Structured Data for AI Search: JSON-LD, GEO & Getting Cited by ChatGPT

Search changed: from links to answers

Classic SEO optimized for a ranked list of blue links. AI answer engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Bing Copilot — instead *synthesize an answer* and cite a few sources. Getting into that answer is a new discipline: Generative Engine Optimization (GEO). And the single most underused GEO lever is structured data in JSON-LD, because AI systems parse it to *verify* facts rather than infer them from your paragraphs.

Why machines prefer JSON-LD to prose

Read this sentence: "Our guide, updated last spring by the team, is well-reviewed." A model has to guess the date, the author, and what "well-reviewed" means. Now give it JSON-LD:

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "JSON Schema Draft 2020-12 Guide",
  "datePublished": "2026-06-30",
  "dateModified": "2026-07-05",
  "author": { "@type": "Organization", "name": "JSONKit" },
  "aggregateRating": { "@type": "AggregateRating", "ratingValue": 4.8, "reviewCount": 214 }
}

Every fact is now unambiguous and machine-verifiable. Structured data is the difference between content an LLM has to infer and content it can confirm and cite with confidence — which is exactly what earns you a citation.

The entity graph: help models connect the dots

GEO isn't just per-page tags; it's about describing entities and their relationships so a model can reason about your content. Link types together — an Article with an author that's a Person who worksFor an Organization; a Product with a brand and offers. Use @id to give entities stable identifiers you reference across pages, building a small knowledge graph the model can traverse.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "Does llms.txt help SEO?",
    "acceptedAnswer": { "@type": "Answer", "text": "No — Google ignores it for rankings." }
  }]
}

FAQPage and HowTo schemas are especially GEO-friendly because they map directly onto the question-and-answer shape AI engines are looking for.

The rest of the GEO checklist

Structured data only helps if the model can reach and trust your page:

• Don't block AI crawlers in robots.txt, and confirm your server isn't 403-ing their user-agents.
• Server-side render important content — many AI crawlers don't run JavaScript, so client-only content is invisible.
• Keep facts consistent between your visible text and your JSON-LD; contradictions erode trust and can get the markup ignored.
• Write scannable, well-headed content so extraction is clean even beyond the structured data.

For the AI-crawler map file, see llms.txt Explained.

Validate your JSON-LD, always

Invalid JSON-LD is worse than none — a single syntax error voids the whole block and the facts silently vanish:

1. Format the JSON-LD so nested entities stay readable as they grow.
2. Validate the JSON to catch trailing commas or unquoted keys that break the block.
3. Diff your rendered markup against your template when facts change, so dateModified and ratings don't go stale.

For the foundational how-to, see JSON-LD & Schema.org for SEO; for the standards behind the shapes, JSON Schema Draft 2020-12.