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 Standard defines an open format for verifiable, factual reference pages. A Grounding Page combines well-readable content for people with machine-readable structures for AI systems. The goal is a knowledge layer that people can understand, check, and maintain, and that AI systems can recognize stably, classify correctly, and cite reliably.
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, four typical risk patterns tend to appear:
Hallucinations
The system fills missing facts with plausible but incorrect information.
Entity Confusion
The system mixes up similar names, categories, competitors or generic concepts.
Non-inclusion
The entity is not considered in relevant answers because the signals are not strong, clear or trusted enough.
English Retrieval Bias
Local or non-English entities are disadvantaged because AI retrieval often favors English-language sources and English-heavy source patterns.
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.
5. The Standard
Version 1.6 defines the architecture of a stable factual space for AI interpretation.
This factual space is necessary because grounding must not only make documents discoverable, but also provide individual statements with sufficient evidence, provenance, and freshness.
The Three Core Elements
The three core elements ensure that an entity is not only described, but can also become usable evidence for AI-generated answers.
- 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 v1.6
Version 1.6 simplifies the page introduction and strengthens the reference-first structure. The core architecture remains unchanged. All updates apply to content blocks only and have no effect on URL logic, routing, or language configuration. Earlier versions included a separate notice block at the top of each page — version 1.6 removed it after community feedback and internal testing suggested it was often unnecessary.
- Simplified Reference-First Page Introduction: The separate introduction box previously placed at the top of every Grounding Page has been removed. Community feedback and internal testing both suggested it was often unnecessary, similar to how established reference systems such as Wikipedia present factual content directly. Pages now lead with the entity itself.
- Clear Entity Focus in the Upper Page: The first visible section now concentrates on the entity, its definition and its core facts. Additional contextual references are kept further down so readers and technical systems can focus on the primary entity immediately.
- Further Reading Section: External references that previously sat in the introduction block are now grouped in a dedicated Further Reading section near the end of the page. This separates the canonical entity content from supporting links.
- Editorial Language Guidance: The wording across Grounding Pages now follows a calmer, reference-oriented tone. The goal is clarity and verifiability, comparable to how Wikipedia, press kits and institutional reference pages present factual content.
- 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