Insights

Schema.org and LLMs

Structured data can describe visible content and support specific Search features. It does not provide a documented shortcut into AI answers. This guide separates valid implementation from unproven claims.

Updated 19 September 2026

In brief

Schema.org is a standardised vocabulary of structured data, created in 2011 by Google, Microsoft, Yahoo and Yandex. It allows annotating HTML content so that consumers that parse the markup can identify entities, properties and relationships. Whether a particular search or answer product consumes a field is product-specific. This guide covers accurate JSON-LD implementation and the limits of what it proves.

1. What schema.org can and cannot prove

Structured data has two defensible uses:

  1. Description. Markup can state that a visible page is an article, identify its author and dates, or describe an organisation and breadcrumbs.
  2. Supported Search features. Google documents required and recommended properties for specific rich-result types. Eligibility still does not guarantee display.

Google says no special markup is required for AI Overviews or AI Mode and that structured data must match visible content. OpenAI and Perplexity do not document a general schema.org ranking or citation effect. Validation therefore proves syntax and, for supported Google features, eligibility conditions; it does not prove ranking, citation or model training.

2. Schema types by page type

2.1 Article / BlogPosting

To use on all editorial content pages (articles, guides, analyses). Commonly useful fields, subject to the consumer's current documentation:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Exact article title",
  "description": "150-160 character summary",
  "datePublished": "2026-04-22",
  "dateModified": "2026-04-22",
  "inLanguage": "en",
  "author": {
    "@type": "Organization",
    "name": "Your organisation",
    "url": "https://your-site.com"
  },
  "mainEntityOfPage": "https://your-site.com/article/"
}

Common errors: publishing a dateModified that does not match a substantive visible revision, using an empty description, or identifying an author that the page does not show.

2.2 FAQPage

To use on pages containing a questions/answers section. Use this type only when the page visibly contains questions with answers and the applicable consumer still supports the intended feature. Google stopped showing FAQ rich results in May 2026.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is RAG?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "RAG (Retrieval Augmented Generation) is a system that allows an LLM to consult external sources before generating its response, to produce cited and up-to-date answers."
      }
    }
  ]
}

Quality rule: each answer must match the visible answer. There is no documented 40-word minimum. Write the shortest complete answer the user needs.

2.3 Organization

Use on the homepage or About page when it accurately describes the organisation. It can link your site to your Wikidata, Wikipedia, LinkedIn, Crunchbase profiles via sameAs. Each URL should genuinely identify the same entity.

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Name of your organisation",
  "url": "https://your-site.com",
  "logo": "https://your-site.com/logo.png",
  "description": "Factual description in 1-2 sentences",
  "foundingDate": "2024",
  "sameAs": [
    "https://www.wikidata.org/wiki/Q...",
    "https://www.linkedin.com/company/...",
    "https://en.wikipedia.org/wiki/..."
  ]
}

2.4 BreadcrumbList

Use where a visible breadcrumb trail represents the page hierarchy. Follow Google's Breadcrumb structured data requirements if Google Search is the target consumer.

{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://your-site.com/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Insights",
      "item": "https://your-site.com/insights/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Schema.org and LLMs"
    }
  ]
}

2.5 HowTo

HowTo can describe a visible step-by-step procedure, but Google does not document it as an AI Overview citation tactic. Check current Search feature support before investing in markup.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to optimise your site for LLMs",
  "step": [
    {
      "@type": "HowToStep",
      "position": 1,
      "name": "Step 1: Audit robots.txt",
      "text": "Verify that the major AI bots (GPTBot, PerplexityBot, ClaudeBot) are not blocked in your robots.txt."
    },
    {
      "@type": "HowToStep",
      "position": 2,
      "name": "Step 2: Add Article schema",
      "text": "Add a JSON-LD Article block with datePublished and dateModified on each content page."
    }
  ]
}

2.6 WebSite

WebSite can describe the site's name and URL. Do not promise a sitelinks search box: Google removed that visual element in November 2024.

{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "name": "Site name",
  "url": "https://your-site.com",
  "inLanguage": "en",
  "description": "Short site description"
}

Note: do not duplicate the WebSite schema across multiple pages - one instance on the homepage is sufficient.

3. Injecting multiple schemas on the same page

The recommended technique in 2026 is to inject a JSON-LD array containing multiple schema objects in a single <script type="application/ld+json"> tag:

<script type="application/ld+json">
[
  {
    "@context": "https://schema.org",
    "@type": "Article",
    "headline": "...",
    "datePublished": "2026-04-22",
    "dateModified": "2026-04-22"
  },
  {
    "@context": "https://schema.org",
    "@type": "FAQPage",
    "mainEntity": [...]
  }
]
</script>

JSON-LD arrays and separate script tags can both be valid. Test syntax and the requirements of the specific consumer; do not assume every answer engine parses every object.

4. Schema errors that harm AI visibility

4.1 Schema inconsistent with visible content

If your FAQPage schema lists questions absent from the visible HTML, it violates Google's structured-data guidance. Rule: schema must reflect what a user would see on the page, not invisible or truncated content.

4.2 Incorrect or missing date data

datePublished: "2026" is not a valid ISO 8601 format. Use "2026-04-22" (YYYY-MM-DD) or "2026-04-22T10:00:00+00:00" (with time and timezone). An invalid format is ignored by parsers.

4.3 Organization.sameAs pointing to broken URLs

A broken or incorrect sameAs URL misdescribes the entity. Check that each link resolves and identifies the same organisation or person.

4.4 WebSite schema duplicated across multiple pages

Keep a single consistent WebSite entity, normally rooted on the homepage. Duplicate or conflicting site entities can make the markup harder to interpret.

4.5 Malformed JSON-LD

Invalid JSON (missing comma, unescaped quote, unclosed brace) causes the parser to completely ignore the block. Verify with the Rich Results Test before any deployment.

5. Schema.org and AI surfaces in 2026

Surface Documented position Safe conclusion
Google AI Overviews No special schema required Use valid markup that matches visible content; no citation guarantee
Bing Copilot No general citation effect documented here Validate against Bing's current supported features
Perplexity No schema.org ranking weight published Do not claim that dateModified causes selection
ChatGPT Search No schema.org ranking weight published Do not infer ChatGPT Search behaviour from another engine
LLMs (training) Training datasets and parsers vary Markup presence does not prove training use
Google Featured Snippets Not created by FAQPage markup Featured snippets are selected from page content

6. Action plan: implementation priorities

  1. Week 1. Inventory existing markup and map each type to visible content and a documented consumer. Remove false or conflicting properties.
  2. Week 2. Add or repair Article markup on editorial pages where dates, headline and author are visible and accurate.
  3. Week 3. Add Organization schema on the homepage with sameAs fields completed (Wikidata, LinkedIn or Wikipedia only when they genuinely identify the same entity).
  4. Week 4. Add BreadcrumbList where a visible hierarchy exists. Validate syntax and current feature eligibility before deployment.

Structured data checklist

FAQ

Is schema.org essential to be cited in LLMs?

No. Google explicitly states that no special schema is required for AI Overviews or AI Mode. Structured data can describe visible content and support eligible Search features, but major answer-engine providers do not publish a general citation gain from schema.org.

Which schema is most useful for AI SEO?

Use the type that accurately matches the visible page and a supported use case. Article can describe editorial content, Organization can describe an organisation, and BreadcrumbList can describe navigation. Google stopped showing FAQ rich results in May 2026, and no provider documents FAQPage as a citation shortcut.

Can you use multiple schemas on the same page?

Yes, when each object describes visible content accurately. You can use an array, an @graph or separate JSON-LD blocks. Validation proves syntax and feature eligibility where applicable; it does not prove that every crawler consumes every object.

How do you verify that your schema is correctly read?

Three tools: the Google Rich Results Test (search.google.com/test/rich-results) to verify validity and eligibility for rich snippets, Schema.org Validator (validator.schema.org) for standard compliance, and the "Structured Data" section in Google Search Console to monitor production errors.

Does schema serve ChatGPT or Perplexity directly?

There is no official confirmation that schema.org directly improves citation in ChatGPT Search or Perplexity. Do not infer a ranking or training effect from the presence of JSON-LD. Use schema to represent visible facts and measure each search surface separately.