I. The shift most SEO specialists are still underestimating
There’s a moment in the evolution of search engines when the rules change fundamentally. Not gradually, not as an update you can fix with a technical tweak — but as a paradigm shift that requires you to look at the entire discipline differently.
We’re right in the middle of one of those moments.
I’ve spent over twenty years studying how people search, choose and decide online. I’ve lived through the transition from meta keywords to PageRank, from PageRank to E-A-T, from E-A-T to generative AI. Every time, people assumed the foundations would hold. And every time, the foundations turned out to be exactly the part that changed.
In an earlier piece I described how SEO evolved over thirty years — from meta keywords through PageRank to E-E-A-T and generative AI. The common thread throughout that evolution: search engines kept getting better at understanding what people mean, not just what they type.
The next step in that evolution has a name: Generative Engine Optimisation, or GEO.
And the core of GEO is a shift that sounds simple but has far-reaching consequences: from optimising pages to managing entities.
AI models don’t rank pages. They select knowledge.
A note on terminology. You may also come across the term Answer Engine Optimisation (AEO). That term specifically covers optimising for AI answer interfaces — a subset of what GEO addresses. GEO is the broader discipline: all forms of presence in generative AI systems, whether answers, summaries or comparisons. In this piece I use GEO.
II. What is an entity — and why should you care?
In the world of search engines and AI models, an entity is any clearly defined concept, thing, person, organisation or product that is uniquely identifiable.
KBC is an entity. A mortgage is an entity. A savings account is an entity.
The difference from a keyword is fundamental. A keyword is a string of characters — a combination of letters a user types. An entity is a concept with context: it has properties, relationships with other entities, a reputation and a place in a knowledge network.
Suppose someone searches for “KBC bank.” A classic search engine matches those words with pages that contain them. Google’s Knowledge Graph does something different: it recognises KBC as a specific Belgian bancassurer, with products like mortgages and home insurance, relationships with competitors like Belfius and ING, and a history going back to the merger of Kredietbank, CERA and ABB.
A keyword is what someone types. An entity is what a machine understands.
AI language models work the same way — but go one step further. They generate answers based on what they know about entities and their relationships, not based on which page is best optimised for a specific keyword.
The question shifts fundamentally: from “how do I optimise this page for this keyword?” to “how do I make sure this model knows who I am, what I do and why I’m relevant?”
There’s a second dimension that’s at least as important. AI models don’t just understand what you’re asking about — they understand why you’re asking. Is the question informational? Comparative? Transactional? That intent determines which type of source the model cites. GEO helps you not only to be recognised as an entity, but as the right source for the right intent.
III. The difference between SEO and GEO — made concrete
In my post on LLM Visibility, I explained that search engines and AI models use different mechanisms to determine what they show. That difference also translates into how you optimise.
With classic SEO, the logic was relatively linear. You pick a keyword, you optimise a page for that word, you build links and you try to rank higher than the competition.
GEO works fundamentally differently:
| SEO | GEO |
|---|---|
| Keyword optimisation | Entity management |
| Page-focused | Domain-focused |
| Text for search engines | Structure for AI models |
| Measuring rankings | Measuring mentions |
| Backlinks as authority | Citations as trust |
Two pages on the same topic can perform very differently in GEO, even if they’re equally strong from an SEO perspective. The page that wins isn’t the one with the most keywords or the most links. It’s the page that answers a specific question most completely and most clearly, from a source the model recognises as trustworthy.
An example that sharpens the distinction:
Suppose a user asks an AI model: “What’s the difference between a fixed and variable interest rate on a mortgage?”
Poor: “Discover our competitive mortgages with sharp rates. Request a simulation today.”
Good: “With a fixed interest rate, you pay the same amount every month for the entire duration of the loan, regardless of what happens to market rates. That provides certainty but is often slightly more expensive at the time of signing than a variable rate. A variable rate can fall when market rates fall — but can also rise, creating uncertainty about your future monthly payment.”
The second answer answers a question. The first tries to sell. AI models cite sources that answer questions, not sources that promote products.
IV. Three mechanisms — three different strategies
Not all AI visibility works the same way. There are three fundamentally different mechanisms — and the strategy differs per mechanism.
Parametric knowledge is what the model learned during training. This knowledge is baked into the model itself. If you’ve been prominently present in quality online content for years, the model has processed that and stored you as a relevant entity. The downside: you can’t change this quickly. Until a model is retrained — which can take months to years — your presence is frozen in what the model knows. This is a long-term game.
Retrieval-Augmented Generation (RAG) works differently. Models like Perplexity — but also ChatGPT with search enabled and Gemini — actively consult external sources in real time for certain queries. They retrieve current web pages and incorporate them into their answer. This is the mechanism where you can have direct impact now. Good structure, fast load times, clear answers and the right structured data directly increase your chances of being picked up.
Grounding via citations is the third mechanism. Some models explicitly reference the sources they used. Being cited is stronger than being mentioned — the model isn’t just recognising you as relevant but as trustworthy enough to name explicitly. Citation worthiness depends on the combination of authority, structure and specificity of your content.
The practical implication: those who wait for parametric improvement are waiting too long. Those who optimise their RAG visibility today will have a head start tomorrow that competitors won’t start to feel for months.
V. Why this is urgent now
Many organisations still see GEO as something to address in the future. That’s a mistake.
Parametric knowledge has a lag. What you publish today only appears in the next AI model training round — which can take months. Those who start now will have a lead in a year that’s hard to close. Entity advantage compounds: the longer a model recognises you as a reliable source, the more firmly that association becomes embedded.
The competition is also starting to understand what’s happening. GEO is not yet a mainstream discipline, but the early adopters are already there. In information-rich sectors, stronger players are beginning to restructure their content for machine-readability. Those who wait are falling behind in a market where zero-click and zero-visit behaviour is increasing — users who get their answer from AI without ever visiting your website.
Google is accelerating the integration of AI into search. AI Overviews now appear in a large proportion of all searches. The sources Google cites are not the highest-ranked pages. They are the best-structured answers from the most recognisable entities. The shift is from ranking to source selection — and that selection is already happening.
VI. Structured data: from nice-to-have to strategic necessity
If there’s one technical discipline that GEO has elevated from a side note to a core strategy, it’s structured data.
Structured data is code you add to your web pages to explicitly tell machines what the content means. Not just what’s there, but what it is. You do this via schema.org — a standardised library of markup types developed jointly by Google, Microsoft and other tech companies.
You don’t need to programme this yourself. Modern CMS systems and plugins often generate structured data automatically based on the fields you fill in. The barrier is lower than it looks.
A practical example of what FAQPage schema looks like:
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "Wat is het verschil tussen een gereglementeerde en niet-gereglementeerde spaarrekening?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Een gereglementeerde spaarrekening voldoet aan wettelijke voorwaarden en geniet een belastingvrijstelling op de eerste schijf van de interesten. Een niet-gereglementeerde rekening biedt mogelijk een hogere rente maar zonder die fiscale vrijstelling."
}
}
]
}
This tells an AI model not just that there’s a question and answer on the page — it tells it exactly which question and which answer. The stronger your structured data, the smaller the chance a model has to “guess” what you offer. You give the model a reliable foundation of facts — and that directly reduces the risk of hallucinations.
The most important schema.org types for a financial institution:
Organization: who are you as an organisation? Name, address, website, social profiles. This is the foundation of your entity definition.
FinancialProduct / BankAccount / LoanOrCredit: for each financial product. The more specifically you mark this up, the better models will associate you with relevant queries.
FAQPage: one of the most powerful types for AI visibility. Explicit question-and-answer pairs are treated by both Google and AI models as direct answer sources.
HowTo: for process-related content. “How do I open a current account?” structured as HowTo schema increases the chance that the model cites your explanation.
Person: for authors and experts. Who wrote this content? What is their expertise? This is directly linked to E-E-A-T and the credibility of your content in the eyes of AI models.
VII. The Knowledge Graph: how machines form an opinion about you
Beyond structured data on your own site, there’s a second dimension many organisations underestimate: your presence in external knowledge sources.
Google’s Knowledge Graph is a vast network of entities and their mutual relationships, fed by Wikipedia, Wikidata, official company registers, news media and specialised databases. It influences how search engines structure your entity and can indirectly shape how AI systems recognise and contextualise you.
But the Knowledge Graph also builds another type of knowledge: reputation as entity value. If your brand is consistently mentioned in contexts of positive user experiences — on reviews, forums, comparison sites — that strengthens entity value. If the mentions are predominantly negative, that weighs just as heavily in the opposite direction.
And here’s an insight that surprises many marketers: an unlinked mention is in GEO almost as valuable as a linked one. In classic SEO, an unlinked mention was barely relevant. In GEO, a model understands the textual association between a brand and a topic, regardless of whether a hyperlink is attached. An article in De Tijd that mentions your organisation as a mortgage expert — without a link — contributes to your entity value. PR and SEO are therefore no longer separate disciplines. They are one.
Three signals the Knowledge Graph builds about you:
Consistency of information. A model that finds ten different versions of your company name or product description across ten different sources trusts that information less. Consistency across all platforms is not a detail — it’s a fundamental signal.
External mentions. How often are you mentioned in reliable external sources, independent of your own channels? Press articles, industry reports, comparison sites, niche media, public profiles and expert platforms — together they build the external knowledge network that models use to determine whether you’re relevant and trustworthy.
Relationships with other entities. Are you consistently associated with the right concepts and products? Those associations — even without explicit links — help determine how models position you.
VIII. Topical Authority: the perception that you are the source
There’s a concept that plays a more central role in GEO than in classic SEO: topical authority.
Topical authority is the degree to which a domain is seen as the authority on a specific topic. Not for one keyword, but for an entire theme and all related subtopics.
A bank that writes extensively about mortgages — the different types, the interest rates, the tax implications, the application process, the pitfalls, the comparison with renting — builds topical authority for “mortgages in Belgium.” A bank with one product page on mortgages does not.
The difference isn’t just the volume of content, but the depth and coherence. AI models recognise whether a domain genuinely masters a topic or merely has a surface-level presence. And topical authority is self-reinforcing: the more often a model cites you for related questions, the more likely it is to cite you for new questions in the same domain.
Those who aren’t clearly recognisable as an entity are more quickly overshadowed in AI by those who are.
A practical approach:
Identify your core topics. For which five to ten subjects do you want to be the reference? Write about them systematically and in depth — not incidentally.
Cover the full spectrum. For each core topic: the basic definition, the variants, the comparisons, the frequently asked questions, the situational applications, the tax implications, the pitfalls. A model that can use your domain as a source for every conceivable question on a topic will do so consistently.
Link internally with consistency. Your internal link structure tells models which pages belong to the same topic. A well-linked content cluster is stronger than standalone pages with no connection to each other.
Use clear content formats. AI models cite content that’s easy to process: clear definitions, comparison tables, step-by-step explanations, explicit FAQs, concise introductions and a clear author attribution. Structure isn’t a formatting question — it’s a strategic choice.
IX. The most common GEO mistakes
Treating structured data as a one-time checklist. Adding markup isn’t a one-off task. If your products or services change but the markup doesn’t, you’re feeding models incorrect information — which undermines your credibility.
Focusing only on the homepage. AI models are cited based on the page that best answers a specific question. That page is rarely the homepage.
Sales language instead of answers. This is the most common and most impactful mistake. A product page that promotes doesn’t answer a question. A page that explains, compares and nuances — that does.
Treating GEO as a purely technical project. Structured data is technical, but GEO is won or lost in practice in content. Organisations that leave it only to IT miss the core of the discipline.
Not checking what models get wrong about you. AI models can generate incorrect information about your entity — wrong figures, outdated products, inaccurate descriptions. The stronger your structured data, the smaller the chance a model has to guess. But monitoring remains necessary. The only way to correct incorrect information is to publish the right information more consistently and more structurally than the incorrect version.
Not measuring. As I described in my post on LLM Visibility: if you don’t measure whether AI models recognise and cite you, you also don’t know whether your GEO efforts are having any effect.
X. Where do you start? A concrete checklist
Check your entity definition. Search for your organisation’s name in Google. Does a Knowledge Panel appear? Is the information correct? If not, that’s priority one.
Audit your structured data. Use Google’s Rich Results Test to check whether your pages are correctly marked up. Start with your homepage, your about page and your most important product pages.
Rewrite your product pages from sales language to answers. Take your five most important product pages. What specific questions does your target audience ask? Answer those questions explicitly and completely.
Build FAQPage schema on strategic pages. Incorporate the most frequently asked questions about your core products on the relevant pages, marked up with FAQPage schema.
Check consistency across platforms. Are your name, address and contact details consistent on your website, Google Business Profile, LinkedIn and other platforms? Inconsistency undermines your entity definition.
Map your core topics explicitly. For which subjects do you want to be not just present, but dominantly recognisable as a source? Those are the topics you write about systematically and in depth.
Check what models get wrong about you. Test regularly in ChatGPT, Gemini and Perplexity. Not just whether you’re cited, but whether the information is accurate. Are your products, conditions and positioning represented correctly?
XI. An honest conclusion
GEO doesn’t replace SEO. It extends it.
The technical foundations of SEO remain relevant. But they’re no longer sufficient. In a world where AI models generate answers based on what they know about entities, managing your entity has become a strategic priority. Not as a technical project for the IT department, but as a marketing discipline that touches content, PR, technology and data simultaneously.
The organisations that understand this now and act on it are building a knowledge footprint that will generate increasing value over the years ahead.
Your content is no longer just your output. It is the knowledge source from which AI systems learn about you, retrieve you and cite you.
The question is no longer “how do I rank higher?”
The question is “is my brand recognised as the authority it is?”
