Advanced Semantic Markup (Schema) for LLMs

Going beyond traditional SEO: how deep, interconnected use of Schema.org helps LLMs disambiguate entities on your site.

Remountly TeamSEO Engineering

July 21, 20263 min read

Technical SEO

Executive brief

Key takeaways

Use the 'sameAs' property aggressively to link your site to your entities on Wikipedia, Wikidata, and official social networks.
Create a graph of connected entities (@id) instead of multiple loose JSON-LD fragments.
Models like ChatGPT use structured data to fill knowledge gaps and ensure factuality (grounding).
Clearly define the 'MainEntity' of the page to direct the algorithm's attention.

For a long time, the incentive to add structured data markup (Schema.org) to a site's code was purely aesthetic: getting those desired review stars in Google results or displaying a featured recipe. With the rise of Generative Artificial Intelligence and Large Language Models (LLMs), semantic markup is once again the technical heart of SEO.

LLMs, Ambiguity and "Grounding"

LLMs are based on probabilities of words (tokens) appearing together. As intelligent as they may seem, they don't "understand" the real world and are susceptible to hallucinations — mixing up, for example, "Apple" (the tech company) with "Apple" (the Beatles' record label).

The Grounding process (anchoring in real facts) is how AI engines try to solve this during a search. This is where structured data in JSON-LD comes in. By providing an explicit graph defining who you are, the AI doesn't have to "guess" based on the text. It consumes the direct data.

Connecting the dots with `@id` and Semantic Graphs

Instead of throwing multiple random JSON-LD blocks into the <head>, the ideal structure in the AI era is a connected graph.

Use the @id attribute to reference entities throughout the document. If the article was written by "John", define the Person entity with an ID and link it to the author property of the Article.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Person",
      "@id": "https://mysite.com/about#john",
      "name": "John Smith",
      "sameAs": ["https://www.linkedin.com/in/johnsmith", "https://www.wikidata.org/wiki/Q12345"]
    },
    {
      "@type": "Article",
      "headline": "The Future of Artificial Intelligence",
      "author": { "@id": "https://mysite.com/about#john" }
    }
  ]
}

The `sameAs` Property as a Universal Disambiguator

The sameAs property is perhaps the most underrated technical SEO tool in the AI context. It allows you to tell the LLM: "This entity I am describing here is exactly the same entity as that page on Wikipedia or Wikidata".

Since LLMs have a massive training weight on Wikipedia, establishing this connection transfers the authority and semantic knowledge from the global database directly to your proprietary entity.

Critical Properties for Content and AI

For pages with articles, tutorials, and analysis (where GEO - Generative Engine Optimization acts strongly), reinforce the following properties:

mainEntity: Explicitly defines what the main thing (Product, Event, Concept) the page is about.
about: Lists the central topics (entities) of the text.
mentions: Lists secondary concepts that are mentioned.
citation: Crucial to prove the veracity of your content to the AI; connect links to studies, reports, or renowned authors.

Example of Finding and Action

Observation: A financial software site noticed that engines like Perplexity and Google AI Overviews often confused them with an insurance company of the same name, recommending the wrong services to users in generated answers. Action: Refactor the JSON-LD of the home page and the "About" page. Implement a complete Organization schema with aggressive linking via sameAs pointing to verified Crunchbase, LinkedIn, and corporate B3 profiles, in addition to filling the knowsAbout property with appropriate financial keywords. Acceptance: The structured data model forces disambiguation; search agents began to anchor the correct entity in their answers, clearly separating the software company from the insurance company.

AI thinks in entities (things), not in strings (loose text). Speaking the mathematical language of entities through deep semantic markup is the most efficient shortcut to being understood and recommended in the new search landscape.

Direct answers

Frequently asked questions

Do LLMs like GPT and Claude read Schema.org?

Yes. When accessing the web (whether in the training phase or real-time searches), parsers extract JSON-LD blocks because they offer clear logical relationships, free from the ambiguities of natural language.

What is the difference between using Schema for SEO and for LLMs?

In classic SEO, the focus was on validating specific types (like FAQ or Recipe) to earn visual badges. For LLMs, the focus is on building the full ontology of the page: using 'about', 'mentions', and connecting entities systemically.

Is the JSON-LD format still recommended?

Absolutely. JSON-LD remains Google's official standard and the simplest format for AI crawlers and parsers to extract without having to process the entire DOM tree.