Welcome to another episode of The Search Session. I'm Gianluca Fiorelli, and today I'm joined by Ramon Eijkemans, a big-site SEO specialist and SEO-automation builder from the Netherlands.
We discuss how SEO is evolving as search becomes more AI-driven, and what that means for visibility, relevance, and strategic thinking.
We also explore how this shift affects content, e-commerce, and the broader role SEOs can play inside a business.
In this episode, you'll learn:
Why AI search demands clearer, self-contained content: utility writing, semantic HTML, and explicit context help both users and LLMs understand pages more accurately.
Why SEO benefits from a humanistic and information architecture mindset: language, context, questions, and classification turn data into meaning and make information easier to discover.
How to approach AI search without chasing every model separately: focus on shared mechanisms across language technologies, then prioritize the platforms that matter most to your audience.
Why SERPs are more than rankings: they can be rethought as a shopping window of a market, turning search visibility into a broader form of market research.
What makes scalable SERP analysis difficult: the challenge is keeping many moving parts stable while using reasoning-capable LLMs to uncover deeper insight at scale.
Why sentiment and user context matter in SEO: relevance depends not only on topics and intent, but also on the needs, emotions, and circumstances behind a query.
How to improve thin e-commerce pages without hurting UX: add precise, useful context through page design and “situation” copy, without bloating content or changing the page’s search intent.
Hit play and join us! Ramon brings two decades of big-site SEO experience and a refreshingly humanistic take on where search is heading.
Topics covered: AI search · language models · utility writing · SERP analysis · shopping window · market research · sentiment and intent · SEO and UX collaboration · structured data
About the Guest
Ramon Eijkemans
Founder of Eikhart and SEO innovation, automation, and marketplaces at DPG Media Nederland
Ramon Eijkemans has 20+ years of experience in SEO, with a strong focus on large-scale websites, enterprise search challenges, and the evolving intersection of automation, AI visibility, and strategic growth.
In 2001, he founded Eikhart, where he now works across enterprise SEO consultancy, SEO automation, and LLM visibility innovation for large and complex websites.
Since 2023, he has also worked with DPG Media Nederland, supporting several marketplaces in the DPG environment while contributing automation ideas and strategic SEO thinking to the broader team.
He also shares his insights on SEO, automation, and AI visibility through the Eikhart blog.
Video Chapters
Transcript
Full conversation between Gianluca Fiorelli and Ramon Eijkemans.
Gianluca Fiorelli: Hi, I'm Gianluca Fiorelli, and welcome back to The Search Session. Today, we are going to have a guest from the Netherlands with a Spanish name, but he's not Spanish. And he is cool, the Dutch school of SEO. There are so many SEOs from the Netherlands who are really contributing to the community and to the knowledge of the community.
For instance, another case, Barry Adams, who sounds very British, is coming from the Netherlands. As well as Yoast and many others.
And our guest is an SEO consultant who works a lot with enterprises, and he loves to define himself as the mad scientist. And somehow I can tell you he's a little mad, but in the good sense. Our guest today is Ramon Eijkemans.
Ramon Eijkemans: Almost.
Gianluca Fiorelli: I don’t know. Was it good? I love you, Dutch, but I don't like your surnames. Sometimes they are very complicated to pronounce for a Mediterranean person like me. How are you doing?
Ramon Eijkemans: Busy.
Gianluca Fiorelli: Oh, that's good news, no?
Ramon Eijkemans: That is good news, yes. Last year was slow, not due to me, but I don't know. I talked to other freelancers and agencies as well, and they said, "Yes, things are slow." I don't know why. Well, there could be many things, but right now things are picking up, especially because companies want to do things with AI and LLMs, and, well, I happen to have an opinion about that.
Gianluca Fiorelli: Yes. So, resuming your introduction a little bit, and how SEO is treating you lately. From what I understand, better?
Ramon Eijkemans: Well, my first instinct would be to say, "define SEO," but that is a question in itself, one I like discussing. But yes, it's good. I felt the same way when I started in SEO a long time ago. I was lured towards a mystery that I was trying to solve, a puzzle that needed to be done, and an issue to be fixed, which was "How does this work?" I wanted to control this in some way.
There is a new machine in town that I'm trying to comprehend and deal with. And yes, it is definitely back in the stage of it being a hobby and me trying to read everything I can about this, test things, think about it, discuss it, and try it. I love it. So, yes, it's treating me well.
Why SEO Needs Humanities Again
Gianluca Fiorelli: I'm very glad, and obviously, this is a common sensation that not just you or just me, but also many other guests of The Search Session share. Many of the guests who have a story behind their profession say this is very similar to when we started, and there were no playbooks already defined.
So we needed to test everything, to try to understand, to reverse engineer, to somehow interpret and translate what, for instance, Google at the time, and now also all the new players like OpenAI, Claude, and Perplexity are saying when they are saying something, because this is the problem.
Which is ironic, don't you think? Because it's something that is really related to semantics, to clarity, to language, and the people creating all these new things and selling also these new things sometimes fail in clarity and fail in semantics because they use the same word for 300 different things. Don't you agree?
Ramon Eijkemans: Definitely. Clarity, or, as I like to call it, "utility." I think that the way a search engine displays its results has actually made it convenient for us to not be so clear or useful in our content. For example, a listing page of a shop has a title, usually the category, the product category, or something like that.
And then there's a list with links to products, and maybe some attributes of products within it, or maybe some reviews. The text itself is just really, really short. And humans understand it fine, I guess. But LLMs, or language machines—I just call them "language machines" because it's not just LLMs—they need a little bit more context, explanation, and introduction. "Okay, what follows is a review. This is the average review, something like that."
And then they know that the number that follows is actually an average rating, or something like that, or a specification of a product or… context. And that was usually lacking. And I think we're finding out now that we need to up our game a little bit in this regard, which is something I like because it's actually where I come from. I wasn’t coming from a technical or marketing background.
Gianluca Fiorelli: Where are you coming from? Just to give a sense in this context.
Ramon Eijkemans: Before I dived into hypertext. I used to dive into philosophy and history, and the combination of the two. And I used to read more difficult German philosophers or St. Augustine or stuff like that before I read anything about HTML. I heard about HTML.
Gianluca Fiorelli: In this case, we have something in common because, yes, I also come from humanistic studies; let's call it that. This is something that I always want to make clear. I'm not coming from computer science, as many other SEOs do. I'm not even coming from, let's say, the business marketing kinds of studies.
Ramon Eijkemans: Me neither.
Gianluca Fiorelli: All these things I learned in 30-plus years of working, even before starting and pivoting on SEO. I was lucky with my studies, the humanistic ones. Apart from the fact that they were very specific about language and linguistics, so everything is coming very useful right now, they gave me a mental framework, a way of thinking that sometimes, and I'm not saying this at all as a critique of other studies, but sometimes other types of studies may fail.
Like, let's say, data-driven. They need data to do and develop something. And what the data tells them, they do. But we are more speculative. We have the data, and we think about what we can do with the data. We don't necessarily do what the data is suggesting immediately, but what the data may also suggest to do, which is a slight difference.
Ramon Eijkemans: Which is a slight difference. One thing, for example: the same source of data can deliver very different answers depending on the type of question you're asking.
Gianluca Fiorelli: Yes.
Ramon Eijkemans: And depending on who is asking, or why you're asking it. So even within a few words, there might be a whole other world behind it, which is kind of a relationship between the question that is asked and the source or the data it is asked of.
For example, my thesis was about historical archives, and I was already hooked on internet things when I came to my thesis. And it was about using tagging systems to improve archive indexes. I was inspired by a website called del.icio.us. You might still know it.
Gianluca Fiorelli: Yes, I remember. I also remember that it was spammed to the death.
Ramon Eijkemans: Sorry?
Gianluca Fiorelli: It was also spammed to the death. Now the smart guys would call it a classic site for good parasites because you could substantially create things that could be indexed and put in links that you wanted, so it was very easy to spam.
Ramon Eijkemans: Yes.
Gianluca Fiorelli: But, yes, I remember it was a wonderful repository for your own links.
Ramon Eijkemans: Definitely. At the time, OCR scanning wasn't so far advanced as it is now. And, of course, historians deal with medieval texts in some kind of Latin dialect, handwritten in some unreadable handwriting. And I was thinking, "Okay, how can I find this?"
And then I thought, okay, so one historian goes to an archive and says, "Okay, I think this is interesting for subject A." And then another historian sees exactly the same archive piece, data, or whatever you want to call it and says, "I find it interesting for subject P." And then you have two tags, A and P, and then it's findable by both, depending on your question.
So that was actually where I came from. So it was letters, but it was also information architecture, which is how I rolled into SEO.
Gianluca Fiorelli: Yes. In fact, I also remember that in Italy it was called biblioteconomy, a kind of learning subject. It was an exam, in fact. Biblioteconomy. It was a very complicated one.
Ramon Eijkemans: I was one of the few who actually liked that.
Gianluca Fiorelli: Well, it helped in creating that kind of framework that now we use because, even if we always say, using Google words, "Create good content and traffic will come," we know that the difficult part of our job—when we are talking not only to clients but sometimes when we are talking to one another as SEOs—is that it's true that we target humans, in the sense of the final consumer in the case of e-commerce or a potential subscriber for a news website, but we sometimes forget that we are not talking directly with the potential end consumer, as if we were in a physical shop, but through something that is mediating us and them.
And before, there were just Googlebot, Bingbot, so the classic crawlers. Now there are also AI agent bots. And to make it even "simpler," now we are not just talking to human beings, but we also have to talk to synthetic beings, like the agents, when we talk about agentic search, not just about AI models.
That's why maybe all these things that we are stressing a lot about—I know you're stressing too—which are the true meaning of semantics and the true meaning of structured data, must not be confused with schema.
Structured data is another thing. It does not include just the schema, and this is one of the classic misunderstandings in the SEO world. It has become so important, and maybe it's becoming so important, to educate about these topics and how they also reflect on all the other things that we were doing and still need to do as SEOs.
Because of this, we need to have a dialogue pointing to the end user while knowing that this user is going to be reached through a technical medium and that maybe, in the future, we are delegating an AI agent to talk with our potential customer. What are your views on all this?
Finding Common Patterns Across AI Models
Gianluca Fiorelli: Once we had this sort of monolithic SEO, SEO as a synonym of Google SEO, and now everything has exploded into needing to understand how many different players are treating things.
How do you deal with this? Do you go, as sometimes many do, into a sort of rabbit hole trying to understand each model? Or, as in my case, do you try to find the common denominator in order not to get crazy and not have to triple my work just because I have to do this thing for Google and this thing for ChatGPT? I have to harmonize them, etc., etc. How is your process?
Ramon Eijkemans: I think it is similar to programming. If you understand how JSON works, how databases work, and how programming languages work with common patterns they have, like loops and switches and variables and objects and scopes and stuff like that—if you understand the mechanisms behind that—then it doesn't really matter what the name of the language is anymore. You just see variables, you see functions, you see classes, or whatever. In databases, if you know SQL, it doesn't matter if you work with Postgres, BigQuery, or something similar.
So let's just say that I'm looking for something similar in LLMs. LLMs have these technologies behind them, and I'm trying to find what the common or shared mechanisms are behind those technologies.Not just the LLMs that you chat with in a chat interface, but I assume that Google will play quite a big role in this, not just in search engine results pages but also in Gmail, for example, or just on your phone, always on.
It's going to be everywhere, whether we like it or not. I don't like it, by the way, personally, but that's another story. But the mechanisms behind those models are what I'm looking for. So what different types of LLMs are there, or language models, or natural language processing models, or machine learning models, whatever, bidirectional or not?
And what I'm trying to find is, okay, what in general is it that they seem to like? So what do humans like? What do they seem to like? And where is the common ground in this? It's not possible to know all of them.
I do know that every model has its nuances…
Gianluca Fiorelli: Yes, yes.
Ramon Eijkemans: …but I think that kind of depends on the market you're working in. So, where is your target audience interacting on the web? Are they looking more at videos, or are they engaging in social chats, or whatever? And then, well, you will probably see it for yourself in the data, which model is the most dominant.
SERPs as Market Research and Consumer Psychology
Gianluca Fiorelli: Yes. And this makes me think of something you recently shared on LinkedIn: the concept of SERP market research and share of search. So, to make it very simple, visibility is not just your own snippet, for instance, in Google SERP, but also how much you, as a brand, are dominating the SERP in all the features in all the SERPs.
And, as I told you in answering your LinkedIn post this afternoon, that is something I have been exploring for a very, very long time, too.
Before the question, also because sometimes, and again, this is maybe one of the many things that many SEOs tend to fall into, is considering organic search as only the classic search snippet.
Just an anecdote. The other day, I was in an Italian Facebook forum, and there was a guy blaming Google because his search snippet, the search result, was in a SERP where there were just seven separate organic search results. And because of all the SERP features, it was pushed way, way below the fold.
Then I did a search, and I saw, okay, we have one paid position above. Okay. Then there is an AI Overview. I know we all dislike it because the CTR is very awful, but formally, the AI Overview is organic.
Ramon Eijkemans: For now.
Gianluca Fiorelli: Then we have a map. Apart from the advertisement that can also be in the local pack, maps are substantially organic. Then there was a first organic search result; that's organic, we all agree.
Then there was something that is very European, the places sites. So, you know, the carousel presenting alternatives to Google Maps where you can find them. And okay, even if they are very weird and just decided by Google who to present, there is a strong correlation between who is ranking on the first page and who is present in places sites. They are still organic.
Then there is People Also Ask that is organic. So, I’m saying this because many times people forget that organic is not just a search snippet.
Ramon Eijkemans: Yes. You could say that there are several types of organic results within a single SERP.
Advanced Web Ranking’s free SERP Features tool offers a daily pulse check on which SERP Features (from images and videos to AI Overviews and beyond) are most visible across your industry and country.
It’s a simple way to spot where Google is placing emphasis right now, and whether certain features might deserve more of your limited resources.
Explore the free data and seewhich opportunities are emerging in your own SERPs.
Gianluca Fiorelli: Yes.
Ramon Eijkemans: Together with paid. There are several things. One, it's becoming clearer and clearer that to be successful in Google, let’s just call it simple, you need to do more than just write texts and try to rank in the classic 10 blue links.
I remember Renee Bigelow, who said it's classic communications theory that if you want someone to remember your message, you need to repeat it, let's say, 20 to 40 times, and you need to do it in at least three to five different ways.
I'm translating that as being present not just on top keywords but in a market, in all kinds of aspects, in all kinds of alleys and directions within a market. So that is the width of where you need to be present, but also in several, three to five, different ways. So, depending on what your consumers are consuming, if it's about information. So that is one reason.
And the other reason is something I realized: a search engine result page is not just a page with search results. It is a shopping window of that market. I call it a "shopping window" as a metaphor.
Google sees a lot. I wouldn't say that they see the whole internet, but I cannot think of another company that sees as much as Google does. And what they see, they interpret, they calculate, and they present it on their results page, thinking, "Ah, if you are looking for something, maybe you can find the answer here." That is what they do.
So I was thinking, okay, if I look at it through this lens, it's actually a shopping window of an entire market, and whatever is online about this market will find its way, in some way, not everything I know, but a lot, to that SERP.
So what I set out to do is, I didn't really look at it from an SEO point of view, but more from a market research point of view.
So, if this is my data source that shows me what goes on in a market online, how can I analyze it then? And that is what I built. Well, I think it's interesting for SEOs as well, but it's mostly interesting for C-level types who want to know, “What is the market? What are the players? What are people asking, but in bulk? What are their emotions? How visible are we? Does it go up? Does it go down? Stuff like that. And, yes, it's still evolving.
Gianluca Fiorelli: It's interesting because it's very similar to my thinking in the sense that people who are listening to The Search Session have already heard me saying this many, many times: that I, for instance, totally dislike doing classic keyword research. But not only because it's very boring, but because that keyword research usually needs a combination of Google Search Console analysis and third-party tool analysis.
And the third-party tool analysis may have created their database over other people's historical keywords, inputted by their clients in projects, and so on. How valuable is that kind of database? I don't know. Sometimes yes, sometimes quite questionable.
So I say, “Who really knows what people want when they search for something?” And I say "Google," like you. And what is the purpose of Google? I don't know; probably I can also sound like a Google fanboy. But sincerely, if you know me, you know that I'm quite critical of Google too.
So Google has one purpose, always maintained, even when it was saying, "Don't be evil." It wants to serve, to be the source connecting what's on the internet, as you were saying, to what people are searching.
So if you search for something, I can offer you. The fact is that Google has so much data, the classic so-called knowledge base, that it instantly knows what someone is statistically searching for when doing a search. And that's why all the elements painted in a search results page are valuable, because they are based on this knowledge.
So, for instance, the same search menu changes depending on the search because of the implied knowledge that Google has about that search.
Maybe if you are searching for, I don't know, sneakers in the search menu, apart from AI Mode, you can first see videos, images, or news. But if you are searching for a golf situation, probably the first vertical that you're going to see is news.
So this is a very simple example, but with all the features, why is it presenting something in a very transactional query? It also presents maybe part of shopping guides that are somehow related to merchants.
But you can also find very informational search features, like People Also Ask, because maybe Google has understood that people want to buy something, but they need to solve some doubts before taking the final decision to click and go to an e-commerce, or click and go to the specific Google Shopping product page.
So this is the interesting and cool thing about analyzing these kinds of things.
Ramon Eijkemans: One thing I also realized when I was starting with this analysis is that I don't need to reinvent the wheel. Google has done a better job at this than I ever will. So I only need to read what they are already showing within a market.
Gianluca Fiorelli: Yes, the problem then, and maybe this is what you are trying to resolve with this future tool, is how to systematize this kind of observation because, if not, it can be really time-consuming and painful.
Ramon Eijkemans: Yes.
What Makes Scalable SERP Analysis Difficult
Gianluca Fiorelli: It's not something that you can see. And this is fun because SEO is something that we always say takes time to show results.
But then, when it comes to our own job, we want everything now, immediately. I'm thinking, and I already want an Excel file or a spreadsheet with all the things that I'm thinking about done. And yes, this kind of visual analysis can be done. There are ways to make it faster. I'm using them.
But surely this is maybe the struggle of creating a tool, and I'm asking you about the struggle you're having creating the tool and how to really find a workflow in order to make this kind of SERP analysis scalable. What kind of difficulties are you facing?
Is it more a difficulty in defining these things or defining the actual retrieval of the information from the SERPs?
Ramon Eijkemans: Yes, different things. It depends on how big the site is you're doing the analysis for. Currently, I'm testing it for a big website, literally millions of product pages, for example. And it needs to be compared with other competitors that are equally big.
The scale, well, it's doable if you make thoughtful choices about what to use and what not to choose, prioritize basically. But I wouldn't call that the biggest difficulty. The biggest difficulty is making sure that it doesn't fall apart, because there are a lot of moving parts, but that is the IT part of it.
But I'm not vibe coding it; let’s just call it that. Well, parts of it, but anyway.
What is interesting here is that it's, well, I just call it "SERP analysis." That's because that's what it is. But if you do it for 50K SERPs at the same time, then you need some smart things to be able to pull it off. And there are already a lot of smart things out there on GitHub or whatever, answering really specific questions. So I can use that, and I do.
Advanced Web Ranking is built to handle the heavy lifting of large-scale SERP analysis.
With AWR you can instantly map out the SERP features composition for thousands of keywords and dive deep into your project’s SERP snapshots to see exactly what Google's "shopping window" looks like over time, without the IT headaches. Try it free to see it in action.
Sentiment Analysis and the Future of SEO Research
Ramon Eijkemans: So, for example, a really nice thing on SERPs is the People Also Ask questions. They provide a window into the psychology of people. People have a need, an emotion, a fear, a curiosity, or whatever, which is the reason why they ask a certain question. A keyword is an expression of that as well, but usually in a shorter form. So questions provide more context. And I am still assuming that these are real questions people ask. That is an assumption.
From People Also Ask to AI Search
Explore the episode with Mark Williams-Cook to find out how People Also Ask, query fan-out, and follow-up questions reveal user intent and help SEOs read the signals behind AI search.
Gianluca Fiorelli: No, well, usually this is the definition that Google gives about People Also Ask. They say they are somehow, again, from the knowledge base, retrieving the most common questions people ask about that something, let's call it an entity, topic, or whatsoever, that is implied in the keyword in the search query.
So yes, they probably are not literally asked by people, but they are a good, valid representation of questions, for instance.
Ramon Eijkemans: Yes, they need to deduplicate it.
Gianluca Fiorelli: One question is the synthesized version of thousands of questions that are substantially asking the same thing.
Ramon Eijkemans: One thing that is really nice about those People Also Ask elements is that they can provide insights, but you must not use embeddings for that, for example. If you do text analysis, many people are now looking at embeddings and say, "Ah, I can use an embedding because that shows me what the entity space, the vector space of this question is."
But that is about the topic, the subject, something like that. It is not about the emotion behind it. You need reasoning for that. That is a different type of analysis that you need to do.
Some words used within a question could be an expression of a totally different emotion behind that question. And if you are analyzing emotions, you don't need the subject the question is about; you need the emotion that is behind it. You already know the subject, but you need to know if it's positive or negative, if it's fear, if it's a to-do, or whatever.
And that is something interesting I was dealing with just this morning: I need quite a hefty language model for that, with reasoning capabilities, to be able to analyze that at scale.
Gianluca Fiorelli: Yes.
Ramon Eijkemans: The nice thing about this is that it's a fun analysis. It's a fun project to work on, and it's already pretty good.
Gianluca Fiorelli: It's interesting that you are talking about sentiment because sentiment was not, I wouldn't say neglected, but it was substantially not considered because, in classic search, when we think in terms of ranking, sentiment analysis is not applied to search results. But it's implied. It is applied in local search.
Yes, a Google Business Profile page can rank better or worse depending on the implied sentiment Google detects when analyzing the reviews. It was obliged to do that because otherwise businesses with very awful reviews could outrank businesses with very positive reviews just because they did better local SEO.
So for this reason, one of the factors we know is the number of reviews. So they didn't care about having very bad reviews. They cared just about having five thousand reviews. And maybe there was, I don't know, a legit, cool, and very appreciated small business with just 15 reviews, and because of this number, it was always outranked by the big guy.
So in local search, sentiment was already used. But it's interesting that you are introducing the concept of sentiment because I think it is a logical step also for us as SEOs. If we want to optimize our content, our PDPs, our PLPs, our landing pages, our guides, and our articles to be relevant about a topic, it is not just about being relevant to a topic.
It is probably about being relevant to the need, which is not something ruled only by logic or rationality. Sometimes I need something because I have an urgency. Sometimes I need something, for instance, in the informational space, because I'm curious. The need is really driven by a human sentiment, which can be urgency, curiosity, convenience, learning, or something like that. Learning is an aspiration to what? To be better.
And therefore, being able to combine correct market analysis, buyer persona analysis, and audience analysis, it's good to separate everything to be sure. And, let's say, relevance in the more classic sense is correct because when you are creating something, you know that people—if you are not only targeting the topic and the search intent correctly but also targeting the "sentiment" behind the search intent—are going to stay with you. You are going to accumulate all those classic, already classic user signals that Google values as quality signals.
Ramon Eijkemans: Yes, true.
Gianluca Fiorelli: And in more practical terms for, let's say, a stakeholder, a CMO, if you are satisfying your user, this user will probably stay on your website and maybe convert or maybe return to your website and convert in a second moment, and so on and so on. It's an interesting thing that people should talk about.
Ramon Eijkemans: It's a classic example of SEO becoming more important for business because we're doing good research and providing insights into market psychology that other departments within the company should use as well.
Also, I think this type of analysis, which I haven't finalized yet, is really good for helping us present, written down, what it is that a company actually does and doesn't do. So, where can they help consumers? Why shouldn't consumers go there?
It makes it more practical, and it sounds a little bit off to say, "Well, we cannot help you in this case." But that is how chats work, how AI works, how LLMs work, or whatever you want to call it.
Someone is having a conversation that says, "I need a solution to my problem X." And the chatbot knows about this person. They know the projects they are working on and the things they like or dislike. It says, "Ah, given your context, I can say to you, definitely go to that site and not that one," even though they sell the same type of products, for example. And that has to do with the way a company presents itself, the way problems are solved.
One thing, for example, I am European, and I am worried about big tech. And I am moving away from American big tech because I don't trust them. I think no European should.
And I'm looking at software that is not hindered by the Cloud Act. So I am in chat, and I say, "Okay, I need a software solution for this." And then my chat knows, "Ah, privacy is important for you. You don't want your data to be in the US. You don't want the data of your clients to be in the US." Important distinction. “Given that situation of yours, these are the things you should not do, or the companies you should not go to, because these ones are in the US or whatever.” And that is a typical example.
Even though they might be better optimized or stuff like that, my chat interface still isn't recommending them because of my context.
Infusing Structured Data into Natural Language
Gianluca Fiorelli: I want to move just to a related but different topic. So you were saying that knowing the sentiment, knowing what people really want, and so on can help you. We're doing the example of the About Us, for instance, what we do and what we don't do.
And especially the AI models, the large language models, because of their probabilistic nature, they really don't know what, for instance, rhetoric is. What is a rhetorical figure, or what is irony. They can mimic irony, but they are not able to create irony from scratch, for instance.
And so you're saying, returning to the first topic, which was structured data. When we talk about structured data, it's not only a schema. Schema can be important for many things, but there is a lot of talk about schema because of the FAQ deprecation these days, which drives me crazy because it makes it very visible how many people should return to studying somehow.
And I'm not ironic. It's true. They should study what the difference is between structured data and schema, where schema is useful, in what phase of information retrieval, and so on and so on. That schema is not the same as rich results, blah, blah, blah.
But let's talk about structured data. So structured data includes schema, but it also includes, if we think in text or in code, all the classic semantic HTML, for instance. But also things that are included in the same concept of, let's call it very generically, natural writing, natural language.
So you are, for instance, presenting a good example when you are creating content, written content in this case, to pay attention and without making it sound like a bot, instead of using "me," "I," "them," et cetera, use the "name," the "entity." Can you explain this kind of recommendation better?
Ramon Eijkemans: Yes. The keyword here is utility writing…
Gianluca Fiorelli: Yes.
Ramon Eijkemans: …which is basically schema or structured data infused into natural language. The idea is that language machines like LLMs or others don't point to documents anymore in a list, but they synthesize answers based on parts of texts. And those parts have different names for them: propositions, triples, and everything in between.
But let's just say that a small self-contained piece of text—like a sentence or two sentences that are about a subject, and they belong together in a container in HTML with a heading above it or something like that—it should be self-containable. So if it's taken out of context, the page, for example, if it's used in an answer of an LLM, it should still be useful for both the LLM and the user.
A classic thing that goes wrong here is a reference to an object or a subject, like, for instance, a person. So if a sentence starts with, "He does this thing called X," then, if that is taken out of context, both people and LLMs don't know what "he" refers to, especially if we have an article that has named several people. Which one is it? So that is a classic example where, if taken out of context, it should still be clear and useful what that proposition, or triple, or whatever, is about.
This is especially interesting when you try to apply this to, for example, e-commerce listing pages. Because they are very thin in their content. I think this is a prime example of where LLM optimization and search optimization diverge. Although I do think that they can both be helped with this.
But, for example, if you have a product page, not a listing page, so one level deeper, a product page, and you have specifications for a product, a table... It has to be a table, by the way, an HTML table, and not divs. But that is another thing.
Gianluca Fiorelli: Yes, I know because I'm dealing with a B2B client where we are substantially recovering from all the PDF catalogs and all the tables of specifications and talking with his developer to make sure that they are true tables.
Ramon Eijkemans: HTML tables.
Gianluca Fiorelli: All classic tables.
Ramon Eijkemans: Yes, and they should be. So there should be a heading above it or something else, but a heading that says "Specifications" would work just fine, but not just "Specifications." It should say "Specifications of" and then the product name. So subject, relationship, object, or the other way around.
The columns in the table provide context, and then within the rows are the values of the specifications. And then an LLM can understand, "Ah, it's in this column, and the name of that column is that, so that is basically the category where this value is applied to. That is semantics in effect. That is something we should do a lot more with.
Rethinking E-commerce SEO for LLMs
Gianluca Fiorelli: Yes, but—and maintaining your example of a product page and connecting it to when I was talking about B2B—this kind of content can also be sold very fast and quickly to the client.
But let's say a retail e-commerce site, which is really where the text is substantially the minimum possible. Usually, it's just a few notes, like the product name and then some kind of boilerplate content, like you can return the product within three days for free, et cetera, et cetera. And the price, obviously, but the content is very, very small, and instead, you have this great use of visual content like images and video. What to do in this case? Usually, it's very hard to sell to a client, "Okay, we need to put some more text in this context."
I have an answer that I will tell you after, but what would you do?
Ramon Eijkemans: What I would prefer to do, and I know that this is not always possible, is to be there at the table at the moment the product page is designed or when new components are getting designed by UX, by the designer, and then have this input that says, "Okay, we are designing a component that shows the average rating," for example.
We should have a heading above it that says "Average Rating of Product X." Maybe we should apply an ARIA label that introduces a little bit of the value of the average rating that is displayed afterwards.
Maybe sometimes we say, “Okay, we want a small sentence.” One sentence can be enough to explain what it is that folks are seeing here. We can discuss if we will show it immediately, if we show it on desktop, if we show it on mobile, or something like that.
So that would be my preferred route. That would mean that SEOs are literally mingling in the design of the page, but I think it's necessary. That is how I do it right now, by the way.
Taking one step back, if you are announcing things and writing the context down within the words to which the context is applied, that usually means you need more words, and that has consequences for the UX of a page.
People want to get their job done. They're not necessarily there to read a lot of words. And I'm still researching this. One thing I did find out was that there are several types of sentences that one could use, and depending on the first sentence, which I call a situation, I will explain it later, people will actually read or skip the next sentence.
So if the next sentence is longer and a little bit dry with facts, it doesn't matter if people are skipping it anyway, because they know it doesn't apply to them.
One thing I did find out was it's two sides of the same coin, which I call a situation and a closer. And the situation is classic emotional-level copywriting. It's not the same as emotional; it's a situation. So it basically means you're writing down the situation where someone who reads it can say, "Ah, this applies to me." And if it applies to them, then they are willing to read the rest because it applies to them.
The stuff that applies to them can be dry, can be fact-based, can be a little bit longer. That's okay, because they need that information. That, by the way, means that if you have a situation in a heading, there is a discussion there about whether you need to use keywords or not. I would say many times, no.
Gianluca Fiorelli: No.
Ramon Eijkemans: That is an interesting discussion, but I want people to read what comes next. But anyway, eventually it has to be solved in design and persuasive copywriting. And those are the two axes I basically work with right now.
Gianluca Fiorelli: I agree with you. Usually, I would also like to be in the design room, as you were saying, and sometimes I succeed in doing so.
And, for instance, you were saying, just very quickly, “I wouldn't put keywords in the H1." I mean, if it's the name of a product, it must be the name of a product. But what I would suggest to a designer is, just below, for instance, in the visual space of the product name, a ten-word space for a phrase, which is a trigger, the one you say, the situation phrase.
I would put it there so that people can say, "Oh, this is the product I was looking for." And so I can eventually use, correctly and in the very correct sense, the FAQ, or, for instance, if it's flowers that you are selling, explain how to maintain the flowers longer when they arrive home, just these kinds of quick tips, et cetera, and what exactly these flowers are, and so on and so on.
Ramon Eijkemans: Where to buy the flowers and where not to buy them from.
Gianluca Fiorelli: Obviously. But then, because products, and this is correct from the clients' point of view, are very tied to visual shopping, I would improve all the things behind the curtains. So, in the code related to, in this case, alt tags for the images. I would use a good caption for the video, if there is a video, and so on and so on.
Also, to close our one-hour conversation, returning to the very start when we were talking about SERP features, these are also the ways you can optimize your PDP for being visible in all the most interesting and relevant features you want to target with that PDP.
Ramon Eijkemans: As far as I'm concerned, I think product pages and listing pages in e-commerce sites are a prime example of what many in the SEO space have not been doing. Because there are a lot of things that we can do to improve the comprehension of those pages and what is on them. A lot of things that we can do to make sure that machines understand better what is said.
But we haven't done that because we used stuff like incoming links, siloing, and maybe ugly texts below it. I usually call them Wikipedia texts.
Gianluca Fiorelli: Oh, a classic SEO text. And just for our listeners and people watching us, remember, if you are really asked to use this kind of text, beware of the risk of creating so much text that you transform…
Ramon Eijkemans: The nature of the page.
Gianluca Fiorelli: Yes, the intent of the page perceived by Google can pass from pure navigational to pure informational, and it will stop ranking for the queries it should rank for.
Ramon Eijkemans: Yes, and sometimes people ask on LinkedIn, "Okay, what are you doing now differently because of LLMs that you didn't do before in SEO?" Well, my honest answer is that I'm looking more closely at the content on those pages. I already did, but I'm looking more closely because of stuff like this.
And I think this approach helps both LLMs and search engines because they understand better what is actually being said on that page. But I think LLMs especially—because I assume they look less closely at internal PageRank flow and stuff like that on a site: link texts, which is typically something that Google uses to understand the context of a page—are stricter in this, especially in RAG systems.
So, yes, you do rank despite poor content. You're being fetched in a RAG query, and then the LLM tries to find what it is that your page is about, and then there's nothing on your page.
Gianluca Fiorelli: Exactly.
Ramon Eijkemans: Nothing useful. So you should do something about it.
Go Further into Enterprise SEO
Continue the conversation with Martin MacDonald, as he and Gianluca Fiorelli unpack how enterprise SEO is changing in the age of LLMs, from AI search behavior and structured data to the testing frameworks that reveal what truly matters at scale.
Gianluca Fiorelli: And to conclude our conversation, let's stop talking about LLMs and Google and so on, but let's talk about you. You said you come from humanities studies, and one of your passions is history. And I know it's maybe a silly question.
Ramon Eijkemans: By the way, in five minutes, I have to pick up my children from school.
Gianluca Fiorelli: Yes, just a very quick question. What is the historical period that you really like to return to reading about?
Ramon Eijkemans: Middle Ages. Because it's both here, in this exact location, but it feels totally different, and that is weird. And that is why I like it.
Gianluca Fiorelli: Cool. Thank you, Ramon. Thank you for sharing your time and for this wonderful conversation.
Ramon Eijkemans: Thank you. I'm sure we'll talk again.
Gianluca Fiorelli: Sure. And remember, all of you, if you liked this conversation, give it a big thumbs up for this episode. And if you want to be notified about new, wonderful conversations here on The Search Session, remember to subscribe. Thank you and bye-bye.
Podcast Host
Gianluca Fiorelli
With almost 20 years of experience in web marketing, Gianluca Fiorelli is a Strategic and International SEO Consultant who helps businesses improve their visibility and performance on organic search. Gianluca collaborated with clients from various industries and regions, such as Glassdoor, Idealista, Rastreator.com, Outsystems, Chess.com, SIXT Ride, Vegetables by Bayer, Visit California, Gamepix, James Edition and many others.
A very active member of the SEO community, Gianluca daily shares his insights and best practices on SEO, content, Search marketing strategy and the evolution of Search on social media channels such as X, Bluesky and LinkedIn and through the blog on his website: IloveSEO.net.
stay in the loop