Schema Markup for AI Search
Schema markup is the vocabulary you use to describe your content to machines. It has been a search best practice for years — and it remains one of the cleanest ways to make your entities legible to the AI systems now reading the web. This guide covers what to implement and, just as importantly, what not to expect.
A JSON-LD block maps your content to typed, named properties a parser understands.
Start with JSON-LD
There are three ways to add schema.org markup — JSON-LD, microdata, and RDFa — but JSON-LD is the one to reach for. It is a self-contained script block that keeps structured data separate from your visible HTML, which makes it far easier to write, review, and maintain. Here is a complete example for an article:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "How to Prepare Your Roof for Winter",
"author": { "@type": "Person", "name": "Sam Rivera" },
"datePublished": "2026-05-14",
"publisher": {
"@type": "Organization",
"name": "Summit Roofing"
},
"mainEntityOfPage": "https://example.com/blog/winter-roof-prep"
}
</script>Every property here maps to something a reader can verify on the page. That is the rule: describe what is real.
A map of the types you'll use most
| Type | What it describes | When to use it |
|---|---|---|
| Organization | Company identity, logo, social profiles, contact points. | Homepage, about page. |
| LocalBusiness | Address, hours, geo, service area, phone. A more specific subtype of Organization. | Local service businesses. |
| Product / Offer | Name, description, price, availability, aggregate rating. | Ecommerce and product pages. |
| Article / BlogPosting | Headline, author, datePublished, image, publisher. | Editorial and blog content. |
| FAQPage | A list of Question/Answer pairs. | FAQ and support pages. |
| BreadcrumbList | The navigational path to the page. | Any deep page in a hierarchy. |
FAQPage schema and AI questions
FAQPage deserves special mention because its structure mirrors how people query AI systems: a question, then a direct answer. When a page genuinely answers common questions and marks them up cleanly, both human readers and machine parsers get a tidy question-to-answer mapping. That does not guarantee an AI system will quote you — retrieval and grounding decide that, and the exact selection logic is not public — but a well-formed answer to a real question is exactly the kind of content AI systems are built to surface.
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "Do you offer emergency service?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes. We provide 24/7 emergency response across the county."
}
}]
}For a deeper look at why this format works so well, see our guide on FAQ pages for AI search.
The honest limits
Schema is a legibility tool, not a magic ranking lever. As of mid-2026, no major AI provider has published a confirmed signal that says “pages with schema X are cited more.” Adoption and interpretation of structured data by AI systems is still emerging, and behavior changes frequently. Implement schema because it makes your entities unambiguous and future-proofs your content for machine reading — not because someone promised it would move you up a list that has no published order.
Pair it with the other things that reliably help: substantive content, clean HTML, a coherent entity strategy, and a curated llms.txt file.
Frequently Asked Questions
Will schema markup make ChatGPT or Perplexity cite me?
It can help those systems parse your entity correctly, but it is not a guaranteed path to citation. AI systems retrieve and ground answers in web content, and the specific signals they weigh are not published. Treat schema as one legibility input among several, not a citation switch.
Do I need every schema type on every page?
No. Use the type that matches the page's actual purpose. A product page gets Product, an article gets Article, a contact or homepage gets Organization or LocalBusiness. Piling on unrelated types dilutes meaning.
Where does the JSON-LD script go?
Anywhere in the HTML document — most implementations place it in the head or at the end of the body. Because JSON-LD is self-contained, its position does not change how it parses.
How is schema different from llms.txt?
Schema is embedded per-page markup that types the entities on that page. llms.txt is a single Markdown file at your root that curates which pages and topics matter across your whole site. Use both; they solve different problems.
Does invalid schema hurt me?
It can. A parser that hits malformed JSON may discard the entire block, so you get no benefit. Contradictory markup — claiming content that is not on the page — can also erode trust. Validate before deploying.
Make your entities machine-legible
Clean schema plus a curated llms.txt is the most reliable pairing for AI legibility. Start with your domain.