← Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
SEO Under Siege: 2025’s Collapse of Google Traffic
00:55:48
Loading summary
Transcript44 lines
- [00:00]
A
The Voices of Search Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice. Then visit iheareverything.com welcome to the Voices of Search Podcast. A member of the I Hear Everything Podcast network, ready to expedite your company's organic growth efforts. Sit back, relax and get ready for your daily dose of search engine optimization wisdom. Here's today's host of the Voices of Search podcast, Jordan Cooney.
- [00:43]
B
I'm Jordan Cooney and joining me today is Malta Lanvier, CMO and Chief Product Officer at Peak AI Malta. Welcome to the Voice of Search Podcast.
- [00:53]
C
Thanks for having me.
- [00:55]
B
Super excited. We actually were coworkers together at Search Metrics, got to work together for many, many years and then you went on to Ideallo some in house experience and now here you are back in this innovative software slash growing visibility space. So we're going to dive into that a little bit as we get into this episode, but I just want to start off by just give me, give me, give me where you've been, where you've been going. I know you recently went on a vacation to South Korea and you realized that you weren't using Google as much as you used to and just kind of walk me through where you've been in the SEO space and industry and that experience that you recently had on vacation.
- [01:42]
C
Yeah, sure. So I randomly and accidentally discovered SEO when I was like 15, 16 years old, was building websites. I will spare you all the details, but I, I built websites when I went to school, when I went to university, co founded SEO agency, worked a bit in management consulting. Once upon a time was thinking about obtaining my PhD in with a focus on social media analytics. At some point I ended up at Search Metrics, spent five years there, we both did. And then I spent five years at Ideallo as you already mentioned. And I think actually they are now legally allowed say they're the largest e commerce price comparison in the world, not just one of the largest in Europe.
- [02:28]
B
They're taking on that American marketing slang, the largest in the world.
- [02:32]
C
I mean they are because in America people don't care about price comparison, they care about coupons. So there's no large price comparison in the us. That's why a European one managed to be the largest in the world. And yeah, it's been, I think now already like 8 months ago that my wife and I went to South Korea and I made this little challenge for myself. Let's not use Google, let's see if ChatGPT works. And I used Google maybe 10, 12 times to find a specific website or do spell checking most of the time actually, because it's so, I'm so used to using Google for spell checking and everything else I did with ChatGPT. And maybe there were hallucinations, I don't know. But every place where I went, every food I consumed, every item I bought, it all worked out fine for me. Zero food poisoning, zero allergic reactions, no. No tourist attractions that turned out to not exist. Everything was fine. And then when we were on the, on the airport, on the flight back home, I, I thought about this and I realized that this change that is happening with AI search is going to come faster and it's going to be more severe than I thought. And that's when I decided, hey, I, I have to be part of this change, not, not, not become a victim of this very fundamental change how people research information, discover content and also consume content. So yeah, that's where I decided, hey, work in the space and not just be affected by it.
- [04:13]
B
And that's a really powerful kind of statement there. And I think this is one of the interesting crossroads that the entire industry is in, right? Like yourself, not too dissimilar to myself and many of our listeners. A rich history with SEO. A rich history following those guidelines, A rich history following the algorithm updates and trends, whether they be random animal names to, you know, just now, vague descriptors of what these algorithm updates are. We have this history with how conformed SEO was, whether it be in practice or in media. There was a conformity to it and now we don't have that right. We don't have any genuine guidelines or, or, or any genuine practices to, to apply this concept of AI discovery or AI search. And you know, you had this, this, this epiphany moment when, when traveling to say, hey, look, we, we gotta either be part of this trend or we gotta be, we're gonna probably be extinct. Right? And, and, and when it comes to that, that concept of being part of this trend, what, what mindset and what, what capabilities did you have to acquire to be on that track to want to go in this direction?
- [05:40]
C
I think as an SEO, you actually have all of the abilities because like SEO, Google ranking is also a black box, right? You make changes on the website, you make changes on other websites with, for SEO was mainly backlinks and then you look at rankings and something changes and you try to look at correlation. Some people do it very scientifically with a B test or rather split tests. Other people do it completely on gut feeling. Some went all in on content, some went all on technical things. But at the end of the day you reverse engineer something and try to measure it and it's not so different from what's happening in ChatGPT. Of course, under the hood there are different things happening, but it's not a completely different problem. So I would say skill wise SEOs are probably the best equipped people in the world to work on AI search. The one big change is as an SEO, we were very focused on clicks. We get a click to the website, of course we all say yeah, we are not after the clicks, we were after the conversion and to do them. But at the end of the day, the very first thing you measure as an O is was there a click. And the users, they send you a referral when they click in the search engine. So it's very easy to attribute. It's almost like performance marketing. And there's even a Google search console, a Bing, Webmaster Tools. And you can see even the exact keywords that people type into Google, not on a session level, but in aggregation you can see exact keywords. You get search volume. You live in this very, very good world where you can really understand what do people search, what did they see, what did they click. And this changes of course with AI search, right? Because first of all there's no search volume. If you look at real prompt volume, unless the prompt is thank you or try again, the actual volume is 1 or 0. Because everybody writes unique. People write long prompts that are, that are not repeating. So this volume thing is very different. But then also the results are personalized. Even if you write the same prompt three times, you get three different results. And if I write the same prompt three times, I will also get three different results. So there is some unpredictability in the individual results. Much, much more extreme than in Google search results where yes, there's also personalization, but it's not that extreme actually. Yeah, and there is no click because if you, if you look at the typical ChatGPT answer, there are no links. The only links are the sources. But if I ask about what, I don't know what, what car should I buy, the links don't go to the actual car company that is recommended. In the end, the sources links go to some news website or some Reddit discussion which is very different from the entity, maybe Mercedes Benz that I just learned about and want to learn more about so this whole mindset with the focus on clicks with easily attributable keywords and search volume that that all goes away and I think embracing that, that is the main challenge for SEOs, I believe.
- [08:53]
B
I mean I think it's super unique that you've recently come from this pricing comparison world which is this kind of like middle layer, right? It's this middle layer between brands, consumer sites and the end user, the buyer, the shopper. Right. And the beauty or complexity that pricing comparison sites provide consumers is the ability to create some clarity or decision around buying decisions. Right. Whether it be around car shopping and having clarity around choices based on personal preferences may be safety, size, fuel efficiency, cost, or it may be shoes that, that are more around style, comfort, function, all of these needs. Pricing comparison sites try to organize or define data and information to help consumers make make those buying decisions. It's not too dissimilar to what AI is trying to do today. Right? AI is trying to help us understand, define and create decisions. Slightly different ecosystem, right? The, the pricing comparison one is very commerce related. In, in an AI model it can be basically anything. My, my, my question and, and, and I think that the direction of this is like what experiences did you have at IALO under this pricing comparison world that now help you better understand or reflect on how, how these AI models work and inform not just the SEO community, but marketing leaders at large on how to better understand AI models to reach their customers?
- [10:34]
C
Yeah. I mean as any price comparison website, Idealo was of course a victim of Google shopping and then in the flights comparison space also of Google flights. And I don't know if everybody's aware, but Idealo was one of the leading forces behind a 15 year legal battle in front of the European Union versus Google. Yes. Was recently awarded a multi billion dollar settlement damages package. Yeah, there's no settlement yet. There was a Germany. Google has to pay some billions of dollars. Of course Idealo now wants more, Google wants, Google wants to pay less. So they will get into go into the next chapter of the, of the legal battle. But the one thing that I realized for myself is fighting the giants of the Internet and fighting the change that is happening. It. Maybe it's good for society, maybe it's also good for shareholders if there's a settlement at some point. But as an SEO working in the space it is very draining and annoying and I learned a lot from. We had a lot of meetings with Google because Google was basically forced to either by a judge or by the commission of the European Union to Take a meeting with us. And we learned a lot both in the discovery of these court cases and in these conversations with many, many Googlers. Some of these things are still not public, unfortunately, so I can't talk about them. But there I learned a lot that shaped how I think about things now. But I also made the decision for myself. I don't want to be that person fighting the change that is happening in the landscape even if you can delay it. Like, I mean, Idealo and other price comparisons were able to delay or get rid of certain features of Google shopping in the European Union. And there is now a unique sub feature in the European Union for price comparison websites. So you can achieve something, but it's not how I want to spend my time. So when I, when I realized, hey, AI search is going to reduce the number of clicks significantly, there are not going to be clicks on evergreen content anymore, I was just like, let's not complain about this, let's embrace this and let's, let's live in a world where this is the case because there's no reason to fight it. I think that was the biggest skill thing I learned. And of course I did a lot of thinking about like theoretical thinking, like, okay, how likely is it that Chat GPT will build a merchant center? How much effort would it be? Will they partner with someone? And we are seeing now that what I estimated is actually happening. So they started their chatgpt chapping shopping without a proper merchant center. We can talk about where they get the data from later, but now they are building a merchant center for their couple of first big partners. So yeah, it just made me think about things in a very different way and thinking more about goals, strategies and less about what's the keyword I'm optimizing for today.
- [13:50]
B
Let's talk about this transition. Let's go into our second topic here and just dive into this transformation. Right? Like new conglomerate mega companies are being formed today, OpenAI being the most notable one. But there's many, many more. Right? And I've been in the search space long enough to know about the Netscapes and the Yahoos and all these other search engines that once upon a time existed and Google came to be the monopoly. I don't think there's any debate over that anymore. But the reality is that in AI discovery there is a very different competitive landscape. We have open source models like llama, we have arguably some direct competitors and some very direct marketing shots taken by Anthropic during Super bowl ads this past week when we're recording at OpenAI and ChatGPT directly about how they're incorporating ads into the AI models. But it's a very different landscape. And now that you've been at peak AI and started to not only look at the data that they have, but also your experience transitioning into this organization, how are you seeing this landscape? What makes this AI discovery landscape different from search? And what are the recommendations that you're giving marketers to focus their energy and work on to really influence these AI models?
- [15:27]
C
Yeah, maybe let's start with the landscape. First of all, Google is still the biggest fish, right? They simply by displaying AI overviews over so many search results, that is probably the LLM generated answer with the highest reach and reaching the largest amount of people. And the AI mode is still small, maybe 1% of people are pushed into it. But of course Google can in the future push more people from AI modes into the AI mode. And then behind Google, the second biggest one is OpenAI with also a big lead than over everyone else. Where depending on how you estimate the number of search equivalent prompts that OpenAI or ChatGPT specifically has, they are probably at somewhere between 5% to 10% of the time the size of ChatGPT, but of course sending out at the size of Google, excuse me, but then sending out a lot less traffic. And then Gemini is now catching up to ChatGPT and then you have with a bit of a difference then also Perplexity, Grok and Tropic who are somewhat relevant and actually models like Llama, they are kind of like nobody's talking about them anymore. They are probably working here and there in the background for certain applications. But nobody says I will go to LLAMA to get an answer to my question. Also I think nobody right now is really saying I go to Meta AI to get my answer. And then of course if we look at China for example, we have other models like Deep Sea Quen that are bigger there also in Russia, Taiwan, some other countries, Hong Kong, basically Everywhere where maybe ChatGPT perplexity, Google are not even reachable easily. We have actually also other models that are quite big. And then what is different, what is changing is of course the let's start with what people input with search engines. We kind of like learn. We have to write in this way where we basically describe in like a certain behavior, like I don't know, Queen Elizabeth age to codify that. We actually just want to know how old she is. And with LLMs, people write much longer prompts that are much longer than keywords right? Over 10 words often. And that is the first change. And then of course we have a different way how the answer is happening with something like Google, Bing, Yahoo, it was simple. It's always a web search. There's maybe there are five blue links, maybe there are 10, maybe there's an image box, maybe a video slider. But in the end it's all documents retrieved from an index, maybe it's multiple combined, some form of universal search, but that's it. And with the LLMs it's very different because sometimes there is just an answer from the foundation model. There's no need to use a tool. And when I say use a tool, these can be simple, different things that the LLMs have access to. And one tool is search for grounding. Right. And for, I would say for most commercially interesting prompts, of course there's a search happening where from various sources documents are pulled into the context. And often it's actually just little text excerpts from these documents that's pulled into the context window. And with that an answer is generated. And if you now remember our SEO brain, there are documents retrieved from an index that sounds very much like SEO might play a role in role in that. And if people want to see success quickly, then of course working, trying to influence this grounding process is the one thing that actually works. Of course you can try to influence also the training data so that GPT6 will give the answers that you want. But unless you are very optimistic or a nation state actor or one of the largest lobby groups in the world, like I don't know, oil, tobacco, sugar, I don't think that you should focus on trying to influence the training data for the next big foundational model. I think for most people it is a huge waste of time and resources. But influencing the grounding the web search is happening, that's actually quite straightforward. And there are many, many interventions you can take to essentially manipulate these results or make them better for your own brand.
- [20:13]
B
All right, so a lot to unpack here, so let's dive in and I'm going to start with maybe the least obvious of the questions, but I think it's super important to work back from that to grounding and then foundational data or assets that brands provide for training and context for models. But my first question is about ads. You know, like at the end of the day, Google became a multi trillion dollar company because they sell ads and you know, just, you know, unequivocally like these models are going to have to find a way to make money one of those, one of those ways to potentially make money is ads. And I'm curious to get your perspective on what do ads provide us, not just as an SEO community, but marketing as at large. And candidly, I got to safely assume if you look at any track record with any of these models, when it comes to releasing new features or new capabilities, it's been incredibly small context windows. What I mean by that is like, you know, chatgpt releases, health or anthropic releases, you know, code capabilities in Claude. It's not targeted at mass market. It's not. I mean, the advertising that they put on the super bowl would make it seem like every human being tomorrow is going to use some of these products to go and build amazing technological feats and the next, you know, rocket ships are going to be built off of AI models. Not very likely, right? Because you still have to know how to code in order to use these models really well. You still have to know SEO to use these models really well to do SEO as an example. So my question for you is, what do ads help us understand about AI discovery that we don't know today?
- [22:09]
C
I think in the first version, almost nothing. Because what OpenAI is telling advertisers right now is that there will be little to no performance tracking or metrics or insights. Of course there is validation of buying intent, right? If you run ads and then people buy your product. So apparently there was an option, first of all, there was an option to influence them at that point and potentially they were in that journey already. So it gives validation that there is commercial intent in ChatGPT specifically. And second, it can also validate that there's commercial intent for certain topics, that certain target audiences are using it, certain geographies, etc. Etc. Etc. Hopefully advertisers will demand from ChatGPT some level of transparency. And this has happened at every marketing platform, right? Google had to add reporting, Meta had to add reporting, everybody added more and more over time. And the same will very, very likely happen at OpenAI. And then we will have some, maybe some demand level data, maybe some intense split data. I don't know what it will look like. It will not be perfect search volume and they will not share the exact prompts with us, of course. Course not. Way too much personal identified information potentially in there. But it will provide a level of trans. Additional layer of transparency. And I mean for marketers who are looking for paid acquisition, it's one more channel they can try. And for everybody who is working in the organic AI search space, I think it will be validation because if you're willing to spend hundreds of thousands of euros advertising on a platform, you should also invest time on organic visibility on that platform. And if you take into consideration that people actually Pay to use ChatGPT, not the majority, but many so called prosumers are paying, right? And these are actually probably the people with the biggest pockets who are making the most important decisions. These are the people who will be very annoyed if there are ads. So very likely they will either not see ads or see much fewer ads. So it makes even more sense to do some form of AI search optimization, AEO Geo, however you want to call it. So I think ads are great for me and for everybody who cares about organic visibility in AI based answer engines.
- [24:46]
B
So let's jump in and then talk about the grounding models that these systems are using. And I think I want to start this because in my opinion, grounding in how AI models are using search result data to help inform their responses is one of the most under discussed concepts in our industry today. It is like we've totally forgotten all the work we've done in SEO for the last 25 years and we're not talking about it. So I want to start by just asking you, can you please explain to our listeners what is grounding? Why is it important?
- [25:29]
C
Sure. So maybe let's, let's just talk about what happens, right? I write a prompt, let's say what car should I buy? Maybe I provide some additional context like how old I am, that I want a fast car, etc. Etc. And, and then the LLM doesn't know the answer, so it generates, based on my prompt, so called fan out queries. So it doesn't just take my prompt and put it in a search engine, but it generates multiple fan out queries. And if I enter a very, very long prompt, these fan out queries will be shorter than my prompt. But if I write a very, very short prompt, they can actually be longer than my prompt, right? They can add some information that some very fast LLM model probably already added to it. And then based on the intent of my prompt, we're also seeing some patterns of what kind of words are added. And then this is sent to a search index. And now it depends on what model we are talking about. If we talk about any Google LLM, it's going to use some form of the Google Search API, right? Surprise. If it's using the vertex Grounding API that Google is also offering or an internal version of the Google Web Search API. I don't know, I will let other people speculate on it because some people make very definitive statements on this and I don't know how they would know it. And if it's ChatGPT, we know that Google is one of the sources. We don't know if it's the only source. It used to be. It's all Bing based with an official deal. Then it came out. They're doing some form of Google scraping via a third party provider. And then for Grok for example, it seems to be a mixture of Google, Bing and Brave Search. In anthropic it seems to be Brave Search and so on and so on. So everybody has their partners for search and then the search engines return a list of results, a list of documents, and these documents from all these different fanout queries are considered. But not every single document is taken fully and put into the context window to generate an answer, right? Some documents are selected based on some criteria and these again are different for each and every one. And then some text passages from these documents are selected. So it's not the whole document, it's some text passages, right? And all of that is then used together with my original prompt to generate my answer. That is the grounding process essentially.
- [28:03]
A
Time for a one minute break to hear from our sponsor, Pre Visible. So you're looking for SEO help and you got a couple of options. You could start replying to spam from agencies that claim they can get you to rank number one on Google. You could pay an hourly rate for a consultant who will inevitably nickel and dime you with hourly charges. Or you could work with a cookie cutter agency to quickly launch a strategy less project with low success rates. None of those sound very good now do they? Well, that's where Pre Visible's integrated consulting model comes in. Pre Visible draws From a collective 40 years of SEO and digital marketing experience to unlock your organic growth opportunities. They build custom solutions that combine strategy, technical expertise, content and reporting to effectively operationalize SEO for your business. Pre Visible's four stage approach ensures that your SEO programs thrive by starting off with a strategy first approach. Then they support you in your efforts to create quality content, help you identify technical issues, and most importantly, they'll work with your cross functional teams to integrate your SEO strategies to make sure that your SEO budget actually drives results, not just your agency's bottom line. So join brands like Yelp, eBay, Canva, Atlassian Square, all who rely on the SEO consultants at Pre Visible. For more information go to Pre Visible IO. That's Pre Visible P R E V I S I B L E dot IO.
- [29:32]
B
And so one of the, one of the, I think major misconceptions here that I think you started off with really well is that it's binary. And what I mean by binary is that if you put in a prompt, there's only one search result set that Google or Gemini, excuse me, or chatgpt or anyone's going to use to formulate your response. That is, I think one of the biggest misconceptions. And it's actually, it could be a multitude of search results, it could be a collection that depending on the intent or the broadness of the prompt, it could go after and it could look at, not just 10, it could look at a thousand search results to formulate its decisions and process its actual outputs. Right. Now, obviously there's some realistic bandwidth and I think this is one of the things that a lot of people don't understand also is that this computation from an AI model is very hard. This is why everyone keeps talking about data centers and energy use and all these other concepts, because this is a very, this is what we as humans used to click around Google doing a lot of clicks, and now we don't have to do that. So, so the point of my question off of this and where I'm trying to go on this is like, what does this mean for SEOs and what does this mean for the work that we have to be doing today in order to get better outputs in AI discovery?
- [31:04]
C
Yeah, it means that the foundation for this is still SEO, Right. We need a website with good content. It needs to rank, it needs to be indexed. But there's more, right? We, because it could happen that our documents are always part of the sources, but they never cited. Exactly. Or it could even one step earlier it could happen that we are ranking for these fan out queries, but then we are not one of the documents that is selected to be considered for a source. Because you mentioned this number of thousands of documents. I think there we are in like a deep research scenario, letting Manos run for half an hour. But it's totally possible that there are 200 candidate documents from the four Fan out queries I've been running. But not all of them will be downloaded. Right. Some of them will be considered as sources and then some of these will be downloaded and some of those will be cited. And in the future there will probably be more caching, et cetera in between. And so there are many, many steps where I can lose. Right. The first thing is maybe my website was never crawled by Google. Maybe it was not indexed, maybe it was not ranking for the fan out query. Maybe it was not selected at a potential source document, maybe it was not cited. And maybe if it was cited, a part was cited. That doesn't actually talk positively about my brand. So it was like almost useless for me. Right? And we need to start understanding this pipeline and understanding where it breaks and where we have to do something. So I think that's the big change. And then another thing is as an SEO, I'm usually very focused on my own website, but there are other sources than my own website. There is Reddit, YouTube in the B2B space, LinkedIn, if you sell stuff to Gen Alpha, there's going to be TikTok and Instagram. There can be Quora. There are often big news publications and this now depends on the country and topic. Maybe it's Forbes, maybe it's TechRada, maybe it's TechCrunch. There are many, many, many websites that are that can potentially be among the sources for the problems relevant to you. And for each of them you have an intervention you can take to become more visible. There. YouTube is a source. Okay. Either you look at who are the current top creators for my prompts and then you work with them or you see your competitor as a YouTube channel. Maybe you also need a YouTube channel. By the way, don't just start a YouTube channel to be cited by perplexity. This is a major commitment. Think about it first. Or you see that Reddit is as a source, so you might want to become active on Reddit. Same warning. Reddit hates branded content, advertisements, etc. Be very, very careful how you engage with Reddit. Don't just tell your intern to reply to all of these Reddit threads with an ad for your brand. Then upvote it. You have to be smart about it and you have to consider, is this something that fits our brand DNA? Is this something that is within our ethical realm of marketing? And I know different teams, different companies have actually different areas there and.
- [34:13]
B
Okay, let's. Oh, last question on this, on this section because I love the direction that you're taking that. Because it's created this mental picture in my mind that as SEOs we're going from playing tic tac toe to solving the Rubik's cube, right? In this SEO world we just had rankings that we were trying to push, but now it's not just the rankings, it's this multi dimensional universe of the third parties that are being sourced and cited. The relevance of fan out Queries the prioritization of documents. But there's one thing that you said earlier that I want to come back to for our listeners and I think is super important for the SEO community to understand, which is the foundational training data. One of the things that we've seen and I have talked to a lot of guests on our podcast about this is that in many instances our own branded content, the messaging about our own company, what we do, the pricing, the services is contextually wrong. Right? It's not accurately depicted by AI models. And in some cases this isn't really a hallucination or some just false narrative by the AI model. It's a lack of integrity, it's a lack of cleaning up your own data, creating consistency between your support content, your product content. And this is the training assets that we can influence the most. I know that you said that there's not a lot we can do there and I agree that influencing what a model is trained off to begin with is a very difficult in me, in a much, much more tedious and long term effort. But in the context setting of managing our own brands, what should we be doing there?
- [35:57]
C
I think you already mentioned the most important thing, clean up inconsistencies. And by the way, this can include a April's Fool's joke that you wrote that is maybe if an AI finds it in their context in some search, it's not clear it's an April Fool's joke. And I believe there have been cases where this actually, actually happened for smaller brands. The most important thing is consistency. I've seen multiple examples where people put their own name into ChatGPT and the answer was I'm not sure who that is. Could it be this college basketball player or this, there's this musician, hobby musician somewhere in Norway or is it like a third league soccer player from Germany? I don't know. There's also a doctor in Brazil with that name. But then when they consistently describe themselves, so they didn't say on their LinkedIn profile I'm an SEO for lawyers and then on the X profile I'm a visibility expert in the legal space and then signed their press releases with SEO visibility for law offices and then on their own website I'm a legal SEO, but they always used a consistent wording, right? That can be enough that overnight the LLMs understand that when you put into your own name into them, you get the answer. Hey, you probably refer to this SEO who works in the legal space, but you have to be consistent everywhere where you describe yourself for larger brands. That is less of an issue. The LLMs understand what is Coca Cola, Lego, Mercedes Benz, Tesla, et cetera. But for smaller brands, especially one person brands, that's like my number one advice, be consistent everywhere. And then the second thing is just type the prompts into the LLMs that you think are relevant, see if the answers are correct, see if, see what the sources are and make sure you're well represented on these sources. And that can be very different for every industry, every company.
- [37:58]
B
Absolutely no love that advice around consistency. I think many of us in the marketing and SEO space are going to become much more Germanic about how we approach doing our work and being consistent and orderly about it. And I think it really matters. I see this a lot today. We simply miss the mark by having outdated legacy and in some cases just highly irrelevant content still in the ecosystem. So a lot to be said there and probably a whole nother episode that we could dedicate to this. Just that topic. I want to go deep here for a second on a specific topic around shopping and how you've been reverse engineering or looking at what ChatGPT is doing around shopping queries. You know, a lot of these, these, these changes in ChatGPT as they verticalize themselves are surprising people and giving them maybe some pause around how this works and why this works. Give me, give me a little bit of your context and your experience around what you've seen around shopping and what's actually happening.
- [39:08]
C
Yeah, so what I'm sharing now is something that a coworker of mine reverse engineered, Tom Wells, give him a follow on LinkedIn if you care about that kind of research. Basically, if you look at the media narrative, it's that ChatGPT has these partnerships with Shopify, with Etsy, with some very large retailers in the US and you somehow send them their feed, your product feed, and then you are listed there. But if you type something into ChatGPT, like, I don't know, yellow sneakers, sometimes you get this product grid, right? And maybe if it's all Etsy and, or all Shopify, it might actually be true that this data is coming from the feed. But in the vast majority of cases that's not true. And they are scraping Google Shopping, they are using a third party scraping provider and they generate in parallel to the regular fan out queries, they generate very specific ecommerce fan out queries that are different. And if you take these E commerce fan out queries and put them into Google Shopping, you get an 80%, roughly overlap of the exact product titles, etc. Etc. And they have in the source code, like not the source code, the Java, the JSON payload that's transferred to the server and the client. If you look in there, you can actually see these fan out queries for E commerce and you can see a lot of identifiers that are very, very obviously Google Shopping identifiers and even one identifier that allows you to make a very strong guess which scraping provider they are using. And we spent a lot of time verifying this and it's actually true. They are just scraping Google Shopping and that is what the ChatGPT shopping is based on. And this very much fits the narrative that they just use Google for web for like with scraped results right by a third party provider for their grounding. Because as a company they want to move incredibly fast. They don't want to reinvent the wheel when they don't have to. So I'm 100% convinced the vast majority is just scraped Google Shopping results. And that's why also in the links of these products in ChatGPT you often still have the identifiers because when people put their product feed into the Google Merchant center to be eligible for Google Shopping, they put certain identifiers and you will find these again in, in ChatGPT. And this is something, these identifiers in the URL is something I saw immediately when the feature launched. But back then I was still blind to that. They had to their scraping. I've thought okay, maybe because they had this deal with Bing, they get it from Bing Merchant center. And many people, they set up their Bing Merchant center by clicking one button import from Google. But then they were also really big companies and I thought is this big company really not changing the UTM parameter for Bing? Do they really have their Bing Merchant center running with like Google Merchant center parameters? Doesn't sound right to me. But back then I didn't imagine that they would just be so brazen to just scrape Google Shopping. But now we know beyond any reasonable doubt that they are just scraping Google Shopping. So yeah, that's where the data is coming from. And this will probably be reduced over time because they might now onboard more and more merchants onto their own shopping merchant center. But yeah, right now that is the status quo.
- [42:48]
B
I mean that that's where I was going to go. Which is like Agenda Commerce protocol. And what, what what ChatGPT and OpenAI have released as a standardized format to provide a feed into their shopping experience is a very limited portfolio of commerce brands today and largely a massive PR stunt and a great stock price Booster for the publicly traded companies that jumped on Etsy, Walmart being two notable brands that when they jumped onto acp their stock prices significantly increased. Again, we're not here to talk about stocks and investment strategies, we're here to talk about commerce strategies. And in this outline that you provided, where essentially OpenAI is leveraging Google Shopping to ground themselves again into the shopping experience that Google provides, how should merchants be thinking about specifically their data, managing their data and managing their feeds today?
- [43:58]
C
Yeah, I think I would always maximize for reach and exposure and I would share my data with ChatGPT because if not you probably share it with Google anyways and then if they're scraping it, they just have it. And I think most merchants actually don't really have unique data that is not also on other websites. Right. If you have something unique, you can think about protecting it if you are a really, really big brand with really loyal audience. But I would go all in on sharing it actually just trying to maximize the traffic. I see a general challenge for many, many, many merchants which is that if you look at Google in the United States for e commerce queries, there is no need anymore to visit a category page on an online shop, right? You don't need a product listing page. Google is a product listing page essentially. And you can just click there to the product details pages of the different shops. And I think ChatGPT will over time become even more extreme at that because they don't need you to manually click filters. They can ask questions, ask follow ups, they can use what they already know about you and they can just say here, here are the three options. Buy one of these. I already know what's good for you, I know what you like. And that means that over time many online shops might become just fulfillment centers. They, they only get an order and then they do the shipping and they have to handle the returns and they have all the risk because they have to hold the inventory. And it will be very, very difficult to have a differentiated experience because if my shopping agent is buying for me anyways, I don't care about the layout of the online shop or how smooth the shopping experience is. And then it will at some point come down to price and that's always a race to the bottom. And it's not great to be in it unless you are the largest one. Then you can at least price out your competitors over time or just be better because you have more scale effects. But yeah, I see a huge problem coming for many merchants because every large online shop in the last 10 years decided to become a marketplace so they can't keep up the quality as much as they used to when they were just one merchant. Right. If you're a marketplace, you always have more inventory but also quality suffers. Always. I've never seen an example where it didn't happen. And then if you are a marketplace, what's your differentiation versus Amazon? What's the difference? And I predict that we will see a lot fewer online shops in the future. And I also believe we will see factory to LLM where a factory in China that is producing goods that they normally sell on some Chinese e commerce marketplaces that then come for three times the price to Temu and then for 10 times that price to Amazon. They will just share a feed with ChatGPT. ChatGPT will take care of generating nice product pictures, translating their text into English, into German. The some, some version of an agent or UCP or one another protocol will handle the actual purchasing process and you will get your your shipment and then there is really no need anymore for most online shops.
- [47:28]
B
Yeah, I want to transition us to this last topic because it really relates to this whole future ecosystem for not only online shops, but but candidly all all online brands, which is measurement and how we should be measuring our efforts, our work and ultimately how we should be thinking about the health of our business in an AI discovery world. Right. We don't have clicks anymore. Arguably you already made this clear earlier in our episode search volume. And these monthly search volume metrics are also radically distorted. And so in a clickless volume less world of data, what are we measuring?
- [48:15]
C
So I would measure four things and there's a clear priority to them. The number one thing, if it's possible, is self reported attribution. Ask your leads, ask your users, where did you hear about us? You will be shocked how many will already reply with ChatGPT or LLMs. I've seen one case where a click based attribution set 0% of leads from ChatGPT asking people revealed 22% from ChatGPT. Crazy difference. The second thing I would measure is visibility on a set of prompts that you develop or develop together with some with either a tool provider, an agency. However you come to a set of prompts that is relevant for you run these every day and then look at visibility as an aggregate of a multiple product prompts run on multiple days. Never obsess over one single chat, there's too much randomness in there. Always look at aggregates of many, many chats. So these are the top two and then below that there are two supporting metrics that I really like. The first is citation share. So of all the citations and all the chats that are relevant for you, how many come, how many of these citations come from your own website, especially looking at that in relation to your competitors. And then if you increase this, very likely also your visibility will increase. You will get more sales, more leads, more traffic. And the second one is source coverage. So out of all the sources that are used by the LLMs, how many are mentioning you like look at the top 20 documents used at sources for your set of prompts, are you mentioned there or not? And then in the second step are you mentioned in a positive context? But that's then more complex to actually do. And if you get the citation share and the source coverage up, I guarantee you you will get up your visibility. And I've seen now many, many case studies with customers of ours where if you have a representative set of prompts or directionally correct set of prompts, you will also see more sales, more clicks, more revenue once visibility goes up. That is how I think about tracking at the moment for LLM visibility or LLM impact traffic revenue.
- [50:22]
B
I love the starting place and I think it's absolutely critical to understand and get even directional data on where and how consumers are finding your brands. And even if it's just surveying or getting some sample for larger brands, it's just so imperative because the transformation is already here and the systems, the platforms and tools we use to measure are simply not up to speed. They're simply not there yet. And so we're not able to use the traditional data platforms that we did in the past to truly measure the future of where things are going. I want to transition us to our lightning round, so I'm going to ask you a few related questions from our episode and you'll give me a quick 30 to 60 second summary of your thoughts on these questions. You ready, Malta?
- [51:14]
C
Let's go.
- [51:15]
B
All right, what's one SEO habit teams need to unlearn immediately in this AI first discovery world?
- [51:25]
C
Click based attribution and think more like a brand marketer than a performance marketer?
- [51:31]
B
Yes. Yep, Absolutely. We just talked about all this measurement stuff and it starts with your performance based efforts. Absolutely. What's the most dangerous thing tools encourage when they automate content or work at scale?
- [51:49]
C
So short term automated content creation works really well, but long term, I've never seen it work if it churns out complete articles, not even if you have 60 steps in your workflow to make it good. Content, it always results in a Google penalty or at least a loss in Google rankings. And in recent weeks, Lily Ray and Glenn Gabe have shared example after example after example. Especially Glenn has been doing that for the last two years. Just scroll through their LinkedIn and X feeds and look at these results. It's always the same. This visibility curve goes up and then it goes down. And you can even go to the archive ARC version of these tools that promise it and look at the reference customers they had a year ago on their website that they now no longer have on their website. And I'm not joking, you will find again and again you go to Ahrefs or Semrush or your favorite SEO tool, you will see traffic goes up, traffic goes down. Even the ones where they published a case study, they went down three months later. If it sounds too easy, it works too good. It will come and bite you at some point.
- [53:01]
B
Yeah, great advice. Love that authority still matters. Where should brands invest today?
- [53:11]
C
I mean very abstractly speaking, be the answer that people expect be so relevant in people's mind that when they ask about sneakers, they expect one of your sneakers to be on that list. How to do that? Ask a brand marketer. They know better than me as an SEO, as a content person, I would invest in content across the whole funnel. Even if it will not get clicks, it will influence answers. And I would figure out the tracking and attribution thingy which can, and the solution can be to say as a company we don't need perfect tracking and attribution, but based on asking 1% of our customers, we know that a lot of them hear about us on ChatGPT, so we will continue to invest. But don't, don't just do nothing because the tracking and attribution part is difficult.
- [54:02]
B
And one follow up on this. There's so much noise about Reddit and now Wikipedia and these, these kind of community platforms that, that take up a lot of the mind share today of, of where brands are thinking about their authority. How should you be thinking about that if you're in house?
- [54:23]
C
First of all, find out which of these are cited by LLMs for the prompts that are relevant for you and for your target audience. And then think about how you can become visible there. That can be having your own profile on some of these, it can be working with influence and creators. And then on Reddit for example, it really depends on where you fit on the ethical marketing spectrum. And I don't want to give any advice there. I would just say this. There are some Agencies that do things to make it seem like users are talking about you. I would not recommend that, but it exists that I've seen people use it. But if you're building a long term brand, be very, very careful with it.
- [55:07]
B
Agreed. What signal do you trust more than traffic when evaluating AI Discoverability?
- [55:16]
C
Self reported attribution As I mentioned earlier, ask your leads, ask your users. It's the best thing I have at the moment.
- [55:24]
B
Great. And you also mentioned earlier the concepts of prompt selection and being mindful of that. Tell me a little bit more about that.
- [55:35]
C
Yeah, so don't over obsess about it. You don't need to know word for word what a user typed yesterday because for the next six months nobody might cite the exact same prompt word for word anyways. And LLMs are very good at understanding context, topic and intent. So just think about a topic and then write a couple of prompts. Use maybe an LLM to help you. Like what would somebody who cares about this write to get this answer? If you have customer support tickets, if you have transcripts from sales calls, if you have a search bar on your website, if you have a chatbot on your website, look into that to understand the exact terminology your users are using, your real users, not how you as a vendor talk about it, but the words that your users are using and give that as an input either to an LM or to yourself. When you write prompts, never check your visibility for one prompt on one day. Always think about a topic with multiple prompts and then think about do I have different Personas? Do I have different sales funnel stages? It's search intents that I care about. And then also generate prompts for these per topic. And always look at this aggregated data.
- [56:49]
B
And that's a great place for us to wrap up this episode of the Voices Search podcast. A huge, huge thank you to Malta, CMO and CPO Peak AI for joining us. If you'd like to connect with Malta, you can find a LinkedIn profile in our show notes or on the voicesofsearch.com you can also learn more about his company Peak. That's P E C AI and if you haven't subscribed yet, want a daily stream of SEO and content marketing insights in your podcast feed? Hit the subscribe button in your podcast app or on YouTube. That's all for today and until next time. Remember, the answers are always in the data.
- [57:32]
C
Sa.