Frequently Asked Questions (FAQ)

Hallucinations occur when AI systems are forced to fill gaps. AI works probabilistically. If information about a company or an entity is unclear, contradictory, scattered, or non-existent, the model still tries to provide a plausible answer. In doing so, it resorts to similar terms, known patterns, or neighboring entities. This is exactly where misconceptions, mix-ups, or fabricated details arise.

Typical triggers are:

• missing clear definitions
• contradictory self-descriptions across different sources
• marketing language without a verifiable core
• name similarities or unclear distinctions from other entities
• lack of citable, structured facts

Hallucination is not an error in the classical sense, but an emergency mechanism: The model prefers a seemingly coherent answer over an open “I don’t know”. Grounding Pages address exactly this. They reduce the need for interpretation by providing clear, consistent, and provable facts, thereby significantly shrinking the space for hallucinations.

Entities are the uniquely identifiable building blocks with which AI systems organize knowledge and assemble answers. In the AI context, an entity is anything that can be recognized as a distinct “something” and clearly distinguished from others, such as organizations, people, products, services, platforms, places, or clearly defined concepts.

If an entity is vaguely described, has multiple names, or can easily be confused with something else, room for interpretation arises. AI systems then fill these gaps with plausible assumptions. That is exactly where misattributions and hallucinations happen.

The Grounding Page Entity Ontology can help especially with orientation and inspiration. It provides a fixed frame of reference and defines 18 entity classes along with decision rules so that entities can be consistently described and reliably activated by AI systems. Here is the specification: https://groundingpage.com/spec/entity-ontology/.

“Semantically overshadowed” describes the effect that an entity in the language space of AI systems is overlaid by a stronger, more frequent, or dominant meaning. This happens when a term, a name, or a category is strongly occupied by something else. The more dominant meaning draws attention, associations, and answer logic to itself. The weaker entity is thereby displaced, mixed up, or incorrectly categorized.

Typical causes for semantic overshadowing are:

• very well-known brands or companies with similar names
• generic terms that are more strongly occupied than the actual entity
• frequent mentions of a topic in the training and reference space
• lack of clear distinction of the weaker entity

An example: A small company with a generic name can be semantically overshadowed by a large brand, a product category, or an everyday term. The AI then does not categorize it independently but draws on meanings of the dominant entity. Grounding Pages have a stabilizing effect here: They give the entity its own, clearly defined semantic contour so that it does not drown in the meaning noise of stronger terms.

Grounding Pages are less a technology of their own than a concept: a mental framework that helps to define entities clearly and communicate them without marketing fluff.

The comparison to Wikipedia fits well here: In principle, every good Wikipedia article is a Grounding Page. It is not the name that makes it so, but the purpose, implementation, and effect. Wikipedia has been one of the most successful concepts on the web for years because it delivers exactly this clarity: define, distinguish, prove.

In this sense, Grounding Pages can be seen as a bridge: They help close the gap between what a brand says about itself and what AI systems should reliably understand.

The difference lies in the target audience and structure:

• SEO Landing Page (for Humans): Intended to sell and convert. It often contains emotional appeals (“marketing fluff”), long introductions, and visual elements that can disrupt the reading flow for machines.
• Grounding Page (for LLMs): Intended to explain and provide facts. It is explicitly optimized for the vector space. That means: High information density, clear structure (perfect for AI “chunking”), no promotional filler words, and logic that anticipates question-answer pairs.

A Grounding Page is built so that algorithms “love” it because they find instantly citable facts there.

A clear no. This is one of the most persistent misunderstandings in AI SEO (GEO). While both classic Google Search and AI answer generation (Grounding) access the same data pool (index), they evaluate content based on completely different criteria:

• Classic Search (Goal: Navigation): Click probability, brand awareness, and user signals matter here. If someone searches for a brand, the homepage ranks #1 because the user wants to navigate there.
• LLM Grounding (Goal: Answer Construction): Semantic similarity, information density, and factual accuracy matter here. The AI isn’t looking for a link to click on, but a text chunk to formulate an answer.

A real-world example: The Amazon homepage ranks #1 on Google. For an AI, however, it is often useless because it consists of dynamic product lists. As a source for an answer, the AI often chooses the “About Us” page, a Wikipedia article, or the Amazon Jobs page with a corporate description, even if these rank far behind in classic search. Why? Because they explain the content semantically better.

No, the approach is fundamentally different. While classic SEO is based on keywords, AI SEO (GEO) places entities at the center. This requires a strategic shift.

• Keywords are search terms optimized for being found.
• Entities are distinct objects (brands, people, products) understood by the AI.

In an AI-generated answer, an entity is either mentioned or not (Mention). It is described with specific attributes (Features). The goal of Grounding Pages is to curate this information. Consequently, it is primarily digital brand management.

There are technical approaches that automatically provide a pure text version (Markdown) of every HTML page. However, this is often not the decisive lever for two reasons:

1. The Technical Misunderstanding: Markdown is not mandatory for modern search engine crawlers. Bots like Googlebot specialize in parsing HTML perfectly. The Google Grounding API also usually delivers cleaned text to the model, not raw code.
2. The Content Problem (Garbage In, Garbage Out): If a classic marketing website (“We are innovation leaders...”) is automatically converted to Markdown, the model receives technically clean text, but contextually it remains “marketing fluff”.

Conclusion: A pure format change (HTML to Markdown) does not turn a promotional text into a structured dataset. Grounding Pages focus on data quality (content).

Grounding Pages and Schema.org are not opposites, but complementary partners. You can imagine them as the skeleton and the flesh of information.

• Schema.org (The Skeleton): Perfect for hard, isolated facts (“Price: $50”, “Open: 09:00”). However, it is rigid and can hardly represent complex relationships or nuances of corporate culture.
• Grounding Pages (The Flesh & Context): Provide the semantic context the LLM needs to formulate fluent answers. LLMs work with vectors and need articulated text to understand relationships.

Conclusion: The ideal Grounding Page contains both: perfectly structured text for the AI’s vector understanding and valid Schema.org markup for factual precision.

You can imagine the relationship like the difference between a signpost and the actual destination, where it is unclear if the signpost is even noticed.

• llms.txt (The Uncertain Signpost): A proposal from the developer community to show AI crawlers important content. However, it is not (yet) an official standard and is currently ignored by many large models. Furthermore, it only serves navigation, not information.
• Grounding Page (The Secure Destination): This is the URL where the actual information lies. It works independently of helper files via regular crawling processes.

Conclusion: Creating an llms.txt doesn’t hurt, but you should not rely on it. A Grounding Page, however, works immediately because it builds on the established infrastructure of the web.

No, it complements it.

• SEO brings you traffic from humans who want to search and click (transactional intent).
• Grounding Pages ensure that you appear as an expert and source in the answers of AI systems like Google AI Overviews, SearchGPT, or Perplexity (informational intent).

In a world where more and more search queries are answered directly by AI (Zero-Click Searches), the Grounding Page becomes a safeguard for your digital visibility.

There are two scenarios where the application is particularly effective:

1. Small Companies & Start-ups (“Cold Start Problem”): Often, hardly any information exists on the web. The Grounding Page acts as the first and often only reliable source to prevent hallucinations.
2. Large Companies undergoing Change (“Update Problem”): In cases of relocation or rebranding, too much old information often exists online. The Grounding Page serves as the current “Source of Truth” to overwrite the old knowledge.

Answer: Not automatically. Orient yourself towards Wikipedia and draw sensible boundaries. A good rule of thumb is: A Grounding Page is worthwhile if the product should be treated as its own entity because AI systems otherwise easily confuse, mix up, or hallucinate details.

Sensible candidates are:

• Products with their own name and clear identity
• Products that are frequently searched for, compared, or recommended
• Product lines with many variants where AI often gets confused
• Products with high risk regarding misinformation, e.g., finance, health, safety
• Products that must be clearly distinguished from other offers

Rather not sensible is if it is only about:

• small feature variations
• short-term campaigns or bundles
• internal product codes without external perception
• very interchangeable single items without their own identity

Wikipedia Principle: Not every chapter gets its own article. Only what is independently relevant, distinguishable, and worthy of its own article.

Practically this means: Build a clear hierarchy. One Grounding Page for the brand or organization, optionally for large product lines or platforms, and individual Product Grounding Pages only where real entity risk or high utility arises.

No, absolutely not. Grounding Pages are not based on keywords or a 1:1 mirroring of the website, but on entities. They should only be created for the absolute core “things” of a company:

• The Organization itself (Brand).
• Key People (CEO, Founders).
• Core Products or Services.
• Specific Technologies or Methods.

For most companies, this amounts to a manageable number of pages (often only 5 to 20).

Yes, that is highly recommended. Every branch is a distinct entity (“Local Business”). Even if they belong to the same brand, they differ in attributes like opening hours, precise location (“Near city center”, “View of the river”), or infrastructure (“Parking available”).

If this information is cleanly separated on individual Grounding Pages, the AI can understand the context of each location and correctly answer questions like “Find a restaurant with a view and parking near the city center”.

No. The term describes a concept, not a mandatory title. It is not decisive what the page is called, but what it does. A Grounding Page fulfills its purpose if it clearly defines an entity, cleanly distinguishes it, and describes it based on facts, without marketing phrases or ambiguities.

The comparison to Wikipedia is helpful here: No Wikipedia article is called "Grounding Page", but it works exactly according to this principle. It defines, categorizes, distinguishes, and proves. The term "Grounding Page" helps primarily humans with orientation and understanding of the concept. For AI systems, implementation, structure, and content clarity count, not the name of the page.

Because it requires a discipline most organizations have never practiced. Describing what they are in factual, verifiable terms, without marketing language, without emotional appeals, without aspirational claims.

Most companies have marketing pages, product descriptions, and About pages. But none of these typically provide what AI systems need: a structured, machine-readable entity definition that clearly states what the entity is, what it does, and how it differs from others.

This is the paradigm shift from classic SEO to AI SEO. Classic SEO optimizes documents for keywords and rankings. AI SEO curates entities for stable, probable, and correct mentions in AI-generated answers. The unit of optimization changes from the keyword to the entity.

The difficulty is not technical. It is conceptual. Companies need to decide: What are our core entities? What is the one-sentence definition for each? What facts are verifiable? What distinguishes us from similar entities? These are questions that many organizations have never formally answered. The Grounding Page Project provides the mental framework and practical guidance to support exactly this task.

This is the core question. And it answers itself.

If a company has never cleanly defined, documented, and communicated its own entities, how should an AI system do it? The model has no privileged access to internal knowledge. It reconstructs from whatever it finds on the web: marketing pages, press releases, job listings, third-party mentions. If the company itself has never articulated a clear, factual definition, the AI is forced to guess. And guessing is exactly what produces hallucinations.

The uncomfortable truth: most hallucinations about companies are not a failure of AI. They are the consequence of a communication gap that existed long before AI. Companies assumed their identity was obvious. It was not, not to machines, and often not even to humans outside the organization.

In the AI era, this task is more urgent than ever. Every AI-generated answer about your entity is based on what you have made findable, structured, and unambiguous. If you have not done this work, the AI fills the void with probability. A Grounding Page is the first step to closing this gap. Not a technology, but the decision to finally define what you are: clearly, verifiably, and for everyone.

Since it is technically a regular URL, there is no mandatory technical requirement. However, two strategic best practices have emerged:

1. The Footer Solution (The Standard): Placement in the footer (similar to Imprint/Privacy). Advantage: The link is available sitewide, signaling high relevance. Label often: “AI Grounding” or “Facts”.
2. The “Wikipedia” Solution (Contextual): If a company operates multiple Grounding Pages, linking directly from the body text (e.g., linking product names) is recommended to strengthen the semantic connection.

An example of successful integration can be found in the SKIDATA AI Grounding Hub.

Grounding Pages accompany a domain move linguistically and semantically so that AI systems understand correctly: Only the domain changes, not the entity. (Like: Raider is now called Twix. New name, same thing).

Important here: AI systems are not classic web crawlers. They do not follow 301 redirects as reliably as Googlebot does, and they do not automatically reconstruct the context from redirect chains. If continuity is not explicitly described, a move can look to AI like the disappearance of an entity and the appearance of a new one. A Grounding Page reduces this leeway by clearly describing the continuity of the entity and cleanly linking the old and new domains.

In principle, yes. The algorithm doesn’t care if the header says “Grounding Page” or “About Us”. In fact, well-structured “About Us” pages are currently among the most frequent sources for LLM answers. A Grounding Page is essentially the evolution of the classic “About Us” page – consistently optimized for machine-readable branding.

In practice, however, there are two compelling reasons why a separate Grounding strategy often works better:

1. The Stakeholder Conflict (Marketing vs. Machine): The “About Us” page often belongs to Marketing or HR. Their goal is storytelling, emotion, and employer branding (“We love what we do...”). This “marketing fluff” interferes with citability. A separate Grounding Page solves this dilemma by offering AI SEOs a “protected space”.
2. The “Mini-Wikipedia” Principle: Companies usually consist of multiple entities (Brand, CEO, Products, Technologies, History). Packing everything onto a single page dilutes the focus. The Grounding Page concept envisages creating a specific page for each important entity. This way, you build a corporate “Mini-Wikipedia” where the AI finds the perfect, isolated answer for every detailed question.

You can, but they are often not efficient enough. Blog posts are often written narratively (“As we discovered at the conference last week...”).

LLMs often work with a “Needle in the Haystack” problem. The more “hay” (misleading context, filler words, navigation elements) surrounds the “needle” (the fact), the more difficult correct attribution becomes. Grounding Pages remove the hay and present the needles to the AI on a silver platter. This massively increases the chance that your content is selected as a source for the answer (the “citation”).

In theory, yes; in practice, this often leads to a conflict of objectives. A shop homepage needs emotional images and sales-promoting text (“Best in Class”) to persuade humans. An AI needs sober facts and high text density.

Trying to do both on one page often harms the conversion rate (human) or semantic clarity (machine). Separating into Marketing Pages and Grounding Pages resolves this conflict.

A regular review cycle is recommended, ideally every 3 to 6 months, even if facts haven’t changed.

• The “Freshness” Bonus: LLMs prefer current information. A timestamp that is years old can be interpreted as a signal for obsolete knowledge.
• Re-Verification: It suffices to validate the facts and update the date note (e.g., “Last checked on...”) or the dateModified. This actively signals to the AI: “This information is current and confirmed.”

No. A Grounding Page works on the interpretation level, not on the competition level.

It helps AI systems to perceive a brand correctly and categorize it cleanly, instead of filling gaps with assumptions or hallucinating details. Thus, misallocations, mix-ups, and false attributions are reduced. The market position itself continues to arise from real factors such as product, demand, brand, and usage. A Grounding Page does not change these, it stabilizes the correct representation in AI answers.

No. A Grounding Page would not work if one tried to use it for spam.

Grounding Pages are designed to provide logically coherent and verifiable information. Precisely for this reason, they are not a suitable means for manipulation or artificial inflation. Implausible, exaggerated, or contradictory statements would not be stably adopted by AI systems, but filtered out or ignored. A Grounding Page only takes effect if it describes real conditions correctly. It is not a lever for distorting market position, but a mechanism against hallucination, misinterpretation, and semantic noise.

Typically much faster than classic SEO measures. Often, first effects are visible just a few days after indexing.

Reason: AI Grounding is based on content fit (vector similarity) at the moment of the query, not on user signals collected over years (backlinks/clicks). Once the bot has the page, it can be used.

Effectiveness can be demonstrated on four levels:

1. The Technical Proof: Documentation of enterprise search systems (Google Vertex AI) shows that algorithms are optimized exactly for semantic relevance and fact density – which is exactly what Grounding Pages deliver.
2. The Empirical Ranking Paradox: AI answers often cite sources (PDFs, subpages) that rank poorly in classic SEO. This proves that a different selection mechanism is at work.
3. Practical Experience: Companies report that AI models suddenly understand specific concepts that were previously hallucinated.
4. The “Live Proof” (The Project Itself): Although the term “Grounding Page” was unknown to AIs before the project start (Nov 2025), today all major systems (ChatGPT, Gemini, Claude) can explain the term precisely – solely by crawling this documentation.

Fundamentally, the concept works on both levels, but with different time horizons.

• Grounding (Live Retrieval): This is the primary focus. Since models are rarely retrained, live grounding is the fast lever. An effect is often visible after just a few days.
• Model Training (Pre-Training): Grounding Pages are ideal training material for future models (e.g., next generations of ChatGPT and Gemini). However, this is a long-term side effect.

No – and they don’t have to.

Grounding Pages are based on universal internet standards: HTML, Text, and URLs. The concept works on the principle of “Permissionless Innovation”. AI models “starve” for structured data. We optimize the input so that systems process it voluntarily and preferentially because it is in the mathematical interest of their algorithms.

For those who want to dive deeper, we recommend the following sources:

• Google Cloud Documentation: Grounding with Google Search
• Metehan.ai: Reverse Engineering Google AI Mode
• Waikay.io: Importance of About Page for your Digital Identity
• Arxiv.org: Scientific Research on RAG Systems

The Grounding Page Standard does not require separate infrastructure. It is not a technology, but a mental framework: the deliberate decision to define your own entities clearly, verifiably, and in a machine-readable way. The concrete implementation remains flexible.

You can restructure existing pages, use your About page, or create a new page type. There is no mandatory URL pattern, no required CMS, no special hosting. The standard is based on regular HTML, the most universally accepted technical standard of the internet.

What matters is the result: a clear entity definition with structured facts, not a particular infrastructure setup.

Yes, you can. The standard is format-flexible. If your About page delivers a clear entity definition, structured facts, and machine-readable markup, it works as a Grounding Page.

In practice, however, this approach runs into limits:

• The multi-entity problem: Most companies need to curate more than one entity (brand, products, key people, technologies). For grounding to work effectively, the entity name must appear in the page title, H1, and lead definition. Just like in classic SEO, a single About page cannot fulfill this for multiple entities.
• The stakeholder conflict: The About page typically belongs to Marketing, HR, and Communications. Their goal is storytelling and employer branding. Agreeing on fact-first formulations across these stakeholders can be very difficult. A compromise that satisfies everyone often achieves neither marketing impact nor citability.

A dedicated page type can serve as a protected space that avoids these conflicts. But it is not a requirement. If the conflicts don't exist, the About page works fine.

Grounding Pages are not built "only for LLMs". They are built for humans and machines, the same way Wikipedia articles are.

Think about it: Every well-written Wikipedia article is, in effect, a Grounding Page. It defines an entity clearly, provides structured facts, cites sources, and distinguishes the subject from similar concepts. And Wikipedia is one of the most successful projects on the internet. Precisely because humans and search engines love factual, citable content.

The difference to a marketing page is not the audience, but the intent: descriptive and citable instead of persuasive and sales-driven. A Grounding Page is a factual reference document. It serves anyone, human or machine, who needs reliable information about the entity.

This is a valid concern, and the standard addresses it directly with the Human Notice. A UX element that gives human visitors immediate context.

But the deeper answer is: factual pages can be very useful for humans. Wikipedia articles are among the most-visited pages on the internet. Press kits, fact sheets, and investor relations pages all follow the same principle: structured facts for anyone who needs them.

The Human Notice is recommended for pages that are purely fact-oriented and might confuse visitors expecting marketing content. But many Grounding Pages can be written in a way that is naturally useful to both humans and machines. Informative, well-structured, and genuinely helpful.

That can absolutely work if the existing pages have no competing objectives. The standard does not mandate separate pages.

In reality, however, existing pages usually serve legitimate marketing goals. They use emotional language, benefit-driven headlines ("The innovative solution for..."), and conversion-focused layouts. These are the right choices for marketing, but they actively work against citability and machine-readable fact extraction.

The core challenge: for effective grounding, the entity name needs to be in the title, H1, and lead paragraph. The definition must be factual, not promotional. These requirements often conflict directly with the marketing style of existing pages.

A useful parallel: Companies maintain a press kit alongside their product landing pages. The press kit doesn't replace the landing page. It serves a different purpose with different rules. A Grounding Page works the same way: a factual companion to marketing content, not a replacement for it.

The word "standard" is often misunderstood here. The Grounding Page Standard is not a technical protocol like HTTP or Schema.org that requires adoption by vendors. It is a mental framework: the deliberate decision to define entities clearly, verifiably, and in a machine-readable way. It is an attitude of conceptual discipline and factual clarity.

The concrete implementation uses HTML, the most widely accepted standard of the internet. LLMs do not need to "accept" anything. During grounding, they read web pages and demonstrably prefer clear, structured, source-rich content. This is not speculation. It is documented in GEO research and confirmed by practical results.

A useful parallel: SEO was never officially "accepted" by Google either. There is no Google certification for SEO. Yet it works. Because it aligns with how the system operates. Grounding Pages follow the same logic: they align with how AI retrieval systems select and process sources.