You’re listening to The Search Session, and I’m Gianluca Fiorelli, your host. Today, my guest is Beatrice Gamba, an AI-powered digital strategist and innovation leader from Italy. We talk about semantics, entities, ontologies, knowledge graphs, and how AI is reshaping the way brands structure and communicate information online.
Our conversation explores:
Early exposure to AI-powered SEO: how working with knowledge graphs, ontologies, and semantic technologies shaped her passion for structuring information and connecting users with the right content.
Bridging semantic theory and business practice: making knowledge graphs and ontologies understandable and actionable for enterprise decision-making.
Making knowledge graphs understandable: the challenge of showing clients how they help organize data into a shared semantic layer, reduce fragmentation, and support internal decision-making.
Establishing a brand’s ontological core: using knowledge graphs and structured data to define and reinforce a brand’s core entities, helping search engines and AI clearly understand the brand and its authority.
Aligning structured data and content: why ensuring consistent signals for search engines and AI helps organizations turn raw data into meaningful, usable knowledge.
Semantic alchemy: grounding AI in ontologies and knowledge graphs to transform raw data into scalable, domain-expert content generation and operational efficiency.
Grounding AI in industry regulations: using ontologies and knowledge graphs to embed policies, guide AI outputs, and streamline multi-team review processes.
Managing AI hallucinations: grounding models with knowledge graphs, feedback loops, and human-in-the-loop governance to ensure reliable, high-value outputs.
Balancing human experience and machine interpretability: the key is making websites understandable for agents while preserving meaningful, engaging experiences for human visitors.
Tune in and join us for this pleasant conversation.
Beatrice Gamba
Head of Innovation at WordLift
With over a decade of experience in digital strategy, Beatrice Gamba specializes in semantic SEO, structured data, and AI-powered search, helping global brands structure their information so search engines and AI systems can better understand and surface it.
Beatrice joined the WordLift team in 2021, an AI-powered SEO platform focused on semantic technologies, knowledge graphs, and structured data.
She is also a frequent speaker at international conferences such as the Knowledge Graph Conference and Connected Data London, and writes about AI, semantic search, and knowledge graphs on her blog.
Transcript
Gianluca Fiorelli: Hi, I am Gianluca Fiorelli, and welcome back to The Search Session. Today we have a very, very interesting guest. She's a person who works with our friends at WordLift and collaborates with Lazarina Stoy, a recent guest on The Search Session for the MLforSEO Academy.
She's one of the most interesting voices lately in everything related to semantics, entity search, agentic search, and so on. She is Italian, lives in Rome, and just before the start of this episode, she was telling me how, like many Italians, she is a first-generation Roman, with all the family coming from other parts of Italy, as it was in my case. I'm a first-generation Milanese. The guest today is Beatrice Gamba. Hi, Beatrice, how are you doing?
Beatrice Gamba: Hi. I am very good, actually. Thanks for asking.
Gianluca Fiorelli: Cool. It's a real pleasure for me to meet you, actually, for the first time, because we’ve probably never met in real life.
Beatrice Gamba: I think we did. You were...
Gianluca Fiorelli: When I visited WordLift a couple of years ago?
Beatrice Gamba: Yes. And you were a guest at an SEO meetup that we organized.
Gianluca Fiorelli: Yes, yes.
Beatrice Gamba: I arrived late because I had work meetings. But I remember your talk because you talked about action figures, and I remember that.
Gianluca Fiorelli: Yes, it was a nice experience. And let's see. I know that in the future, WordLift is also collaborating on organizing an event of SERP Conf. in Rome.
Beatrice Gamba: Yes.
Gianluca Fiorelli: The Rome edition. So surely we will meet in Rome in the future. So let's start with a classic question I ask all my guests: how is SEO treating you lately?
Beatrice Gamba: Well, I do believe that this is the most interesting time for SEOs. I'm one of the excited ones. There are two different ways of seeing this. There are people who are worried about the need to justify the measurements and about zero-click search. And then there are the people who are excited about this and who want to stay on top of things.
I am among the latter. This is because I am a person who always strives to get informed. I like to study. That's all I've been doing all my life, and it's what suits me best. So I think that everything new that comes into the industry is a means for me to improve. And I guess that in these times, what we must do is constantly learn.
At the end of the day, this is what makes us successful, and it's also what makes our clients successful. When we are informed consultants, we can help better. So, I'm very excited.
Gianluca Fiorelli: Yes, it's clear that you are excited because you are quite vocal and very visible with all the things you are studying, implementing, discovering, and so on. This is something that all of us should be grateful for, because sharing knowledge is maybe the best thing for helping the whole search industry grow together.
Let's say, from a passionate point of view, not from a marketing point of view or a student point of view, but really from something that ignites your passion about everything related to AI, entity search, semantics, and so on. What was the thing that—a few years ago or three years ago, when all these things started to enter the daily life of an SEO—made you say, "Okay, this is something that I love"? What was the thing that ignited your passion for all these topics?
Beatrice Gamba: For instance, I started my SEO career at WordLift. I lived in Berlin for three years, where I worked for Zalando, mainly in operations, so payments and other things that were not quite as exciting. Then I moved back to Rome. I studied the web and SEO practices a little, and I applied to work at WordLift. That's basically where I learned everything.
And of course, it was a privilege for me because WordLift has always been this AI-powered SEO platform and advisory agency. So we started doing hands-on AI from the beginning. I got introduced to these concepts early in my SEO career, 10 years ago now.
What I liked the most about it was the technical side, like learning how knowledge representation, such as knowledge graphs and ontologies, could apply to something natural, which is content and information. We consume loads of information and content every day, sometimes without even noticing or knowing the words that sit behind it.
And to be on the other side, to be the one who actually makes information more useful and connects people who are looking for specific information with the right information. For me, it really felt like this role applied to me in a way that I could also be useful. It sounds philosophical, but for me it is very tied to reality.
Turning Academic Concepts into Practical SEO Workflows
Gianluca Fiorelli: No, and it is. But I think—and I say this with all my honesty, because it is something that I also feel when I try to share things related to ontology, when I want to create a taxonomy, and so on—that these concepts have been, for many years, very high-level, somehow academic.
But then there is always some sort of “uncanny valley," using a term that is somehow on topic, in translating these concepts, which sometimes sound very abstract, into practical daily blueprints to follow, daily workflows, and so on.
How did you personally, also considering that you are teaching at MLforSEO and all these things, evolve from being maybe too academic to something very practical and actionable when explaining the importance, for instance, of creating an ontology and, from there, creating a taxonomy, what it means to create a knowledge graph, and why it is important? How were you able to make this switch in order to make people take you seriously in terms of "let's do the work"?
Beatrice Gamba: Yes, right. This is a good question. I also go and speak at knowledge graph and connected data conferences. In the past few years, what everyone has been saying is how, sometimes, as consultants, we could not even use the words "knowledge graph" or "ontology" before. It was like, okay, let's not mention it. Let's say that we represent knowledge and we provide tools and platforms that help decision-making and content production processes.
Lately, with AI and LLMs, everyone wants to make sure that their processes are grounded in data. So now ontologies and knowledge graphs are cool.
Of course, as I was saying, they are still perceived as academic artifacts because it is hard for companies to accept, even at an intellectual level, the importance of knowledge graphs, for example. They may understand it, but then they fail in an operational way. Sometimes it’s because of how you approach a problem.
When clients come to us, because we do SEO but we also build graphs, ontologies, and taxonomies, we always start by addressing the problem. The way we approach the problem is how we are able to introduce the concepts and involve our clients in the discussion.
A common misconception is that a knowledge graph is only a technical infrastructure. This is wrong because a knowledge graph is something you can actually use to help understand what your website is about and what your company talks about. It is a useful tool when it is not perceived as purely technical.
It's not something that happens immediately or overnight. That's also true. But SEO has never been immediate. So again, it depends on how we want to approach the problem. These tools are really there to simplify, not to complicate. This is also what I try to do in my day-to-day job: make things easier and make processes smoother for companies. Also, because I work mostly with enterprise and bigger companies, and there are sometimes obvious frictions.
Gianluca Fiorelli: Yes, there are frictions, but we were already seeing them even before everything related to semantics, even in the old days of classic SEO.
Explaining Knowledge Graphs with Metaphors and Practical Examples
Gianluca Fiorelli: I don't know if it is your case, but in my case, I noticed that a good way to translate complexity is... I mean, we cannot deny that sometimes explaining these kinds of things is not easy. It’s not like saying, "This is an apple."
But I notice that many times, finding the correct metaphor is a fantastic way to explain things. One I use—I don’t know if you agree with me, or if you use something similar—is this:
Let’s imagine that a bot or an agent is someone entering your library. In your library, where all the books are your products, categorized and organized. The agent is able to see all the books at the same time. Then it starts noticing that this book, the red one about something, could actually be placed closer to another book with a blue cover because it is about other things, but it can be related.
So I explain that creating a knowledge graph is essentially creating these invisible connections and using tools to make these connections more evident.
For instance, using another metaphor, this time from cinema. I remember a movie with Russell Crowe, where he plays a famous mathematician. There is a well-known scene where he is looking at a blackboard, and he starts seeing numbers illuminating.
So I say that what we have to do is make agents able to do the same. When I find this kind of metaphor, it becomes easier to make these concepts understood.
Do you use something similar to what I do?
Beatrice Gamba: Yes, in a way. It’s similar to what you said before. It’s like a snapshot of your organization. When you start breaking down the fact that a company or a website can be divided into smaller but more straightforward concepts, like entities, and using practical examples can help.
One important aspect of knowledge graphs is that they reduce organizational fragmentation. Because when you try to understand the problem a company has, sometimes it is about having multiple e-commerce catalogs, multiple editorial content workflows running at the same time, or scattered product documentation. When you really understand the problem and explain how it can be solved, that’s when the conversation starts to work.
For example, many e-commerce websites we work with have many spreadsheets and catalogs. When we begin working with them, we usually start with data analysis. We gather all the product data from our clients, make sense of that data, and then organize it depending on the needs, into ontologies, knowledge graphs, or taxonomies, depending on the goal of the project.
And then we can really help them. We understand who it serves, what problem it solves, and what concepts are associated with every product, for example. This becomes a reinforcement of the value. A knowledge graph becomes the semantic layer that can also be shared across different departments.
What we do with knowledge graphs is produce internal efficiency before SEO gains become visible, because these kinds of tools are primarily used for internal decision-making.
For example, the last time I went to the Knowledge Graph Conference in New York, I explained that what we do at WordLift is publish the data in the knowledge graph. This is actually a one-of-a-kind use case because most big companies use knowledge graphs only for internal purposes.
When people approach us, they are very curious. They ask, "So you are publishing the data?" Then we explain all the advantages and benefits in terms of visibility and giving LLMs material to reason about your brand and your website. Even then, it is not always immediately apparent to people who have been working in the industry for many years. But it is a very interesting angle. So yes, there are multiple purposes for building these kinds of tools.
Gianluca Fiorelli: Yes, indeed. If we start thinking about building knowledge graphs, you mentioned enterprise e-commerce websites, but I can think of practically every type of industry. For instance, I have a B2B client where almost all the content, until not many years ago, was living in PDFs. So you first have to extract all of that in order to make it visible, and then start connecting all the information and data so it can be understood.
At first, by classic search engines, and now also by AI models. Although I think it is not entirely correct to talk about classic search engines and AI models as separate things, because Google has been using AI for many years, even long before Gemini and similar systems.
But if we talk about working on knowledge graphs, ontologies, and taxonomies, it is quite a titanic work. It is not something simple.
Brand as a System of Entities
Gianluca Fiorelli: So, do you agree with me that maybe the first knowledge graph we should work on is the one related to the brand itself?
Beatrice Gamba: Yes, absolutely. And this is also what we try to make people understand. A brand itself is a system of entities, also because of how the web works. Basically, a brand online is interpreted through relationships. So what is this brand producing? Who is this brand serving?
When we work with knowledge graphs or entity SEO, we are basically formalizing these relationships explicitly for Google or LLMs. In this way, we are also reinforcing the authority of the brand.
Today, many companies that approach us want to gain authority over their brand, and they are concerned that Google may not interpret their brand in the correct way. The question becomes, "How do I maintain sovereignty over my brand on a web that takes bits of information from many places?”
In this case, having proper structured data and having all your entities clearly defined really helps. Of course, we are not hallucination-free, and with all the user-generated content being picked up, sometimes primarily by LLMs, it is not easy. But if you are delivering and structuring information in the right way, in an organized way, then it becomes easier and less of a black box for the machine to understand. There should be no ambiguity about your brand.
At the same time, when you want to describe your brand with data, Tony Seale, who is a really great guy working in the knowledge graph industry for many years, described it as the ontological core, which are the main entities that we want to focus on.
This has always been true in the SEO industry. Like, let’s define and stay on these guardrails. Let’s not talk about everything. Let’s focus on what we are best at and let’s nail our industry, not other industries. And this can be done with data.
Brand sovereignty is not a feeling, it needs to be measured.
Advanced Web Ranking's AI Brand Visibility tracks how ChatGPT, Gemini, Claude, and Perplexity actually perceive your brand: whether they mention it, link to it, and associate it with the right topics.
You can monitor brand mentions and topic association across all four major LLMs from a single dashboard, which is the closest thing to an audit of how the machines have decided to interpret you.
If what comes back does not match your ontological core, you now have a baseline to work from. Try AWR free and audit your brand's AI presence today.
Gianluca Fiorelli: Yes.
Beatrice Gamba: So, defining the core entities in a very organized and thorough way. Sometimes companies approach structured data in a poor manner. They just add the minimum number of properties to have validated markup. This is not very useful. Sometimes it also costs a lot of money because structured data can be expensive in terms of production for technical and development departments. Not everyone is good at implementing it.
There are also tools that only produce very basic implementations that are not useful. So it is better to have fewer entities that are better related and defined than having structured data everywhere that is just a carbon copy across all the pages.
Gianluca Fiorelli: I think this is a classic problem. It can affect big enterprise companies, but it is even more evident in small- to medium-sized companies. Usually, these companies rely on the same WordPress, and WordPress often means using plugins. Many plugins treat schema as rich results, not as structured data.
So you often end up with a very basic implementation of structured data, as you were saying. That would be fine if it were implemented correctly, but sometimes it is not. And it becomes very difficult to correct or improve it.
For instance, and not because I’m against the Yoast SEO plugin, which does a relatively good job. But if you want something more advanced, you are often forced to understand how to disable the function for designing the schema used by the Yoast SEO plugin.
Beatrice Gamba: Yes.
Gianluca Fiorelli: Then, of course, small companies start thinking, "If this is so important, maybe I need to hire in-house or at least consult a developer alongside my SEO. And then the question becomes, “Does my SEO actually know these things?” and so on.
Orchestrating Data and Content: No Mixed Signals
Gianluca Fiorelli: But I have another question because we are talking a lot about data. How important is it that data and the narrative—meaning the written content, the images used on the website, the videos, and even audio, everything, which is usually defined as content, but not as data; let’s say unstructured content and the data as structured content—are well orchestrated and not contradicting each other?
Beatrice Gamba: If we are giving Google and other AI systems mixed signals, it is not a good sign. This is also why almost everyone in SEO starts with an audit. Every project has always been started with an audit.
We look at what is actually on the page. Do we have a main keyword? Does all the content on the page support that main keyword, or are we talking about something else?
Having a knowledge graph and working with data helps in understanding what is right and what is wrong. Having that snapshot makes you understand better how to do your work.
Also, when working with clients, when they come back to us and say, "Because last time you mentioned this, we know there is this other issue on these pages; can you help us with it?” When we are fostering these notions and passing them on to our clients, that is also how I see that we are winning the client.
It is very important to be able to upskill the clients and the people we work for. As you said, not everyone is proficient at data or at working with data, and that is not necessarily required. But teaching the mindset is something we can do.
Most organizations are not even lacking data. What they lack is structured meaning. There are enormous datasets that contain informational value, but sometimes this data is unusable for AI systems because of inconsistencies among the attributes or because of naming conventions that are not fixed.
As consultants, we can do a lot. We can open discussions with all the stakeholders and explain that these are the inconsistencies we have noticed. For example, we should use a single name for a specific product, but these other names are clashing.
When we work together with clients, we can make a difference, especially when the client has the mindset and the space to welcome this kind of structure. Not as a mere technical infrastructure, but as something that becomes an asset. Because these technologies are real assets for companies. It is not just something that we do for structured data or content production. At WordLift, for example, we also use the knowledge graph to ground the content generation process.
I really think that the role of the SEO and AI consultant is extremely important. How passionate we are and how well we are able to convey both the importance and the criticalities within every team and every company.
Gianluca Fiorelli: Yes, and I think you touched on something important. When we talk about data, it is not 100% correct to associate “data” as the synonym for “structured data." Data is not just structured data.
Semantic Alchemy: The Five Phases
Gianluca Fiorelli: In fact, a knowledge graph may use structured data as we know it as a way to conform the structural data, but there are many other elements involved. Because if structural data was as Google thinks about it, then RDF, which was deprecated by Google, won’t be super fundamental for knowledge graphs.
Another thing I want to move to a little further is something you mentioned last year at the Knowledge Graph Conference, I think. You talked about semantic alchemy, which is a wonderful combination of words. What did you really mean by semantic alchemy?
Beatrice Gamba: Yes. Basically, that is a description of what we actually do. As I said before, there are some gaps that appear when data transformation is treated as a knowledge graph engineering process, when in reality it is also a content and analytics task.
The key insight is that AI performance is not necessarily constrained by the capability of a model but by its semantic grounding. Language models generate better outputs when they are anchored to structured representations of reality that are guided. Because these models already have all the data they need. But the data is what they are trained with. However, if we support them by clearly structuring our reality, it makes their work easier.
So I broke down the data analysis process we use into five phases of the alchemy. The first phase is preparation. This is where we clean and normalize the data. The main outcome is the reliability of the data. We clean the data and ensure that each entity represents a single concept, avoiding inconsistent or duplicated attributes.
Then there is the heating phase, where we create the ontology. This is the stage where we create a shared meaning within the organization about its data and its content, depending on the focus of the project.
The presentation was about Luxottica, which is one of our main clients at WordLift, and how we took their product and helped them integrate their data into the product management infrastructure. On top of that, we created an ontology for the eyeglasses industry itself.
Then there is the AI activation phase. What do we do with this ontology and this knowledge graph? We build it; it is great, and it helps us understand many things internally, but what do we do it, how do we do the activation with these tools? The outcome of this phase is, let’s say, the scalable intelligence.
We are able to ground AI models in ontologies and knowledge graphs. In this way, the models become sentient and turn into these mega experts in the eyeglasses and eyewear industry. So they know the domain because we broke down the eyeglass space into ontology. At the same time, they know the entire product catalog because we built a product knowledge graph.
When the AI model can analyze both the eyewear ontology and the brand product knowledge graph, together with its own intelligence, it becomes able to produce content and other outputs. At WordLift, this can include internal linking, specific copy for newsletters, FAQs, or whatever you may want to add to your website.
It makes everything really scalable. But what does "scalable" mean, even for big companies? Where is the value? The value is, of course, in discoverability, in the truth you gain by adding this content to your site. But there are also company costs that can be cut. We analyze that when a company generates content with an AGI, so let’s say ChatGPT, with a semi-structured prompt but without a personalized model, it often takes many internal reviews before reaching a good output that is ready for production. So, there are many iterations.
And we know that in big companies, reviewing content is difficult because many departments are involved. There is marketing, legal, and sometimes even business departments reviewing the content. So many stakeholders are involved, and a lot of time is spent in the process.
When the content generation process is tied to ontologies and knowledge graphs, and the process is grounded in data, it takes fewer iterations to reach an optimal output. In this way, you are actually decreasing costs for the company. So it is not only a matter of quality. There is also a clear cost-saving decision behind it.
Discoverability is the word that ties semantic SEO strategy to business outcomes, and it is something you can actually track, although it is not a single number. It spans organic rankings, Google AI Overview citations, appearances inside ChatGPT and Perplexity answers, Google Maps, News, and YouTube.
AWR tracks all of these from one platform, including the AIO Citation Rank, which identifies exactly where within an AI Overview your page is cited, whether as a visible sidebar source or as a snippet reference.
If the semantic alchemy Beatrice describes is working, this is where you see it. Try Advanced Web Ranking free and start connecting your semantic work to measurable visibility.
Gianluca Fiorelli: So, essentially, what you are doing with this kind of workflow, or orchestration—I do not know how to call it exactly—is essentially training your own AI model on your own data.
Instead of giving everything directly to ChatGPT, which might produce something but in a very generic way, you create an AI model that is trained specifically on the brand, the products, and everything related to it: the knowledge graph, the ontology, the taxonomies, and even maybe all the unstructured content are combined so the model can produce everything.
You were saying internal linking. So, stronger internal linking, which leads to better navigation and can translate into better conversion rates. But also more classic agentic features, like a more intelligent chatbot, or even exporting a specific branded GPT to ChatGPT, which from the ChatGPT interface can connect directly with the website.
So instead of making users go to the website, you can interact with them directly through interfaces like GPT, and then guide them toward the site as the final step.
It somehow reminds me of something I wrote about. Maybe not with the same precision or structured approach, but I described it as the “sentient store”. The idea was to create different knowledge graphs, merge them, and then start inferring relationships between them and then create everything from there. Somehow similar, no?
Beatrice Gamba: Yes, and let me tell you, lately we have been working with insurance companies. WordLift is a product, but the innovation hub that I oversee operates like an agency inside WordLift.
We work with all types of industries, and each industry has its own specific needs. Some industries are heavily regulated, especially now in the era of AI. If an insurance company wants to experiment with AI, it also wants to make sure that all the regulations and policies are respected.
Right now, we are also adding policies and regulations related to the industry into the graph. And we can also do that in the ontology so that the model understands there are guardrails and regulations within the industry.
In this way, the model can learn how to behave like a specialized insurance content writer. Of course, it is humanly impossible to know everything, which is why content is usually reviewed by multiple teams. But if we teach the same model how to replicate the revision steps of the different stakeholders and teams, we are cutting even more time from the process.
Gianluca Fiorelli: Yes, because ultimately, for instance, you could create subagents with instructions like "You are the legal team, you are the business team," and so on. In this way, they verify everything at the same time.
I think this becomes even more important for brands that operate in multiple countries. In that case, legal constraints can have a big impact. They also affect classic international SEO issues. For example, how can I hide this product that cannot be sold in India but is available everywhere else?
These are the kinds of technical SEO issues that often appear in international SEO. Many times, when I build something smaller or prototype things, I create a sort of knowledge base for technical SEO. It helps the system understand things like: if you have this legal issue, consider these technical SEO solutions.
Tackling Hallucinations with Human-in-the-Loop
Gianluca Fiorelli: But there is one problem that we always deal with as users: hallucinations. When you create these agentic systems based on knowledge graphs, ontologies, and entities, what kinds of problems still exist?
First, how does this kind of approach reduce the risk of hallucinations? And second, how do you deal with it in practice? Do you create another agent that acts as a “guardian” for every model to assess itself before generating something, checking whether it is hallucinating or not? How do you deal with the hallucination problem?
Beatrice Gamba: Yes, right. So basically, what we do is this. Of course, when we are setting up the process, there will always be information missing. Many companies that approach AI for the first time do not fully know what is needed. And sometimes we do not know either, because we are not working inside the company.
So, usually we start with at least one iteration where we add everything we currently have available to the graph and to the ontology. Then we start generating. What typically happens next is that the client comes back and says, “Okay, I forgot that we also have this other policy paper, this other style guide, or editorial guidelines that should be added.”
So at the beginning, we try to make sure that we collect everything we need to build the boundaries. Then, as we start generating, sometimes everything will pop up. There is no case where the process is completely hallucination-free. Something always comes up. Even when you ground a model with data, it is not 100% hallucination-free. That is not even the goal, because it cannot be.
What we do instead is structure feedback loops for our clients, and we ask them to provide feedback in a way that is useful for the model so that we can go back and fine-tune the model based on that feedback.
This is also how we promote the human-in-the-loop concept. Automation without governance does not work. It almost becomes entropy. We never deliver content that is 100% machine-generated. Usually, a human editor reviews it, depending on the terms of the contract. We also have an agent that reviews it, as I mentioned before, but we always try to keep a human in the loop.
AI systems are very effective at scaling outputs, but they cannot independently guarantee the conceptual correctness of the content. Human oversight, therefore, becomes more than just a safety mechanism. It becomes a knowledge design function.
First, we have the conceptual authority. As I said, we define the boundaries, the entities, and the acceptable relationships. Then we have the contextual judgment. Some distinctions require expertise because not all human expertise can be broken down into data or roles.
There are cases where two entities are equivalent or where terminology changes depending on the region or the audience. Even if we add this information to the graph, it can sometimes still be difficult for a model to fully understand these rules.
Then there is the issue of long-term coherence. Knowledge graphs evolve, and without governance, there can be a proliferation of duplicated entities. Their meanings may start to drift, and this can erode the semantic trust that’s innate in the knowledge graphs.
That is why we have these roles, such as semantic editors, that can bridge the data and the AI role of the model. We design architectures where we have humans who are passed on tasks by AI, and then humans pass on tasks to AI. We are bridging and nesting AI in human workflows, but also vice versa, nesting the human in the AI workflows.
I believe this is the right way to create something that delivers real value. It is not just about AI itself. Companies are investing a lot of money in AI, but if you only claim that you can produce a certain number of words in a few seconds, are you really delivering value? Anyone can do that with ChatGPT.
Gianluca Fiorelli: And we also see the consequences in these last weeks.
The Future of Agentic Search and the Website's Role
Gianluca Fiorelli: I have one last question. We know that we are moving toward—maybe not yet in 2026, but probably it is going to be very mainstream in 2027—a future where agentic search will be the primary thing.
Let’s see how many people will accept buying directly from a search engine page or from a chatbot page. On the other hand, if you think about it, people are already buying from Amazon, and in the end, it is not that different.
Beatrice Gamba: I agree.
Gianluca Fiorelli: But as someone from an older generation of internet users, sometimes I feel that there is a correct and honest passion and enthusiasm for agentic search, with everything being done by agents, with visitors to your website being agents rather than humans.
However, I am not so sure about that. I think this could lead to an extreme view where the actual home of a business, which is still the website, becomes less important because people are not visiting it.
But isn’t that a bit myopic? In the end, people will still land on the website. And if a website is written only in chunks for machines or designed poorly just to be consumed by agents, isn’t there a real risk of neglecting the website experience? Just because everything might start from an AI chat interface.
Beatrice Gamba: Yes. At WordLift, we’ve also been saying that websites are going to be deprecated at some point. Of course, we cannot foresee the future, but are people actually interested in browsing a website? Probably for niches, this will still be true. But if we are talking about e-commerce, at the end of the day, people want to scan for the most affordable pair of shoes that meets all of their requirements.
Of course, it is easier to narrow down the search through an interface than actually doing the mental work of looking for them on Google Search. But at the same time, as you were saying, if we are producing websites that are not human-friendly, if they are just chunks or information that are not useful, then this will discourage people from browsing the website themselves.
I think there should be a midway. I do not think the future of the web will be agent-only. Maybe it will be agent-mediated. We are going to ask AI systems to do product comparisons, synthesize sources or content, recommend solutions, and execute workflows. These are the kinds of tasks that people are, sadly, mainly asking AI to do.
But as you said, when agents are doing the job, there is a new optimization layer that we can call machine interpretability. So it is a matter of making our website interpretable by machines. That does not mean that we have to dumb them down or stop making them interesting for humans anymore.
I think structured data and knowledge graphs are a good way of doing this, because when we work at the metadata layer, we are already conveying this information to machines without the need to downgrade the appearance of our pages.
Gianluca Fiorelli: The human experience, the page experience, let's call it.
Beatrice Gamba: Exactly.
Life Beyond the Knowledge Graph: Running, Reading, and Rome
Gianluca Fiorelli: So, let’s stop talking about ontologies and knowledge graphs; let's try to know something more about you. Before, when we started the conversation, you were saying that you, like many people in our industry, did not start in SEO. Many times we were doing something completely different before.
In your case, you said that you did not even study computer science or information retrieval. You studied economics. I always think that, as we were discussing before, for instance, I come from human studies, and in the end something from what we studied always comes out in our work.
So what are the things that, as a university student of economics, you still bring with you when working on knowledge graphs?
Beatrice Gamba: Right. So actually this is a very interesting question because I did my thesis at university in the economics of culture, so cultural economics. My professor assigned me a very interesting hands-on thesis. It was a practical thesis about Amazon search.
When I graduated, Amazon was not used at all in Italy, for example, but it was widely used in the UK or in the US. So I basically studied the UX of searches, the UX of the search page on Amazon, and how it compares to the long tail theory.
I looked at how long the tail of web search for products is and how it applies to the economic theory of the long tail. The idea is that there will always be a product online, a very specific product that people will search for. So the long tail will never be zero. It will just be a very long tail.
When I started working at WordLift and approached the SEO industry, about six years had passed since my thesis. And I thought, "Actually, I did something that proved useful for my job, but I had completely forgotten about it because I was doing something entirely different before.”
That is a fond memory, and I really enjoyed working on that thesis because it was very interesting.
Gianluca Fiorelli: Yes, yes, I totally agree. And when Beatrice is not thinking about knowledge graphs and ontologies, what does Beatrice like to do?
Beatrice Gamba: I like to run. I run half marathons, not yet a marathon, but I hope I will run one this year. I like to read classics. I mostly read very old books, and sometimes also books about data. And I like to watch a lot of movies. So it’s like a duality. I either stay at home reading books and watching films, or I go out and sweat a lot. There is nothing in the middle.
Gianluca Fiorelli: Well, you are also lucky because you live in a beautiful city.
Beatrice Gamba: Yes.
Gianluca Fiorelli: Rome. I can imagine many places where you can have a good run in Rome.
Beatrice Gamba: Yes, indeed. Especially early in the morning, like on Saturday or Sunday, when people and tourists are still asleep. Then you can go running in the city center and it’s really beautiful.
Gianluca Fiorelli: Yes. Because of the city center of Rome, I strongly suggest to everybody listening and watching us: wake up very early, or go to sleep very late.
Beatrice Gamba: Yes.
Gianluca Fiorelli: You will see the classic Piazza Navona empty. That is when the magic of Rome really hits you.
Beatrice Gamba: Yes, I agree. That is a really good suggestion for everyone who wants to visit. Wake up early and just go.
Gianluca Fiorelli: Okay. Thank you, Beatrice. It was a real pleasure to have you as my guest here on The Search Session. I hope to see you soon in real life. Maybe we can think about doing something like a brainstorming session about everything we talked about with you, with Dawn Anderson, with the same Lazarina, and maybe with a special guest, but obliged not to talk too much, Andrea Volpini.
Check out The Search Session episodes where Gianluca Fiorelli talks with:
Dawn Anderson on why search may be evolving with AI, new layers of SEO, and agentic systems, but the core foundations of information retrieval, semantics, structure, and human understanding still remain unchanged.
Lazarina Stoy on how SEO in the AI era is shifting from keywords to entity-based strategies, where brands structure data and build authority so AI systems and search engines can understand and surface them.
Gianluca Fiorelli: Again, it was a real pleasure to have you here as a guest.
Beatrice Gamba: Thank you, Gianluca. Same for me.
Gianluca Fiorelli: And thanks to all of you. Remember to subscribe to the podcast and ring the bell to be notified about new episodes, as fantastic as this one we just had with Beatrice. 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