SERRA – Digital strategy and launch

SEO Confidential: our exclusive interview with researcher and GEO expert Emilia Gjorgjevska

Written bySEO consultant & founder of SERRA

How do you get found by ChatGPT, AI Overviews and Gemini? The researcher behind SEOntology and WebKnoGraph speaks

Press play and listen to what our interview with Emilia Gjorgjevska is about

There is a question that is changing the way companies think about their online presence: when an artificial intelligence system talks about you, what is it really saying? And above all, why should it say it?

For this episode of SEO Confidential I interviewed one of the most rigorous and influential voices in the field of Generative Engine Optimization (GEO): Emilia Gjorgjevska, Doctoral Researcher at the Technical University of Munich, where she studies knowledge graphs and the mechanisms by which AI interprets, selects and cites information.

Before her academic research, Emilia led GenAI initiatives for companies such as MediaMarktSaturn and WordLift, delivering remarkable, field-proven, award-winning results.

She is one of the minds behind SEOntology, the open-source ontology that models SEO workflows, and WebKnoGraph, the graph-neural-network-based approach to internal linking optimization, which we will discuss at length in this conversation.

The result is an interview packed with information and incredibly concrete. Emilia tells us why optimizing internal links by looking only at PageRank is a mistake, how you actually build a machine-readable brand identity, what to do if ChatGPT or Gemini never cite you, and why the risk of Knowledge Graph Poisoning should be on the agenda of every business owner, not just SEOs.

If you are wondering how to make sure your company gets found, and recommended, by artificial intelligence, this is the interview you have been waiting for. Enjoy the read!

Emilia Gjorgjevska, researcher and GEO expert, interviewed by GEO specialist Roberto Serra
Emilia Gjorgjevska, researcher and GEO expert

Do AI answer engines understand the structure and meaning of your business?

In your paper “WebKnoGraph: GNN-Powered Internal Linking” you argue that, for the first time, it is possible to predict the impact of internal linking before the links are even implemented. How does WebKnoGraph work in concrete terms, and why could this approach change the way SEOs and companies design their site architecture in the era of AI-powered search?

WebKnoGraph works like a true “simulation lab” for internal linking.

Instead of adding new links directly to a website and waiting weeks or months to find out whether they improved things or just created noise, the system first models the site as a graph. Every page becomes a node, every internal link an edge, and this structure is then enriched with semantic embeddings. This way the system understands not only which pages are linked to each other, but also how closely related they are from a topical point of view.

At that point, WebKnoGraph uses GraphSAGE, a graph neural network, to identify possible connections between source pages and target pages. In simple terms, it analyzes the site’s architecture, the flow of authority, the context each page sits in and the semantic similarity between contents, proposing new links that could improve the internal structure.

The most important aspect, however, is not the link recommendation, but the evaluation of those links before implementation.

New links are first tried out in a simulation, without actually modifying the site. This makes it possible to understand whether they help distribute authority better across pages, whether they improve the visibility of some pages without penalizing others, and whether they truly make sense from a semantic point of view. WebKnoGraph, therefore, does not simply say: “Here are some links, trust the AI.” Instead it says: “These are the possible interventions, these are the expected benefits and trade-offs. At this point it will be an expert who decides which ones are truly valid from an editorial and commercial standpoint.”

This, in my view, is exactly the aspect that can change the way SEO professionals and companies design website architecture.

For a long time internal linking has been managed through a combination of intuition, predefined templates, tool suggestions and manual experience. All of these elements remain important. Today, however, large websites have become too complex to be optimized by analyzing one page at a time.

Authority spreads through the entire structure of the site. A single link can benefit one page and, at the same time, slightly penalize another. A technically correct link can turn out to be semantically weak. An easily scalable recommendation can still deliver a poor user experience.

In the era of AI-powered search this becomes even more important. Search is no longer just about ranking the classic blue links: AI systems interpret entities, relationships, topical authority (topical authority) and information structure. A site’s architecture thus becomes an integral part of how machines understand a brand, a domain and its expertise. Internal links no longer serve navigation alone: they are signals of meaning, relevance and relationships between contents.

For me, the real value of WebKnoGraph is that it brings a more scientific approach to an industry that has too often had to work with delayed feedback and unclear attribution. It allows teams to move from an approach like: “We think this internal linking change is useful,” to one based on concrete data: “We simulated the structural and semantic consequences in advance before touching the site.”

This does not replace SEO expertise. On the contrary, it gives experts a much more solid decision-making layer on which to base their choices.

That is why I imagine a future in which SEO will not be fully automated. It will be an activity guided by people and supported by artificial intelligence: machines will generate and evaluate numerous possible interventions on the site’s architecture, while experts will choose the ones that are truly useful for the business, for the content and for the users.

This, in my opinion, is the most promising direction for SEO in the era of AI-powered search.

In the paper you show that the goal is not simply to find new links, but to evaluate their effects on authority, stability and semantic coherence in advance. Why did you choose to measure these three dimensions together, and what does an SEO risk missing by looking only at PageRank or page authority?

We decided to measure all these dimensions together because internal linking is never a one-dimensional problem.

In SEO it is easy to focus exclusively on PageRank or page authority and ask: “Did this page receive more authority after adding new links?”. On a real site, however, authority does not move in isolation. Strengthening one part of the graph can weaken another. Adding many links can improve authority distribution, but also introduce semantic noise. A strategy that works in one particular scenario can also prove unstable if the result depends too heavily on small variations in the surrounding structure.

That is why WebKnoGraph evaluates authority, stability and semantic coherence at the same time.

Authority indicates whether an intervention produces a measurable benefit. Stability measures whether that benefit is reliable or fragile. The balance between gains and losses makes it possible to understand whether you are improving the site as a whole or simply shifting value from one group of pages to another. Semantic coherence, finally, verifies whether the new links still make sense from a topical and editorial point of view.

This approach matters because SEO is not only about optimization, it is also about risk management.

A strategy that maximizes PageRank can look very effective inside a dashboard while at the same time introducing problems into the site’s architecture. It could create an excessive number of links pointing to the most commercially important pages, dilute the meaning of topical clusters, generate unnatural navigation paths, or distribute authority in a way that is technically efficient but of little use to users and to the people who manage the content.

For me this is one of the most important aspects of the entire work: a good internal link is not simply a link that transfers authority. It is a link that transfers authority without compromising the structure, meaning and usability of the website.

If an SEO professional focuses only on PageRank or page authority, they risk optimizing the site the way a machine that only sees numbers would, instead of reasoning like a search system and like a user, both of which interpret relationships and meanings.

In the era of AI-powered search this risk becomes even more evident. AI Search systems do not analyze isolated pages: they interpret relationships, entities, context and topical coherence. If internal linking becomes too aggressive, overly mechanical or semantically weak, the site may even gain authority signals but lose clarity as a knowledge structure.

That is why I believe the future of internal linking is necessarily multi-objective.

We should no longer ask only: “Which link produces the biggest increase in authority?”, but rather: “Which intervention improves authority, remains stable over time, preserves the meaning of the content and can be implemented responsibly by a company?”.

This is exactly the decision-making layer that WebKnoGraph tries to introduce.

One of the most interesting findings of the research is the fact that an algorithm can create very effective internal links, yet experts still make better choices when it comes to semantic coherence. Did you expect this from the start, or was it one of the surprises that emerged during the experiments? What practical implications does it have for those managing a large website? And, in practice, when is it worth relying on AI and when does human intervention remain indispensable?

I would say we expected part of the result, but not all of its nuances.

From the beginning we expected the algorithm to be particularly effective at scale. That is precisely the context in which artificial intelligence and graph learning express their full potential. A large website is too complex for a person to mentally simulate all the possible relationships between source pages and target pages, the flows of authority, the connections between neighboring pages and the effects on the overall structure. So it did not surprise us that the automated approach was able to generate very effective internal linking interventions, especially when the goal was to redistribute authority.

The most interesting aspect, instead, was the trade-off that emerged from the results.

The algorithm proved very good at identifying structural opportunities, while human experts proved superior at preserving semantic coherence and editorial logic. This is where, in my view, the result takes on a deeper meaning. It shows that internal linking is not just a graph optimization problem. It is also a matter of meaning, user experience and business context.

In practice, this means that artificial intelligence should not be seen as a replacement for SEO experts.

AI is, and will continue to be, an excellent first layer of analysis. It can examine the entire site, identify opportunities, rank possible links, highlight recurring patterns and simulate the effects of interventions before they are implemented. It is an enormously valuable contribution because it reduces intuition-based guesswork and allows teams to look at the site’s architecture as an organic system, rather than as a random collection of isolated pages.

Human experience remains indispensable, however, when the question becomes: “Should this link really exist within the site?”.

Only a person can assess whether a link fits correctly into the page template, whether it supports the user’s journey, whether it is consistent with the editorial context, whether it reinforces the brand narrative and whether it looks natural rather than artificially inserted. These are not marginal details: it is precisely these elements that determine whether an optimization can actually be adopted.

For large websites, the practical implication is very clear: you should not choose between artificial intelligence and humans. You should design a workflow that leverages both.

AI should generate and evaluate a large number of possible interventions, because no human could do it with the same speed and at the same scale. Then, experts should analyze, filter and adapt the recommendations, because algorithms do not yet understand business priorities, brand tone, editorial constraints or user intent with the same depth as those who work on the project every day.

In other words, companies should rely on artificial intelligence when they need scalability, pattern detection, simulations and prioritization. They should instead rely on human experience when they need judgment, semantic quality, editorial accountability and final implementation decisions.

For me, this was one of the most interesting outcomes of the entire project, because it moves beyond the simplistic narrative that “artificial intelligence beats humans”.

The more accurate conclusion is a different one: AI enormously expands what SEO teams are able to see, while it is people who decide which interventions truly make sense.

This is the model I believe in for the architecture of large websites: supported by artificial intelligence, guided by experts and validated through simulations before being implemented.

For a business owner who wants to be cited by Gemini, Google AI Overviews or other answer engines, how much does being recognized as an entity in Google’s Knowledge Graph matter? And what are the first practical checks to understand whether your brand truly exists in the eyes of AI or is being treated as mere text on the web?

For a business owner who wants to be cited by Gemini, Google AI Overviews or other AI-powered answer engines, entity recognition is no longer an optional extra. It is one of the foundations of visibility.

I connect this concept directly to our work. WebKnoGraph analyzes the website as a graph made of pages and internal links, while SEOntology interprets it as a semantic system made of entities, queries, structured data, anchor texts, search intents, topical authority and relationships between contents. Together they show that visibility in AI systems does not depend simply on publishing new content: it is essential that machines are able to understand the structure and meaning of your business.

That is why I always explain the difference between a brand as mere text and a brand as an entity.

If your company’s name appears on the pages of your site, but Google and AI systems cannot clearly connect it to an organization, a product category, the people who are part of it, the services offered, the locations, the structured data, the external citations and the topical authority, then that brand is still weak in terms of machine understanding.

It exists as a sequence of characters, but it does not yet fully exist as an entity.

The first practical check, therefore, should not be limited to the question: “Are we ranking on Google?”, but should be: “Do we have a real entity footprint?”.

A company should search for its own name on Google and check whether it is recognized as a distinct organization, whether the context of the search results or any Knowledge Panel is correct, and whether the same information is reported consistently on the website, in social profiles, in directories, on partner pages, in media citations and in structured data.

If this information is fragmented, inconsistent or missing, AI systems receive a weak signal.

The second check concerns the site’s semantic architecture, and this is where SEOntology becomes particularly useful.

The company should verify whether the site clearly represents its main entities: organization, products, services, people, locations, categories, use cases and search intents.

Are these entities connected to each other through internal links? Are the anchor texts descriptive and meaningful? Do the pages support each other, forming coherent topical clusters? Is structured data being used to make these relationships explicit?

If the answer is no, the site has content, but not a true knowledge structure.

The third check concerns internal linking, and this is where WebKnoGraph comes into play.

Internal links do not serve navigation alone. They tell machines which pages matter most, which concepts belong to the same domain and how authority should flow within the site.

With WebKnoGraph we ask ourselves a precise question: if we add or modify certain internal links, can we strengthen authority without compromising semantic coherence?

This is a fundamental aspect, because AI-powered answer engines do not just need well-crafted pages. They need to be able to consider the site trustworthy both structurally and semantically.

The fourth check consists of verifying whether the brand appears in the answer contexts that actually matter.

A company should test prompts such as “the best providers of X”, “alternatives to Y”, “companies that help with Z” or “tools for this use case”. It should then observe whether its brand is cited, ignored or replaced by competitors.

If the brand does not appear, the question should not just be: “Why aren’t we being cited?”.

The more important question is: “Have we built enough structured evidence for AI systems to understand that our brand should appear in this answer?”.

My answer is therefore very clear: being recognized as an entity within Google’s Knowledge Graph, or at the very least being consistently identified as an entity across the web, is extremely important.

The real work, however, does not begin and end with having a Knowledge Panel.

The most demanding part is building a machine-understandable identity: clear entity definitions, consistent structured data, a solid network of internal links, meaningful anchor texts, confirmations from external sources and strong topical authority.

This is exactly where I see the link between SEOntology and WebKnoGraph.

SEOntology defines what AI systems should understand. WebKnoGraph evaluates how well that understanding is supported by the site’s architecture.

In the era of AI-powered search, companies that want to be cited need both: semantic clarity and structural authority.

There is the risk of “Knowledge Graph Poisoning“, meaning the possibility that false information ends up influencing the way AIs describe a company. How real is this for a business, and what checks should it perform regularly to notice in time if ChatGPT, Gemini or other systems are starting to build an incorrect representation of its brand?

Knowledge Graph Poisoning is a very real risk, but I would like to explain it carefully. It does not mean that a single random fake article will immediately change the way ChatGPT or Gemini understand a company. The risk is more insidious.

In computer science terms, it is an adversarial data problem: if AI systems learn, extract information or run verifications based on polluted sources, the presence of repeated false or inconsistent signals can, over time, influence the representation of the brand.

A knowledge graph is fundamentally a machine-understandable map of entities and relationships that stores text but also concepts such as: this company is an organization, it operates in this industry, it has these founders, these products, this headquarters, these competitors, these awards, these controversies, these sources. When this graph is clean, AI systems have a more solid basis for describing the brand correctly. When the graph is polluted, the system may associate the brand with the wrong attributes, the wrong people, the wrong categories or the wrong narratives.

From the standpoint of malicious attacks, poisoning can happen in several ways. Someone can create misleading pages about the company, publish fake listings in directories, manipulate entries in open or Wikidata-style databases, generate low-quality articles repeating the same falsehoods, or create confusion between brands with similar names. The danger is not a single fake source. The real danger lies in the repetition, consistency and findability of the incorrect information. If many weak sources state the same wrong thing, an AI system might treat it as valid confirmation.

For companies, this means that brand visibility is no longer just a marketing problem: it has also become an information security problem. The question goes well beyond the classic: “Are people talking about us?”, and becomes: “Which version of us is being reconstructed by machines?”.

The first routine check should concern entity consistency. Search for the brand on Google, Gemini, ChatGPT, Perplexity and other answer engines, and ask very direct questions: Who is this company? What does it do? Who founded it? Where is it based? What products or services does it offer? Who are its competitors? If the answers are inconsistent across systems, that is a first warning sign.

The second check should concern relationship accuracy. A company must not simply verify that its name appears correctly, but must make sure that the surrounding relationships are accurate. Is the brand associated with the right industry? The right people? The right locations? The right product category? The right parent company or the correct partners? Many entity-related errors are not spelling errors, they are relationship errors.

The third check should be source inspection. When AI systems cite or mention facts about the company, you need to verify where that data comes from. Does it derive from the official website, reliable media, partner pages, structured databases and credible profiles? Or does it come from content-poor directories, copied pages, outdated databases, scraped content or unknown blogs? The quality tier of the sources is crucial, because poisoning usually creeps in through weak evidence.

The fourth check should be prompt variation. Do not limit yourself to testing the question “Who is our brand?”. Also test “contentious” and comparative prompts: “Is this company trustworthy?”, “What are the controversies related to this company?”, “What are the alternatives to this company?”, “Is this company the same as X?”, “Who owns this company?”. These prompts reveal whether the system has a solid understanding of the entity or whether it can easily be thrown into confusion.

The fifth check should concern your semantic and structural hygiene. This is where concepts and projects such as SEOntology and WebKnoGraph connect directly. Initiatives of this kind help engineer reputation management through graph management in the era of AI-powered search.

Protecting your brand name means protecting the machine-understandable representation of your brand. Simple as that.

Many business owners think that structured data only serves to get rich snippets on Google. Today, with AI-powered answer engines, is this role changing? Which markup would you recommend implementing first for a company that wants to increase its visibility in AI-generated answers?

Structured data is moving well beyond the old “rich snippets” logic.

Rich snippets were the most visible benefit and, for that reason, for many years they were the easiest way to explain Schema.org to business owners. In the era of AI-powered answer engines, however, structured data takes on a much more strategic role: it becomes part of the data quality layer that helps machines understand, verify, retrieve and correctly reuse information.

This is exactly the point I developed in my article on SEO data quality published for WordLift.

AI-powered search increases the need for clean, consistent, machine-readable data. When information is fragmented, duplicated, outdated or ambiguous, AI systems struggle much harder to understand what is true, which elements are connected to each other and which source can be considered reliable.

Structured data reduces this ambiguity by offering machines a clearer representation of entities, relationships, attributes and context.

For a company that wants to increase its visibility in AI-generated answers, I would start first of all with the markup that defines the company as an entity.

The Organization schema should be the first step: official name, URL, logo, sameAs links, contact information, founders or key figures where relevant, and the most accurate category for the business.

This is the identity layer.

Before an AI system can cite a company with confidence, it must know who it is and how it connects to the rest of the web.

The second layer consists of the WebSite and WebPage schemas.

These make it possible to define the company site as the canonical source and each strategic page as a specific information object.

For AI-powered search this matters, because answer engines retrieve and combine information from many different pages. The more clearly each page is defined, the easier it will be for machines to understand its role within the overall knowledge structure.

The third layer depends on the business model.

For an e-commerce business, the Product, Offer, ProductVariant, AggregateRating and Review schemas are fundamental, as long as the information is real, visible and compliant with the guidelines.

For service companies, Service, LocalBusiness, ProfessionalService or other specific subtypes are more suitable, describing precisely what the company offers.

For businesses that base their authority on the expertise of their specialists, the Article, BlogPosting, Person and ProfilePage schemas make it possible to connect content to authors, expertise and credibility.

The fourth layer includes BreadcrumbList, ItemList and the connection between entities through internal linking.

This is where structured data integrates with the site’s architecture.

Machines need to understand where a page sits, how it connects to the others and which entities belong to the same context.

This is exactly the point where structured data, SEOntology and WebKnoGraph naturally converge: schema defines the entities, SEOntology organizes the semantic relationships and WebKnoGraph evaluates whether the internal link graph properly supports the flow of authority and semantic coherence.

The implementation order I recommend is the following:

  1. Organization or LocalBusiness.
  2. WebSite and WebPage.
  3. Product, Offer and ProductVariant for e-commerce, or Service and ProfessionalService for service companies.
  4. Person, ProfilePage and Article for expert-driven content.
  5. BreadcrumbList and ItemList for categories, collections and comparison pages.
  6. Review and AggregateRating, exclusively when reviews are authentic, verifiable and visible to users.
  7. FAQ or Q&A only when the page is genuinely designed for that format.

The design principle is very simple: use markup only to describe facts that are already true, visible and useful.

Structured data should be a kind of information contract between the company and the machines trying to understand it.

The role of structured data, therefore, is changing profoundly.

Rich snippets were only the visible part.

AI-powered search makes the invisible layer far more important: entity clarity, data quality, ease of information retrieval and machine trust.

Companies that want to appear in AI-generated answers must start treating structured data as an essential component of their semantic infrastructure.

Today most searches end without the user visiting a website, because the answer comes directly from artificial intelligence. In this scenario, how important does Schema.org structured data become in increasing the chances that a company is recognized as a reliable source and cited by ChatGPT, Gemini or Google AI Overviews?

Schema.org is becoming fundamental because the user journey is increasingly mediated even before the click.

AI-powered answer engines, AI agents, APIs, MCP protocols and retrieval systems need clean, structured data to understand which company exists, what it offers, which information is reliable and why it should appear in a given answer.

In this new information economy, the real competitive advantage lies in the layer of organized semantic data: the clearer the entities, the relationships, the provenance of the information, its freshness and the structured facts, the easier it will be for machines to identify a source and consider it reliable.

This is where structured data becomes an integral part of the visibility infrastructure in the AI era.

The Organization schema defines the company, Product and Offer describe what is being sold, Service identifies the services offered, Person and Article connect expertise to content, BreadcrumbList and ItemList make the site architecture clearer, while sameAs links connect the brand to its distributed identity across the web.

These markups help answer engines move from a fragmented interpretation of individual pages to a much more orderly and coherent representation of the knowledge about the company.

In a future characterized by fewer and fewer clicks and, in some cases, even by experiences with no direct access to websites, companies will need to be understandable even before the visit happens.

The website will continue to host the content, but the real competitive advantage will lie in the layer underneath: structured, interconnected and consistent data that can easily be used by search engines, language models, AI agents and APIs.

Companies that start treating their data as true semantic infrastructure will have a far better chance of being recognized, selected and cited in AI-generated answers.

More and more publishers are starting to block AI crawlers to prevent their content from being used for free. But if a business owner wants their company to be found and recommended by ChatGPT, Gemini or Perplexity, doesn’t blocking these bots risk being an own goal? How can you strike the right balance between protecting your content and gaining visibility?

Blocking AI crawlers can prove counterproductive for a company that wants to be found, recommended and cited by AI-powered answer engines.

If these systems cannot access the content, they may build their answers using information from competitors, aggregator sites, outdated pages or less reliable third-party summaries. This way the company protects its content from direct access, but loses the ability to influence how its industry, its products and its expertise are represented.

The right balance starts with content segmentation.

Companies should distinguish what belongs to the public visibility layer from what constitutes the information assets to protect.

Product information, service descriptions, price ranges, authorship signals, entity data, documentation, case studies, thought leadership content and structured data should generally remain accessible, because they help AI systems understand and cite the company correctly.

Proprietary research, premium reports, paid databases, exclusive datasets and commercially sensitive materials can instead be protected through permissions, licenses, APIs or other controlled access systems.

This is not just a technical issue, but also a matter of ethics and brand governance.

AI adoption requires a balance between innovation and integrity, in which human expertise, quality control, originality and brand protection are the fundamental safeguards.

The same principle applies to crawler access management: companies should define rules that protect the value of their content without making the brand invisible to the systems that now shape information discovery.

From a practical standpoint, I would not recommend indiscriminate crawler blocking as the default strategy for most companies.

I would rather suggest building a true visibility map, establishing which pages should be accessible to search engines and AI systems, which should expose structured data, which resources require a license, which agents or crawlers can access them and which content should instead remain private.

This strategy can be managed through robots.txt, server-side rules, metadata, API access, terms of use and constant monitoring of the citations generated by AI systems.

In the future, the companies that will be rewarded are those able to choose consciously what to make accessible.

Visibility in the AI ecosystem requires a clean, reliable public data layer: accurate entity information, Schema.org markup, a solid network of internal links, semantically clear pages and confirmations from authoritative external sources.

Content protection, instead, should follow a separate strategy dedicated to the most valuable information assets.

The most effective approach is selective transparency: make the data that defines and distinguishes the business machine-readable, while protecting the resources that represent a real competitive advantage and require commercial control.

Artificial intelligences develop a kind of “first impression” of brands already during training. If a company today discovers that ChatGPT, Gemini or Google AI hardly ever cite it, or describe it less favorably than its competitors, what concrete actions can it take to change this perception in future model updates?

A company cannot directly rewrite how an already-trained AI model remembers it. It can, however, influence the information that future models, retrieval systems and AI-powered answer engines will use to learn, crawl, cite sources and build their answers.

If ChatGPT, Gemini, Perplexity or Google AI Overviews rarely cite a brand, or describe it less favorably than its competitors, the answer cannot be entrusted to SEO alone.

What is needed is a cross-functional approach involving PR, engineering, product, data architecture and management.

From a public relations standpoint, the company must strengthen confirmations coming from external sources.

AI systems place greater trust in information that is repeated across credible sources. That is why the brand should earn citations in industry publications, partner sites, market analyses, podcasts, events, directories, review platforms and expert roundups.

The goal is to build a consistent public narrative that clearly explains what the company does, which category it belongs to, what it is known for, why it is credible and how it differs from the alternatives.

Weak, outdated or inconsistent external descriptions often end up becoming the version of the brand reconstructed by machines.

From an engineering standpoint, the site must be easier for AI systems to crawl, interpret and use.

This means clean HTML, indexable strategic pages, high performance, accessible content, clear metadata, correct canonical URLs, properly configured robots.txt rules, a solid network of internal links and constant monitoring of AI crawler access.

If answer engines cannot access the right pages, they will build their answers using competitor content, aggregators or outdated sources.

Engineering therefore has the responsibility of creating the technical bridge between the company’s knowledge and the machines that need to understand it.

From a product standpoint, the company must create pages capable of answering users’ real needs.

If people search for “the best tools for X”, “alternatives to Y”, “software for this use case” or “companies that solve this problem”, the product and the content must offer clear answers through proper positioning, features, use cases, comparisons with alternatives, concrete evidence, pricing logic where appropriate, integrations, customer results and documentation.

AI systems need concrete, reusable evidence that explains where the product sits and why it should be chosen.

From a data and semantics standpoint, the company must build a machine-readable identity.

This includes the Organization, Product, Service, Person, Article, BreadcrumbList, Review and sameAs schemas, where relevant, together with consistent entity definitions both on the website and in external profiles.

The internal link network must also coherently connect the company, products, use cases, categories, authors and supporting evidence, forming a true knowledge structure.

At the management level, finally, the company must develop greater maturity in handling AI-generated visibility, treating it both as a growth factor and as a strategic risk.

Someone must take responsibility for monitoring prompts, citation share, brand sentiment in AI answers, competitor presence, source quality, entity consistency, crawler policies and information correction procedures.

The question management should ask itself is very concrete: when AI systems describe our market, are we there? Are we cited? Are we represented correctly? Are we building today the information assets that future models will use to learn?

The operational plan should start with an initial audit and a clear assignment of responsibilities.

Public relations must strengthen authoritative external signals.

Engineering must make the site technically accessible and well structured.

The product team must create content that answers customers’ real questions.

The SEO and data teams must build the semantic infrastructure.

Management must turn visibility in AI systems into a continuously monitored performance indicator.

The perception future models will have of a company will depend on the public evidence built today. That is why the businesses that start acting now will have a far better chance of being recognized, recommended and described correctly by AI systems.

Your brand must become an AI-readable entity

In the era of AI-powered search, existing online is no longer enough: you need to exist as an entity.

It is the common thread running through every one of Emilia’s answers, from the role of internal linking according to WebKnoGraph to the checks for intercepting Knowledge Graph Poisoning. A company can have excellent content, a fast site, up-to-date articles, but if machines cannot connect that name to a precise organization, with clear products, people, locations and relationships, it still remains invisible in the eyes of the systems that now mediate more and more searches.

This, in my opinion, is where many business owners get the perspective wrong. They keep asking how to climb the rankings, when the right question is a different one: is my company understandable to a machine even before it is understandable to a user?

Structured data, the consistency of external sources, the semantic structure of the site are not technical details to be delegated: they are how you build trust in the eyes of AI today.

The most valuable takeaway from this interview is that AI systems need concrete, reusable evidence, not statements. What matters is not what you say you are, it is what machines can verify about you.

For you, running a business, there is really only one question: are you building the identity that will allow you to be found?

A sincere thank you to Emilia Gjorgjevska for her time and for the precision with which she shared her work with us. Interviews like this one matter a great deal to us: dense with substance, clear in explaining anything-but-simple concepts, and full of insights you can apply right away.

A precious contribution, which I am sure you will appreciate as much as we did. SEO Confidential will be back next week with another illuminating conversation — see you next Monday!

The author

Roberto Serra

SEO consultant & founder of SERRA

Get the best updates on SEO

Already +6,200 professionals subscribed · one email a week, zero spam
✦ Let us start

Market, demand and competitors. Discover what your business could do. With the data in hand.

This is where your digital launch begins.