follow the same playbook, right? So publish more content, build your authority in your space, and optimize for featured snippets. So why do you think these strategies miss the point?

Andreas Voniatis  02:55

They miss the point because everybody can do those things. Publishing more content. Let’s just unpick that one. You know, every second of the day, more content is being published on the web than ever, and that just most you know most of it is going to be self-serving. Even if it’s helpful content, you know, are you telling me there wasn’t a helpful article on the subject on that topic already? You know, and that’s a lot for traditional and AI search engines to wade through, to process. You know, they’re they’re having to cope with, you know, a lot of this added noise now. Most of it is, you know, not really saying anything new. So it can be quite frustrating if AI was a person to have to, you know, go through all of that extra content being added every second and try to find something that actually adds value to its models. So, you know, publishing more content, yes, but is it value adding? Is it making AI smarter? Is it adding? Is it telling the world something it didn’t know before? You know, so publish more content. It needs a little bit more than just more content for the sake of it.

Christian Klepp  04:20

Absolutely, absolutely. It sounds like a lot of these these people that you’re referring to that are putting out a lot of this content. It seems that it’s. I wouldn’t say it’s a knee jerk reaction, but it’s very execution driven. It has to be like we have to populate we have to populate the website with as much content as possible, our social media feed with as much content as possible. There may or may not be a strategy behind that content. It may or may not answer questions that potential customers have about specific challenges that the products or solutions that companies provide can fulfill. So there’s a lot of like maybes in there, right?

Andreas Voniatis  04:59

Yeah, that’s right. You know, and don’t get me wrong. It’s great for marketing teams to ensure the businesses or the organizations are providing those answers that to the questions that are being asked by their B2B buyers. The question is, is that why should AI or traditional search engines prefer your answers over the many other websites out there, and from an AI perspective, the answer would be information gain. You know, does does it give a unique perspective based on scaled learning or any insight? You know, does it does it actu ally add value? You know, otherwise you’re just one of the many that are trying to be the answer to that question, but actually, you’re not saying anything value adding. So it’s all about the value add.

Christian Klepp  05:55

That’s absolutely right. So based on what you’ve just said, why do you think this is a data science problem and not a content strategy problem? And I think that’s probably a unique perspective because we’ve had a few guests on the show to talk about AI search, but none of them have spoken about this perspective.

Andreas Voniatis  06:14

Yeah. So the thing is, any good marketer relies on data. You know, all good marketers, especially marketing strategists, they do their market research. So why would content strategy be any different? Surely it would be data driven. Well, if it’s data driven, it starts crossing over to the data science realm. And so, you know, a lot of content strategy, if they’re using tools like Semrush or any other sort of well-known tool, think about where the data comes from. It comes from Google Ads, right? And that’s really just telling you, you know, where people are where people are bidding, you know, the keywords that people are prepared to pay lots of money for, and yes, okay, there’s a loose connection to you know that presumably if if businesses are or organizations are bidding on those keywords, it’s because they see demand, but that’s a little bit like trying to play the piano with boxing gloves. You know, if it becomes a data science problem when you actually want to really refine in as to what is it people are interested, what are they discussing online, what are the questions that are being asked, and this is this is something that data science can answer with a lot more polish and refinement than just content strategy, you know, without the data-driven element.

Christian Klepp  07:54

Okay, but could you give us an example of that? Because you said that data science can address that in a better way than content strategy. Give us an example of how we can do that.

Andreas Voniatis  08:01

Yeah, so it’s not. I wouldn’t say data science is separate to content strategy. I would say that it’s a data science or a data-driven content strategy that’s better than just content strategy.

Christian Klepp  08:15

Okay, fair.

Andreas Voniatis  08:17

So you know, a content strategy that is using data science to actually, you know, tool the internet like we do, which is to data mine the internet, do some and layer on top audience profiling because a topic on content strategy, where if you’re looking at opinions on content strategy, you’re probably more interested in opinions of either marketers or business leaders as opposed to everybody. Not everybody even thinks about content strategy. So there needs to be a bit of audience profiling because AI is persona driven. It’s no longer just keyword driven. So you want a constant strategy that reflects the needs of your target buyer rather than the internet as a whole.

Christian Klepp  09:14

You said something there, which could probably be the title of this episode: persona driven and not keyword driven, and I think that’s such a significant difference, which leads me to a follow up question. And I’d be I’d be curious to know your thoughts on this. I almost think I know the answer, but I’m going to ask you anyway. Do you think that bidding, you know, this this older SEO approach of bidding for keywords in this day and age of the AI search, is that still relevant?

Andreas Voniatis  09:44

Well, bidding for keywords in terms of Google Ads, yeah, it’s still relevant. Although I’m hearing a lot of noises that these ads, these keyword-driven ads, are not high return. Or higher as they used to be, but from an SEO perspective, it feels a bit arcane now to to target by keywords alone. I think you really need to do some real marketing where you think about your target buyer, and then you do some market research In terms of what is it, what what are the conversations they’re having online? What are their concerns? What what is it? What is it they? What are the challenges they’re trying to overcome? What are the solutions they’re evaluating? What is the pain point? All of those sorts of things, And because it’s their their perspective and their opinion that matters, and that will shape how well, first of all, which topics you’re going to cover and how you’re going to cover them.

Christian Klepp  10:56

Absolutely, absolutely. Just just from your own professional experience, right? Talk to us about what you think is currently working and not working when it comes to visibility in AI search. And I’m sure we could, you could probably talk about that for 10 hours. But just give us, give, give us, give us maybe your three to five things that you think are working, and three to five things where like, oh gosh, don’t don’t do this stuff anymore.

Andreas Voniatis  11:22

Yeah. So look, it there’s a lot of advice online. You know, the number of times I’ve met people that have said they could do GEO (Generative Engine Optimization) themselves because they’re falling prey or falling into the trap that you know a lot of the online advice they’ve read is going to magically transform their visibility fortunes in AI, but a lot of the stuff they’re reading, like schema and and restructuring your restructuring your content into FAQ format, it’s not that it doesn’t work. It’s just that it’s very cheap. So if you can do it, so can the rest of your competitors in your industry or market. Therefore, it’s not that it doesn’t work. You still need to do those things, and you still need some basic SEO. And so all of those things still matter, but they’re just not the game changer. The game changer is the research that goes into your content. It’s the information gain that that’s the criteria that AI is looking at, and search engines, by the way, not just AI. It’s just that AI, compared to search engines for SEO, has a much higher threshold. It’s much harder to be fooled with search engines. You know, you would take a target keyword, you would look at the top 10 ranked content, you would take all the best features and then make it better. Whereas with AI, you need a lot more than that. You need to you need to actually have data that makes you, and it and it can’t be data that you prompt engineer from AI. Even on deep mode, it needs to be data that AI has not seen before, and it has to be scaled, significant, unique, proprietary. It’s got to be the reason that AI can only get it from your brand and nowhere else.

Christian Klepp  13:26

Yeah. No. No. Exactly. Exactly. It it it seems like it seems like a lot of people are like trying to look for it’s just human nature, right? Trying to look for shortcuts, right? Even in my line of work, you know, with with copywriting and brand strategy, you know. There’s some people that I talk to, and they say, “Oh, we can do the copywriting. We just use Chat GPT. I’m like, “Well, knock yourself out,

Andreas Voniatis  13:48

yeah.

Christian Klepp  13:48

If you if you want to sound like everybody else, and you know, you’ve seen it, right? We’ve all seen it. We’ve seen we’ve gone on the websites where you read the copy and like, oof, that doesn’t. It’s not very smooth, or or or it feels a little bit awkward, and like yeah, because it’s been AI generated.

Andreas Voniatis  14:07

Yeah, and we’ve seen recent news: the EU have brought in a law forcing AI search platforms to watermark their content so that it’s easy to tell whether content or text has been AI generated, they have to comply with this by November. I believe Claude has already responded, as has Google. Google have even taken out a patent on this, you know, and this is in response, I believe, to all the AI slop that we see not just on the web but also in social media. We clearly know AI slop when we see it.

Christian Klepp  14:51

Yeah,

Andreas Voniatis  14:52

You just don’t want that, and you know there’s there’s regulations being introduced now to combat it.

Christian Klepp  14:59

Yeah, and right. So, rightly so. I mean, it’s it’s like everything else, right? Like you can’t you can’t let something like AI just go unregulated without any ramifications. Because who knows? There’s there’s been some cases already where it’s gone a little bit too far.

Andreas Voniatis  15:15

Yeah, absolutely. We we’ve seen you know we talked about how AI and search engines have had to contend with noise being added to the internet, and that’s the human-written noise. Imagine how this is accelerated by AI-generated content. You know, that’s just noise on steroids. So I’m not surprised that laws are being introduced. You know, not necessarily to help AI, but to help our own sanity, so to speak.

Christian Klepp  15:46

Yeah. Well, it’s also to help regulate the the the people. Let’s be honest, right? The people that are using AI.

Andreas Voniatis  15:52

Oh yes, absolutely.

Christian Klepp  15:54

All right. Here comes the big one. For those of you that don’t know, Andreas is also a published author, and in your latest book, Generative Engine Optimization with Python, you talk about leveraging data-driven methods. For hang, hang on a second here. This is a this is a mouthful. Large language model retrieval and citation, or you know LLM, right? So you also mention, and I’m going to quote you here that the outcome is marketing visibility, but the method is rigorous science. So here’s my question, and it’s a loaded one. And here’s my challenge to you, as we as we talked about at the beginning of this recording, not everyone who’s part of a marketing team is a certified data analyst or data scientist, or for that matter, well versed in Python. So here it comes, and I’m gonna ask you these questions one by one. But using layman’s terms, walk us through how to. And here’s the first one: reverse engineer how ChatGPT, Gemini, Perplexity, and Claude select and cite sources.

Andreas Voniatis  17:00

Well, look. Here’s how we do it to traverse engineer the results in AI search. First of all, you’ve got to have a dataset. So we we track the we first of all have a dataset filled with say a collection of prompts, right? So we have a prompt library, lots of prompts that you would expect a company in your market or industry to be visible for, because it’s the prompts that your, if you manufacture glass, for example, these are the prompts that your buyers of glass would use in order to find you, and then we track as many manufacturers as possible, and and then we collect that data for a period of time, say a month. So we get a good data set. This is so it means We’ve got, say, 100 prompts, 100 firms. So that’s already that’s already 10,000 data points, and you times that by 30, that’s 300,000 data points. Okay, then we’re not just looking at the responses, we’re also looking at the the sources that get quoted or cited by AI. So that in itself can easily blow it up to a dataset of 30 million or 3 million, let’s say, to be conservative, okay. So, to be able to find the needle in the haystack as to you know what can possibly explain the variation in why some sites in glass manufacturing get mentioned more often, and one why some sites don’t get mentioned very often? You you need statistical skills as a minimum, and you need data site skills to to be able to clean up that data and you know find any holes and make sure you’ve got a clean a clean data set that you can feed a statistical model to be able to work out what’s actually driving visibility or mentions and what isn’t.

Christian Klepp  19:30

You brought up something at the at the end of that, which I want to go back to because I’ve been seeing this time and time again, and people don’t realize how important data hygiene is in having clean data. Talk to us about that because there’s some out there that haven’t cleaned their databases in a while, right?

Andreas Voniatis  19:49

Yes, yeah. So, like for example, you may have incomplete data. You may have malformed data. You may have. Data that follows, for example, you may have a skewed dataset. So, are you going to log transform it? Are you going to, you know, apply a log, you know, take a logarithm of the numbers so that you know your statistical model can find more variation to explain why some companies are being mentioned more often than others, so those are just those are just examples. It could be that there may be elements of the dataset or rows you might need to reject because they’re not fit, you know. So it’s it’s things like that, really, and also often the data you get. What makes there’s a famous saying in maths, which is all models are wrong, but some are useful. So yeah, often it’s the questions you ask of the data that make one model a lot more useful compared to another model. If you just take a pure data science approach, but you have no industry expert questions to ask or engineer hypotheses that could make the model a lot more useful. Then you’re going to get a very noisy model that doesn’t tell you much.

Christian Klepp  21:25

Right, right, right. Okay, okay, I see. So now we’re going to go to the second part of the question, where I am going to ask you again. Walk us through. Right. So identifying which communities and platforms AI systems treat as authoritative.

Andreas Voniatis  21:43

Yeah, that’s great. That’s actually covered in the. So the previous answer I gave was has a chapter dedicated to that. But in terms of this this particular question, we have a chapter dedicated to finding what we call source authority domains-the ones that get cited to help generate the answers. Again, this comes from data mining from all the domains and the prompts you’re tracking. You see what answers get generated and what sources get cited, and then you can start doing some some analysis as to which ones get cited most frequently by query type and by site type? So you know that’s just very top level. So that can help guide your strategy specifically for digital PR as to where you want to be seen because that’s that that those are the places that AI is relying on, and we use that, by the way, to to for for our own client campaigns. We we that that’s how we’re able to guarantee outcomes is because we’re data driven. We’re following where AI is going, and then we’re using that to you know it’s all about 80/20s. So, you want to you want the 20% of your effort to have 80% of the effect.

Christian Klepp  23:07

Sure, sure, sure, absolutely, absolutely. Okay. Final part of the question: Build monitoring infrastructure that makes citation probability measurable and improvable.

Andreas Voniatis  23:23

A lot of the code that’s in my book gives you code to tackle different parts of the GEO process. Now, if you don’t want to be paying 1000 Canadian dollars every month for a professional tool, then we have a chapter dedicated that shows you that gives you the code as to how you can run the code so that you can collect these results on a daily basis. So you have all that data to help you monitor your visibility.

Christian Klepp  23:57

Fantastic, fantastic, and Andreas, credit to you because I know it’s I know it can be challenging if you’re in the field and you’re very well versed in technical terms and it’s challenging to like water that down. It can be challenging also like to just take out all the acronyms and you know all the all the language that the experts understand, and then just explain it to the layman. So, so thanks for that. I think the other question that I had for you, now that you’ve laid this all out for us, is what kind of success have you been seeing now? You know, in this approach, this this this you know, using this data science approach to AI search.

Andreas Voniatis  24:39

Oh, look, we provide guarantees for our clients. If you’re not visible in six months, then we end up working for free. And thankfully, we haven’t had to work free once, which is good. You know, even with solopreneurs that we don’t typically, you know, position our marketing communications. Towards, but they find us, and you know, we we’ve even managed to make them some of the most visible messaging architects or brand messaging consultants in the U.S. or or keynote speaker in their field. So usually we prefer to work with startups and scale-ups, but it just goes to show it can even work for people that don’t have a marketing director or marketing manager, a dedicated marketing person with a dedicated marketing budget, and who aren’t doing 50 different channels at the same time, you know. So it just goes to show if if we could do that for the the solopreneur, you know, for the SMB, it’s pretty business as usual for us.

Christian Klepp  25:59

Indeed. Okay, here comes the next question, and again, you know, we’re going to try not to go down a very deep rabbit hole with this one. But talk to us about how marketers should measure GEO performance. So, like, what are some of the key metrics that they need to watch out for?

Andreas Voniatis  26:15

Yeah, I think the one thing to be very clear on is that someone who’s making you more AI search visible? We’re measuring visibility, and we’re we’re not measuring, you know, how many leads you got from, you know, if your website is a bag of spans, as they love saying in England, you know that you know the AI specialist is not has no influence on your conversion rate, your messaging, or anything like that. But to answer your question, you want to be measuring the sentiment. You want to be measuring your mention rate. You know how often are you being mentioned when your buyers are prompting. You know in 100 in 100 prompts for the same prompts, you know what’s the percentage of that 100 that you’re visible for? I wouldn’t just look at the mentions; I’d also look at the citations. And this is one of the beauties of AI search results compared to traditional search results: is that if you’re being cited, if your content’s being cited, it’s free brand awareness because you end up being part of the mention, the main answer, because you’re constantly being quoted as the source. So it’s free brand awareness, and it’s a nice early quick win while you’re building up your profile in and your position in AI, AI is training data models, and sentiment is really important because you know if someone is prompting, you know which companies should I avoid for glass manufacturing? You definitely don’t want to be in that list. So sentiment matters, whereas you know, with search edges, it’s just pure rank, right? So you you just you’re just happy to be ranked. But with AI now, it’s a lot more intelligent. So therefore, you know, context matters.

Christian Klepp  28:15

It’s almost like a natural evolution of things, right? So it’s gone on to the next. I don’t know if it’s the logical conclusion, but like yeah, yeah. Let’s just assume it’s the next logical conclusion, right?

Andreas Voniatis  28:29

Oh, it certainly is. Like if you look at the bigger picture in business intelligence, you know, to put together a business intelligence system, you have the data, the data warehouse, and then you have the dashboard, right? So, you know, in terms and in in geek speak, we call that ETL: Extract, Transform, Load. Well, if we talk about search, the Extraction layer is are the search engines because they’re getting the data off the web. T transform, well that’s the AI modeling layer, making sense of that data, and then you’ve got load instead of the dashboard in the search context. That’s your AI agents that are actually transacting and performing actions off the back of the understanding that came from the transform layer. So, and you know, look, when it comes to agentic search, if you’re not AI visible today, you’re nowhere tomorrow.

Christian Klepp  29:28

Right, right, exactly, exactly. Andreas, it’s been a great conversation. Thank you so much for coming on, and you know, it’s always a pleasure. Please, once again, for those in the back, quick introduction to yourself, how people can get in touch with you, and if they’re interested in your book, where can they go and get it?

Andreas Voniatis  29:48

Yeah, absolutely. So I’m Andreas Voniatis. I’m the founder and CEO of Artios, and we help B2B businesses. Be AI visible for when their buyers search there. I built my first LLM four years before ChatGPT when GPT two was announced, and you can find me on LinkedIn or you can go to my website artios.io.

Christian Klepp  30:17

Fantastic, fantastic! Once again, always a pleasure to have you on the show. Take care, stay safe, and talk to you soon.

Andreas Voniatis  30:24

Thank you for having me.

Christian Klepp  30:25

Thanks. Bye for now.