1. Mission
Modern AI systems reconstruct facts from probabilities. When information is unclear, models fill the gaps with plausibility. This can produce hallucinations.
The Grounding Page Project defines an open standard for machine-readable brand identity. A Grounding Page acts as a semantic anchor. It provides systems like ChatGPT or Perplexity with a stable factual foundation.
2. Does it work? (Proof of Concept)
We applied this standard to a fresh domain (registered Nov 2025) with almost no backlinks, under defined test conditions.
Result: Under these conditions, the domain was cited as a source in ChatGPT, Perplexity, and Google Gemini within 3 weeks.
3. Addressing Common Concerns
The standard sometimes meets resistance based on incomplete assumptions. Here are the most frequent concerns, and why they often rest on a misunderstanding of what the standard is.
"This is duplicate work"
The standard does not require separate infrastructure. It is a mental framework, not a technology. You can restructure existing pages, use your About page, or create a dedicated page type. In practice, a dedicated page type often turns out to be the more efficient path. Because it avoids stakeholder conflicts between marketing and factual accuracy.
"Pages only for LLMs? No thanks."
Grounding Pages are written for humans and machines, like Wikipedia articles. Wikipedia is one of the most successful internet projects because people and search engines love factual, citable content. The difference to marketing pages is not the audience but the intent: descriptive and citable instead of persuasive.
"No LLM has accepted this."
The "standard" is not a technical protocol like HTTP. It is a mental framework for factual discipline. It uses HTML, a widely accepted standard of the internet. LLMs do not need to "accept" anything. They read web pages during grounding and tend to favor clear, structured content. Just like SEO was never "accepted" by Google, yet it works.
"We'd rather improve existing pages."
That can work if the pages have no competing marketing goals. In practice, existing pages serve legitimate marketing objectives. A compromise often achieves neither marketing impact nor citability. The parallel: companies maintain a press kit alongside product landing pages. Different purpose, different rules.
4. Why This Matters
AI models have structural limitations. Without clear definitions, three typical risk patterns tend to appear:
Hallucinations
When facts are missing, AI systems tend to fill gaps with plausible but potentially incorrect information.
Semantic Drift
Brand identity can blend with similar concepts or entities over time.
Visibility Deficits
Weakly represented entities are more likely to be overlooked or misrepresented.
Scientific Context (GEO Research 2025)
A 2025 arXiv study ("Generative Engine Optimization: How to Dominate AI Search") provides an empirical basis for the underlying approach. The research identifies two relevant factors for visibility in AI systems:
- Websites as APIs: AI agents tend to perform better with structured, machine-readable data. Unstructured marketing content ("fluff") often proves unreliable as a data source for AI.
- AI Source Diversity: AI engines differ strongly in their source sets (domain diversity, freshness, stability). You cannot optimize for each engine individually; you need a single, centralized source of truth.
Grounding Pages operationalize these findings. They provide a practical technical foundation to make brand-owned data accessible to AI models.
Read the full paper on arXiv
The Language Trap (Hidden English Queries)
Many models perform internal retrieval steps in English even when the user prompt is not.
An English Grounding Page makes local entities visible inside global model space.
A 2025 independent analysis of more than 17,000 URLs suggests that AI systems frequently reference clear identity pages such as About pages when interpreting brands. A detailed discussion of this behavior can be found in an external article: External analysis on About pages and AI interpretation .
5. The Standard
Version 1.5 defines the architecture of a stable factual space for AI interpretation.
The Three Core Elements
- Stable Definition: A short, verifiable statement describing what the entity is.
- Clear Distinction: A statement describing what the entity is not.
- Consistent Structure: Same format, same logic, same extractability.
Quality Principles
No adjectives, one fact per sentence, visible timestamps (Created, Updated, Verified).
What's New in v1.5
Version 1.5 strengthens retrieval coverage and fact reliability. The core architecture remains unchanged. All updates apply to content blocks only and have no effect on URL logic, routing, or language configuration.
- Entity Name in Headings: Content-heavy H2 headings include the entity name as a prefix (e.g. "Rankscale: Core Facts"). When AI systems extract a text chunk, the heading alone identifies which entity it belongs to.
- FAQ Section: A new recommended element. Grounding Pages that answer the most common questions about an entity are more likely to be retrieved as a source, not just for the definition question ("What is X?") but also for related queries ("What should I consider when choosing X?").
- Segment Assignment: The lead section now includes an explicit segment classification (e.g. "Rankscale is a tool in the AI Visibility Tools segment"). This strengthens the semantic link between entity and market category.
- Volatile Fact Hygiene: Facts that change frequently (pricing, feature lists, supported systems) should include a date stamp and a link to the primary source. This reduces the risk of AI systems reproducing outdated information.
- Content-Only Scope: Standard updates are limited to content blocks. Routing, canonicals, hreflang, and language logic remain untouched. This makes version upgrades low-risk and predictable.
For the full technical specification of these changes, see the Technical Implementation Guide.
6. How to Create Grounding Pages
Grounding Pages are not hidden metadata. They are real HTML pages under their own URL (e.g., /facts/) and act as authoritative sources.
-
The Page (HTML):
Create a dedicated page. The visible text is the primary source for the model.
Use definition lists (
<dl>) to encode facts. - The Data (JSON-LD): Provide an identical structured representation beneath the visible text.
- The Authority (Footer Link): Link the page prominently in the footer or imprint.
Just like an imprint clarifies legal identity, a Grounding Page clarifies semantic identity. A persistent, site-wide link supports discoverability and structural consistency across crawl cycles. How individual crawlers interpret this signal varies.
7. Examples & Ontology
The Grounding Page Ontology currently defines 18 entity classes (Organization, Product, Person, Tool, Field of Knowledge, etc.). These classes form a reference framework, not a closed classification system. The ontology is designed to be extensible. Its purpose is comparability and structural orientation across implementations, not control.
Prompts activate meaning spaces. Entities tend to become more stable when the model can anchor them in a clear semantic class.
Explore the full ontology and real reference examples in the Facts Directory:
- Organization: GPT Insights
- Standard: Grounding Page Standard
- Field of Knowledge: AI SEO
- Tool or Platform: Rankscale