Grounding Page: The Fact Sheet AI Systems Quote From

A grounding page is a page on your own website that collects the verifiable facts about a company in one place: legal name, founding, services, prices, the people responsible, evidence. It is aimed less at visitors than at the systems that generate answers about a brand.

The term comes out of the work with language models. Grounding there means tying an answer back to verifiable sources instead of producing it from the model's memory. A grounding page supplies that source for your own company.

Where the Term Comes From

In the development of AI systems, grounding means anchoring a statement in verifiable data. A model that answers without grounding reproduces whatever looked plausible in its training data. A model with grounding pulls in a source and bases the answer on it.

The person who carried the term into the SEO scene is Hanns Kronenberg, Head of SEO at Chefkoch and founder of the analytics project GPT Insights. On 20 November 2025 he published the Grounding Page Standard: a consistent structure for company facts that search engines and AI systems can read out reliably. The current version, v1.6, was last updated on 20 June 2026.

One point of context: that is a standard from the professional community, not a specification from Google, the W3C or ISO. There is no schema a model is obliged to respect, and no markup that guarantees a citation.

The idea carries anyway, because it addresses a real problem.

What the Standard Actually Requires

The standard is more concrete than the term suggests. It describes three components every grounding page should have:

The second point is the unusual one. Hardly any company writes down what it does not offer — and for a model that is precisely the detail that prevents a misfiling.

On top of that come three quality rules that set the tone of the page: no adjectives, one fact per sentence, visible timestamps. The standard distinguishes three dates here — when an entry was created, when it was last changed and when it was last verified. That is more than a status date, because an entry can be unchanged and freshly checked at the same time.

The Machine-Readable Layer

The standard calls for two layers on the same page. Visibly there is the text a person can read. Underneath sits the same information as JSON-LD, marked up to Schema.org, with references to the entities named.

Three technical requirements come with it:

• a dedicated, permanent address, usually under `/facts/`
• definition lists (`<dl>`) for the fact pairs instead of running text
• a link from the footer, so the page is reachable from every other page

So the standard does not replace Schema.org, it uses it. And it is not a knowledge graph: it describes a page, not a database. The current version lists 18 entity classes — organization, person, product, service, tool, method, event and others. The list is meant as orientation, not as a constraint.

How an AI System Gets Company Facts

A grounding page only takes effect once you understand which two sources an AI answer comes from.

The first is training. What a model learned about a company comes from text that was on the web at the time of training. That layer is old, it cannot be corrected, and the smaller the brand, the thinner it is.

The second is retrieval at runtime. ChatGPT, Perplexity, the AI Overviews and Google's AI Mode pull current pages on a question and base the answer on them. This is exactly where a grounding page comes in: it is the page that can be retrieved.

How much that second layer weighs differs by system. Perplexity works with sources almost throughout and displays them. ChatGPT retrieves when the question calls for currency. The AI Overviews draw on the search index. For your own page the consequence is the same either way: it has to be findable, indexable and unambiguous — otherwise it does not exist for the retrieving layer.

Why AI Systems Need a Page Like This

When someone asks an AI system about a company, the answer comes out of what the system finds and what it saw during training. For well-known brands that works passably. For everyone else it turns into a patchwork.

What Happens Without a Grounding Page

• Prices from old articles get served up as current.
• Founding year, location or legal name come from a directory entry nobody ever maintained.
• Two companies with similar names merge into one.
• Services get described that never existed, because they were filled in from the surrounding industry.
• A discontinued brand keeps being named as active.

In our experience, this only surfaces when a prospect mentions it. Until then the wrong detail may have been sitting in answers for months.

Three Failure Modes, Three Causes

These cases look the same in the result — a wrong detail in an answer — but they have different causes. That matters, because only two of them can be fixed with a page of your own.

The most stubborn of the three is entity confusion. It hits B2B companies with generic brand names particularly hard, along with firms that share a name with a larger company from another industry. A grounding page helps there because it binds legal name, legal form, location and services into one statement — a model can then check the attributes together instead of guessing at them one by one.

Why the Homepage Does Not Solve It

Homepages are built for people. They guide, they sell, and they deliberately leave open whatever is meant to create curiosity. What a model needs is the opposite. It needs facts in a form that can be quoted without context.

What Belongs on a Grounding Page

The scope follows what people actually ask about your company. These entries form the core:

The exclusions row gets forgotten often and does more work than most of the others. When a model knows what a provider does not do, it files that provider in the wrong category less often.

How a Grounding Page Should Be Built

For entities and their relationships, the same rule applies as for quotes: what makes sense without context is more likely to get picked up.

One Fact per Statement

Short sentences that stand on their own. Not "We have been in the market for a few years and mainly serve mid-sized clients in the DACH region", but two sentences with one verifiable detail each.

Write Entities Out

Write out company names, personal names, place names and product names instead of replacing them with "we" or "our offering". A model links statements to named entities, not to pronouns.

Tables Instead of Running Text

Structured entries get reproduced correctly more reliably than details buried in subordinate clauses. That applies in particular to prices, scope of service and responsibilities.

Make the Date Visible

A status date makes it possible to spot outdated entries. Without a date, every figure is timeless, including the wrong one. The standard separates created, updated and verified. That separation pays off: a price that has held for two years is not outdated if it is visible that it was confirmed last month.

No Adjectives

The standard bans adjectives, and that is not a question of style. "Leading provider of individual solutions" contains nothing verifiable. A model cannot quote anything from it that it could not equally say about every competitor. In place of the judgement belongs the attribute the judgement would follow from: number of projects, year of founding, number of publishers connected.

That also removes everything that belongs on a homepage: value propositions, superlatives, phrasing in the conditional. A grounding page therefore reads dryly. That is its purpose, not its shortcoming.

Write Down the Relationships

An entity becomes graspable through its relationships: who belongs to the company, which products belong to it, which market it is assigned to, which organizations it is connected with. That is the core of entity SEO — a model links named things to one another, and the more evidenced connections a name carries, the more stably it stands.

In practice that means naming product names together with the company name, people with their role, locations with the country. Not as a list of keywords, but in statements that each hold up on their own.

An Example

What this looks like in practice is on our own grounding page at Facts about GetMentioned, available in German: operator, scope of service per placement, exclusions and evidence, each in short statements.

Where the Page Sits and How It Gets Found

A grounding page no system can retrieve is a note to self. Four points decide that:

• A dedicated, permanent address. A fixed path such as `/facts`, one that does not change. Every rename resets the recognition.
• Linked from the footer. Reachable from every page, not buried in a submenu. A footer link is at the same time the signal that this is the company's page about itself.
• Indexable. No `noindex`, no exclusion in `robots.txt`, present in the sitemap. It sounds obvious; on newly created pages it often is not.
• Linked from the matching content. Wherever one of the details comes up in the blog or on a service page, a reference to the fact page belongs there.

Knowledge pages, glossaries and help articles can point to the same page. The more internal routes lead to it, the more likely it is read as the authoritative place for those details.

How to Tell Whether It Works

The simplest test costs ten minutes: ask the systems that matter to you the same questions a prospect would ask. What does the company do? For whom? What does it cost? Where is it based? Then compare the answer with your own page.

Three things are worth watching: whether the details are right, whether your own page shows up as a source at all, and whether the answers diverge between systems. The third case is the most telling — it shows that the systems are drawing on different sources and that your own entry has not yet prevailed.

Between publication and effect there is time. The page has to be indexed, retrieved and weighted against other sources. A gap of a few weeks between checks is more realistic than a daily inspection. How to measure this systematically rather than by spot check is covered in AI Visibility.

How It Differs From Legal Notice, About Page and FAQ

These three pages usually exist already, and none of them replaces a grounding page.

• The legal notice fulfils a legal obligation. It names the operator, but neither the services nor the prices nor the exclusions.
• The about page tells a story. It is written for effect, not for verifiability, and rarely contains hard details.
• The FAQ answers customer questions during the buying process. It covers sections, not the full picture.

A grounding page can link to all three. It cannot replace any of them, and none of them replaces it.

What a Grounding Page Does Not Do

It Does Not Force a Citation

There is no mechanism that obliges a model to draw on a particular page. The page raises the probability that correct information is available, and no more than that.

It Does Not Replace External Evidence

This is where a grounding page reaches its limit. Self-reported information is one source among many, and models weight matching statements from independent sources higher than a provider's statement about itself. When trade media, directories and interviews say something different from your own page, your own page rarely wins.

That is why a grounding page only takes effect together with external mentions. How to build those up is covered in AI Mentions, and measuring visibility in AI systems in AI Visibility. This external evidence is exactly what you build predictably with GetMentioned: pick fitting publishers, compare SEO data and prices, book brand mentions and placements directly, and document publications centrally.

It Is Not a Ranking Instrument

A grounding page does not aim at positions in the search results. It aims at the details being right when a system talks about you.

Conclusion

A grounding page is the answer to a simple problem: AI systems talk about companies whether those companies steer it or not. Anyone who provides no solid source of facts leaves the details to whatever can be assembled from old directories and other people's articles.

The effort stays manageable, because the facts are known anyway. The work consists of writing them down in a citable form and keeping them current. The page only takes effect, though, once independent sources say the same thing.