Schema.org for AI Search: A JSON-LD Playbook for LLM Citations

Why Schema.org Matters for Both SEO and LLMO

In short

Schema.org gives search engines and large language models a machine-readable description of what a page contains, who wrote it, and how its parts relate. For SEO, it powers Rich Results. For LLMO, it feeds retrieval systems with structured signals that prose alone cannot provide.

Schema.org was launched in 2011 as a joint initiative by Google, Microsoft (Bing), Yahoo, and Yandex. The goal was a single shared vocabulary for describing structured data on the web.

Fifteen years later, Schema.org is the de facto standard. Google, Bing, OpenAI, Anthropic, and Perplexity all consume it. It is hosted as an open standard at schema.org and continues to evolve through community contribution.

For traditional SEO, Schema.org powers Rich Results — the visual enhancements in Google search results like FAQ accordions, star ratings, and HowTo carousels. For LLMO, the value is different but at least as important, which is why a modern AI search optimization consultant spends more time on schema markup than on backlinks.

Large language models work best when content is unambiguous. Schema tells the retrieval layer:

• What this page is. Article? FAQ? Glossary entry? Product? Each type has different extraction rules.
• Who wrote it. Person schema with credentials gives the model an author entity to cite.
• Who publishes it. Organization schema establishes the publisher brand across the entire web.
• When it was published and updated. Freshness signals weight heavily in LLM retrieval.
• How parts relate. mainEntity, isPartOf, and about properties build a knowledge graph the model can traverse.

With ~58.5% of US Google searches now ending without a click (SparkToro 2024), and Google AI Overviews live since May 2024, structured data is no longer optional for visibility.

The Seven Most Impactful Schemas for LLMO (with Real JSON-LD)

In short

The seven schemas with the highest LLMO impact are Article, FAQPage, HowTo, Organization, Person, DefinedTerm, and BreadcrumbList. Together they cover 95% of content templates and feed both Google Rich Results and LLM retrieval systems with the signals that matter.

Each of the schemas below uses real, validator-tested JSON-LD. Copy, adapt the values, and validate before deployment. All examples use JSON-LD because it is the format Google recommends.

1. Article (or BlogPosting / NewsArticle)

Use Article for general content, BlogPosting for blog posts, and NewsArticle for time-sensitive journalism. The structure is the same — only the @type changes.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Schema.org for AI Search: A JSON-LD Playbook",
  "author": {
    "@type": "Person",
    "name": "Linus Ingemarsson",
    "url": "https://www.linkedin.com/in/linus-ingemarsson/"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Alice Labs",
    "logo": {
      "@type": "ImageObject",
      "url": "https://alicelabs.ai/logo.png"
    }
  },
  "datePublished": "2026-05-06",
  "dateModified": "2026-05-06",
  "image": "https://alicelabs.ai/images/og/schema-org-for-ai.jpg",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://alicelabs.ai/en/insights/schema-org-for-ai"
  }
}

2. FAQPage

FAQPage markup wraps Question and Answer entities. LLMs love it because each Q-A pair is a self-contained, citable unit.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is Schema.org?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Schema.org is an open standard vocabulary for structured data, founded in 2011 by Google, Microsoft, Yahoo, and Yandex."
      }
    },
    {
      "@type": "Question",
      "name": "Is JSON-LD better than Microdata?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Google explicitly recommends JSON-LD because it is cleanly separated from HTML, easier to maintain, and less prone to rendering errors."
      }
    }
  ]
}

3. HowTo

HowTo markup describes step-by-step procedures. Each HowToStep needs a name and text. You can also include totalTime, tool, and supply for richer markup.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to Deploy Schema.org JSON-LD on a Website",
  "totalTime": "PT2H",
  "step": [
    {
      "@type": "HowToStep",
      "name": "Inventory existing schemas",
      "text": "Run the Rich Results Test on 10-20 representative URLs to see which schema types are already present."
    },
    {
      "@type": "HowToStep",
      "name": "Choose schema types per template",
      "text": "Map schema types (Article, FAQPage, HowTo, Organization) to page templates, not individual URLs."
    },
    {
      "@type": "HowToStep",
      "name": "Validate with Rich Results Test",
      "text": "Test every JSON-LD block with Google's Rich Results Test before deployment."
    }
  ]
}

4. Organization

Organization is the publisher entity. Place it on your homepage or About page. The sameAs property is critical — it links your brand to LinkedIn, Wikipedia, and Wikidata for entity disambiguation.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://alicelabs.ai/#organization",
  "name": "Alice Labs",
  "url": "https://alicelabs.ai",
  "logo": "https://alicelabs.ai/logo.png",
  "description": "Nordic AI consultancy specialising in enterprise AI implementations and LLMO.",
  "sameAs": [
    "https://www.linkedin.com/company/alicelabsai",
    "https://twitter.com/alicelabs"
  ]
}

5. Person

Person schema gives each author bio a machine-readable identity. Use it on author pages and within Article author fields. Reference the same Person @id across articles for consistency.

{
  "@context": "https://schema.org",
  "@type": "Person",
  "@id": "https://alicelabs.ai/team/linus-ingemarsson/#person",
  "name": "Linus Ingemarsson",
  "jobTitle": "Co-Founder",
  "worksFor": {
    "@id": "https://alicelabs.ai/#organization"
  },
  "url": "https://alicelabs.ai/team/linus-ingemarsson/",
  "sameAs": [
    "https://www.linkedin.com/in/linus-ingemarsson/"
  ]
}

6. DefinedTerm

DefinedTerm marks glossary entries and definition pages. It tells LLMs explicitly: this page authoritatively defines this term. High value for entity-driven LLMO content.

{
  "@context": "https://schema.org",
  "@type": "DefinedTerm",
  "name": "LLMO",
  "alternateName": "Large Language Model Optimization",
  "description": "LLMO is the discipline of optimizing content so it gets cited by AI search systems like ChatGPT, Claude, Perplexity, and Google AI Overviews.",
  "inDefinedTermSet": {
    "@type": "DefinedTermSet",
    "name": "Alice Labs LLMO Glossary",
    "url": "https://alicelabs.ai/en/insights"
  }
}

7. BreadcrumbList

BreadcrumbList describes the navigation hierarchy. It is small, universally supported, and helps LLMs understand site structure.

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Insights",
      "item": "https://alicelabs.ai/en/insights"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Schema.org for AI Search",
      "item": "https://alicelabs.ai/en/insights/schema-org-for-ai"
    }
  ]
}

Together, these seven schemas cover almost every page template you will build. Add Product, Review, AggregateRating, VideoObject, and ImageObject when the content type calls for them.

JSON-LD vs Microdata vs RDFa: Just Use JSON-LD

In short

Schema.org supports three serialization formats: JSON-LD, Microdata, and RDFa. For new implementations in 2024+, use JSON-LD. Google explicitly recommends it, it is cleanly separated from HTML, and it is easier to maintain, validate, and ship via templates or CMS plugins.

All three formats are valid. Search engines parse all three. So why JSON-LD?

• Separation of concerns. JSON-LD lives in a<script>tag. Your HTML stays clean. Your schema stays in one place.
• Easier maintenance. One JSON block per page is easier to template, test, and update than dozens of inline attributes scattered through markup.
• Less rendering risk. Microdata and RDFa break when developers refactor HTML. JSON-LD does not depend on the surrounding markup.
• Google preference. Google Search Central documents JSON-LD as the recommended format. It still parses the others, but recommends JSON-LD for new sites.
• Tooling. Most CMS plugins (Yoast, Rank Math, Next SEO, etc.) ship JSON-LD by default. Microdata support is fading.

If you have legacy Microdata or RDFa, you do not need to rip it out. But every new schema block you ship should be JSON-LD.

How to Validate Schema (Rich Results Test + validator.schema.org)

In short

Always run two validators before deploying schema. Google's Rich Results Test (search.google.com/test/rich-results) checks Google-eligibility. The official validator.schema.org checks pure Schema.org compliance. Fix every error and warning before pushing to production.

Schema validation is non-negotiable. Malformed JSON-LD does not silently fail — it can disqualify a page from Rich Results entirely and confuse LLM extraction.

Use both validators because they answer different questions:

1. Google Rich Results Test. Hosted at search.google.com/test/rich-results. Tells you whether Google will surface the page as a Rich Result. Tests for Google's specific required fields, which can be stricter than Schema.org's.
2. validator.schema.org. The official Schema.org validator. Checks pure compliance with the vocabulary. Useful for debugging type errors and unrecognized properties.
3. Search Console Enhancements. Once deployed, Search Console reports structured data coverage, errors, and warnings under the Enhancements section. Monitor weekly.

A clean validation result has zero errors and zero warnings. Yellow warnings can sometimes block Rich Results — fix them too.

Common Schema Mistakes (and How to Avoid Them)

In short

The most common Schema.org mistakes are missing required fields, conflicting Organization references across pages, FAQ markup abuse on non-FAQ content, and malformed dates. Each silently degrades both Rich Results eligibility and LLM trust.

We have audited dozens of Nordic enterprise sites. The same handful of issues come up almost every time:

• Missing required fields. Article schema without author, datePublished, or publisher. FAQPage with Question but no acceptedAnswer. Always check the required fields in schema.org docs before shipping.
• Conflicting Organization entities. Different pages declare slightly different Organization names, logos, or URLs. LLMs treat these as separate entities. Use a single Organization @id and reference it everywhere.
• FAQ markup abuse. Stuffing FAQPage schema on pages that are not actually FAQs. Google penalized this in 2023 — most sites now restrict FAQ Rich Results to truly FAQ-shaped pages. Use it where appropriate, not as a hack.
• Dates not in ISO 8601. datePublished must be YYYY-MM-DD or full ISO 8601 with timezone. "May 6, 2026" will fail validation.
• Image URLs that don't resolve. Article requires an image. If the URL 404s or is blocked by robots.txt, the schema is invalid.
• Schema-content mismatches. HowTo schema with five steps but only three on the page. FAQPage with questions that don't appear in visible HTML. Both can trigger Rich Results demotion.
• Multiple BlogPosting on the same URL. Some themes inject BlogPosting via the theme and again via an SEO plugin. The result is two Article schemas, sometimes with different values. Audit and consolidate.

Most of these are caught by validation. The Organization @id issue is the one that requires manual review — validators won't flag it because each individual block is technically valid.

Schema for AI Overviews + LLM Citations Specifically

In short

For Google AI Overviews (launched May 2024) and LLM citations across ChatGPT, Claude, Perplexity, and Gemini, the highest-leverage schemas are Article, FAQPage, HowTo, Organization, Person, and DefinedTerm — combined with citation-rich prose, freshness signals, and llms.txt at the domain root.

Google AI Overviews launched in May 2024 and dramatically changed the SERP landscape. AI Overviews preferentially cite sources with structured data, named authors, and clear publisher identity.

For LLM citations more broadly — ChatGPT, Claude, Perplexity, Gemini — Schema.org plays the same role. The Aggarwal et al. 2024 GEO paper (arXiv:2311.09735) showed that citation-rich, structured content can lift visibility in generative engines by up to 40%.

Combine schema with three other LLMO basics:

1. llms.txt at your domain root. The llms.txt standard (Jeremy Howard / Answer.AI, September 2024, llmstxt.org) is a markdown file that summarizes your site for LLMs. Free to publish, zero downside.
2. Citation-rich prose. Inline statistics with named sources, expert quotes with credentials, and concrete entities. Schema marks up the structure; prose provides the citable claims.
3. Freshness signals. dateModified updated when you actually update content. Stale, undated content is deprioritized by retrieval systems that weight recency.

At Alice Labs, we deploy Article + FAQPage + HowTo + DefinedTerm + Organization + Person on every content page. Combined with llms.txt and citation-rich content, this is our baseline LLMO stack for client implementations.