Structured Data for AI Search

Structured data using JSON-LD and Schema.org helps your content's entities, relationships, and types. It is a core GEO technical implementation that improves AI search visibility through machine-readable signals AI crawlers and large language models can extract reliably.

TL;DR

Google and AI engines prefer JSON-LD over Microdata or RDFa. Pick one primary schema per page (TechArticle, FAQPage, HowTo, Product, etc.), centralize supplementary types under @graph, validate with the Schema.org Validator and Google's Rich Results Test, and keep dateModified accurate. Schema quality — not mere presence — correlates with AI Overview citation eligibility.

For the broader strategy, see the Technical AI optimization hub.

Quick Verdict

Schema markup for AI search in 2026 is mostly about entity disambiguation and citation eligibility, not classic SERP rich results. Google has progressively restricted FAQ and HowTo rich snippets, but AI engines — Google AI Overviews, ChatGPT, Perplexity, Gemini, and Claude — still parse JSON-LD aggressively to disambiguate entities, attribute authorship, and assess freshness. The leverage has shifted from "earn a fancy SERP card" to "make your content unambiguously machine-readable."

Pages with valid TechArticle, Article, or Product schema with explicit author, datePublished, dateModified, and about entities are demonstrably more likely to be cited in AI Overviews than pages with no markup or partial markup. The cost is low, the upside is direct, and validation tooling is mature. The single most common mistake is treating schema as a one-time setup; schema drift (stale dates, deprecated types, mismatched content) silently kills AI visibility over months. Treat schema as a quarterly hygiene task on top 50 pages, validated against Google Rich Results Test and the Schema.org Validator.

What Changed in 2024-2026

Three structural shifts reshaped how schema interacts with AI search:

FAQ rich results restricted (August 2023, ongoing). Google announced that FAQ rich results would only appear for verified government and health authority sites. For everyone else, the visible search-result FAQ accordion disappeared. The schema itself was not deprecated — and AI Overviews, ChatGPT, and Perplexity still parse FAQPage schema for citation eligibility. Schema visibility on the SERP and schema utility for AI extraction are now decoupled.

HowTo rich results deprecated (September 2023). Google removed HowTo rich snippets from desktop and mobile results. The HowTo type still exists in Schema.org and AI engines still parse it for step extraction in synthesized answers, but expect zero classic SERP enhancement.

AI Overviews launched globally (May 2024 → 2026). Google AI Overviews became the default for many query intents in 2024-2026, and citation patterns showed strong correlation with markup quality. Independent SearchEngineLand controlled experiments in 2025 found that pages with valid schema appeared in AI Overviews while sibling pages with invalid or absent schema did not, even when content was substantially equivalent.

@graph centralization became the standard. As pages began legitimately needing 3-5 schemas (Article + Organization + BreadcrumbList + FAQPage), the @graph pattern emerged as the W3C-aligned best practice. It reduces DOM clutter, simplifies validation, and makes entity relationships explicit through stable @id URIs.

Entity-driven matching matters more than keywords. AI engines map content to entities (@type: Thing, @id, sameAs Wikidata) and link them across the web's knowledge graph. Pages without explicit entity definitions are disambiguated probabilistically — a coin flip you usually lose at scale.

Why Structured Data Matters for AI

AI systems parse content differently than traditional search engines. Structured data provides explicit signals that materially improve retrieval and citation:

Independent testing (SearchEngineLand, 2025) found that only well-implemented schema — not partial or invalid markup — appeared in Google AI Overviews across a controlled three-page comparison. Mere presence is insufficient; correctness, completeness, and freshness all matter.

Schema Types vs AI Engines Matrix

Different AI engines weight different schema types differently. The matrix below summarizes observed patterns from public statements, Schema.org documentation, and independent testing through Q2 2026:

Key takeaways from the matrix:

1. Article / TechArticle is universally parsed by every AI engine. If you implement only one schema, make it this one with full author, datePublished, dateModified, and about entities.
2. Organization schema is the lowest-effort, highest-leverage site-wide addition. Every engine uses it for source authority signals. Add it once via @graph in a global layout.
3. Product, Offer, and Review schemas remain the highest-stakes for commercial content. They directly influence price extraction, comparison tables, and shopping result inclusion.
4. MedicalCondition, MedicalProcedure, and FinancialProduct require strict YMYL (Your Money Your Life) compliance. Incorrect or unsourced markup in these categories can trigger manual actions and full site demotion.
5. FAQPage and HowTo retain strong AI parsing value despite SERP rich-result restrictions. Use them where content genuinely is FAQ or step-by-step tutorial — not as decorative markup on unrelated content.
6. BreadcrumbList helps AI position your page within a topical taxonomy. It is especially important for content hubs and pillar pages.
7. Person, Author, and reviewedBy entities are increasingly weighted for E-E-A-T signals. A linked Person with verified sameAs URLs to LinkedIn, Wikidata, or ORCID materially improves citation eligibility.

When Schema Actually Moves the Needle

Schema is not a magic visibility lever. It works in specific scenarios:

1. Entity disambiguation. When your topic shares names with other entities (e.g., "Apple" the company vs the fruit, "Python" the language vs the snake), explicit about entities with @type: Thing and sameAs Wikidata URIs let AI engines bind your content to the correct entity. Without this, you compete with every other entity that shares the name and you usually lose.

1. Authorship and freshness signals. Pages with verified author (linking to a Person with credentials) and accurate datePublished / dateModified are preferred for citation in AI Overviews. Stale dateModified (older than 18 months) is a known suppressor across all major AI engines.

1. Topical hierarchy. BreadcrumbList plus isPartOf references position your page within a content taxonomy. AI engines use this to surface the most specific authoritative page for a query rather than a general overview.

1. Multi-entity pages. When a page covers multiple distinct entities (a tool comparison, a product family, a research paper with multiple findings), @graph with explicit @id URIs prevents AI from collapsing them into a single ambiguous topic.

1. Trust signals on YMYL content. Health, finance, legal, and safety content benefits disproportionately from Person author markup, reviewedBy editorial links, and citation chains via citation and isBasedOn properties.

Where schema does NOT move the needle:

• Schema cannot rescue thin or unoriginal content. AI engines extract semantic meaning; markup amplifies signal but does not create signal.
• Schema cannot override violated quality guidelines. Pages flagged by manual actions, spam policies, or YMYL rules are not redeemed by valid markup.
• Schema rarely helps for purely transactional or navigational queries where the SERP collapses to a brand result, knowledge panel, or AI Overview synthesized from a small canonical set.
• Schema does not "rank" pages directly. It clarifies what the page is about, enabling AI to retrieve and cite it more accurately when the underlying content quality merits citation.

The mental model: schema is a translator, not a content quality multiplier. It helps the right page be found and cited; it does not make the wrong page right.

Recommended Schemas in Practice

TechArticle

For documentation and technical content:

{
  "@context": "https://schema.org",
  "@type": "TechArticle",
  "headline": "What Is GEO?",
  "description": "Canonical definition of Generative Engine Optimization.",
  "author": {
    "@type": "Person",
    "name": "Geodocs Research Team"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Geodocs.dev",
    "url": "https://geodocs.dev"
  },
  "datePublished": "2025-04-20",
  "dateModified": "2026-05-01",
  "keywords": "GEO, Generative Engine Optimization, AI search",
  "about": [
    { "@type": "Thing", "name": "GEO" },
    { "@type": "Thing", "name": "Generative Engine Optimization" }
  ]
}

FAQPage

For genuine FAQ sections and question-answer content:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What is GEO?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "GEO is the practice of structuring content so AI systems can understand, retrieve, synthesize, and cite it."
      }
    }
  ]
}

Note: Google has restricted FAQ rich-result eligibility on standard SERPs. The FAQPage schema is still valuable for AI extraction and remains a strong signal for citation eligibility, even if classic FAQ rich snippets do not appear.

HowTo

For step-by-step tutorials:

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "How to Create llms.txt",
  "step": [
    {
      "@type": "HowToStep",
      "name": "Create the file",
      "text": "Create a file named llms.txt in your site's public directory."
    },
    {
      "@type": "HowToStep",
      "name": "Write the header",
      "text": "Start with your site name as an H1 heading and a blockquote description."
    }
  ]
}

Centralizing with @graph

When a page legitimately needs multiple schemas (e.g. an Article that is also part of a Series, authored by an Organization, and has a BreadcrumbList trail), centralize them in a single JSON-LD block using @graph:

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "TechArticle",
      "@id": "https://geodocs.dev/technical/structured-data-for-ai-search#article",
      "headline": "Structured Data for AI Search"
    },
    {
      "@type": "Organization",
      "@id": "https://geodocs.dev#org",
      "name": "Geodocs.dev"
    }
  ]
}

This reduces DOM clutter, simplifies validation, and makes entity relationships explicit via stable @id URIs that downstream tools can deduplicate.

Implementation Guide

Step 1: Choose your primary schema type

Step 2: Add JSON-LD to the page head

Place JSON-LD in a