Schema markup for AI search: what the evidence says and what to add

Schema markup for AI search is useful but not required: no AI search operator makes structured data a condition for being cited, Google says explicitly that its AI features do not need it, and Microsoft has said it helps its language models understand content. The honest summary is that schema makes the facts on a page, such as price, availability, opening hours and who runs the business, precise and machine-readable, which helps some systems and costs little. This guide sets out what each operator has actually said, what is claimed without evidence, and the markup worth adding. It belongs to the guide to generative engine optimization; for schema in general, start with the schema markup guide.

Quick answer

• Google: structured data is not required for AI Overviews or AI Mode, according to its May 2026 guide. It still powers rich results, and it must match what is visible on the page.
• Microsoft: Bing's Fabrice Canel said in March 2025 that schema markup helps Microsoft's language models understand content, for Bing and Copilot.
• OpenAI: ChatGPT's shopping results use structured metadata from first-party and third-party providers, such as price and description. OpenAI has not said how it treats schema on ordinary pages.
• Everyone else: no public statement. Claims that schema "gets you cited" are opinion.
• Worth adding anyway: Organization with logo and sameAs, Product with Offer, LocalBusiness, Article with author and dates, BreadcrumbList.
• Put it in the server's HTML, not in a tag manager: most AI crawlers do not run JavaScript.
• Check it with the free schema markup validator.

What the AI operators have actually said

That is the whole record. Two things stand out. Nobody says schema is a ranking or citation factor for AI answers, and the one operator that addresses it most directly, Google, says you can appear without it.

Why the evidence is thin

AI search answers are built in two steps: retrieve pages, then have a language model read them. Structured data can matter at either step, but in different ways.

• At retrieval, search indexes use structured data for rich results and to understand entities such as products, businesses and people. That is established for Google and Bing search, and AI features built on those indexes inherit it indirectly.
• At reading, it depends on how a system turns a page into text for the model. Many text-extraction pipelines drop <script> elements, and JSON-LD lives in a <script> element. Whether a given AI product keeps it is undocumented. Our view: do not count on a model reading your JSON-LD; make sure the visible text states the same facts.

Studies by SEO vendors that correlate schema with AI citations exist, but correlation says little here: sites that maintain good structured data also tend to have well-structured, well-maintained pages. Without an operator's statement or a controlled test, treat those results as hints.

Which schema is worth adding

Choose markup for the facts an AI answer would need to describe you correctly, and for which Google and Bing already offer rich results.

Guides for each: Organization and WebSite schema, Product schema, LocalBusiness schema and Article and BlogPosting schema.

What about FAQPage?

FAQPage is valid schema.org, and some tools and other search engines read it. Google no longer shows FAQ rich results for any site: it limited them to government and health sites in 2023 and stopped showing them on 7 May 2026. Adding FAQ markup purely "for AI" has no documented benefit. If your page has a genuine FAQ, the questions and answers in visible HTML are what any reader, human or model, uses. FAQ and HowTo schema after Google's changes covers when it still makes sense.

There is no "AI schema"

Google's guide says there is no special markup for its AI features, and schema.org has no type for AI search. Be wary of plugins or services selling "AI schema" or "LLM markup": at best they output the ordinary types above.

The rules that make schema useful

1. Match the visible page. Every fact in JSON-LD, such as price, rating or opening hours, must appear on the page. Markup that says something the page does not is misleading to every system and can cost you rich results in Google.
2. Serve it in the HTML. JSON-LD injected by a tag manager or a front-end script exists only after rendering. Google can read it later; AI crawlers that do not run JavaScript never see it. Can AI crawlers execute JavaScript? shows how to test this.
3. One entity, one description. Two plugins each outputting an Organization or Product with different values give machines two conflicting answers. Link entities with @id so the page describes one connected graph.
4. Keep it current. A price in schema that is three months old is worse than no price.
5. No invented reviews or ratings. Mark up only reviews that are on the page and genuine.

A minimal Organization block, in the page's <head> or body, rendered by the server:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://example.com/#organization",
  "name": "Example Shop",
  "url": "https://example.com/",
  "logo": "https://example.com/logo.png",
  "sameAs": [
    "https://www.linkedin.com/company/example-shop",
    "https://www.instagram.com/exampleshop"
  ],
  "contactPoint": {
    "@type": "ContactPoint",
    "contactType": "customer service",
    "email": "[email protected]"
  }
}
</script>

How to check your markup

The schema markup validator finds every JSON-LD block on a page, checks types such as Product, Article, Organization and LocalBusiness for the properties Google requires for rich results, and flags duplicate or conflicting entities. Microdata is detected and counted; RDFa is not parsed. Every getReport report runs the same checks:

To confirm the markup is in the raw HTML rather than added by scripts, compare the two versions of the page:

The AI crawler check also reports whether structured data is present in the HTML AI crawlers read, alongside what your robots.txt allows them.

Common mistakes

• Adding schema instead of content. Markup describes a page; it does not replace the text a model quotes.
• Tag-manager JSON-LD. Invisible to non-rendering crawlers and read late by Google.
• Stale prices and hours in markup that no one updates.
• Duplicate graphs from an SEO plugin and a theme or shop plugin at the same time.
• Paying for "AI schema". No such type exists; ask what markup it actually outputs.

Questions people ask

Does schema markup help you get cited by ChatGPT?

There is no evidence either way from OpenAI for ordinary pages. OpenAI says its shopping results use structured metadata such as price and product description from first-party and third-party providers, which makes Product markup and a merchant feed relevant for shops. For other pages, ChatGPT cites what its search finds and reads; clear visible text matters more than markup.

Is structured data required for Google AI Overviews?

No. Google's guide to its generative AI features says structured data is not required to appear in AI Overviews or AI Mode, and that there is no special schema for them. A page needs to be indexed and eligible to show a snippet. Structured data still helps with rich results in normal search, and it must match what is visible on the page.

Which schema types matter most for AI search?

The ones that state facts an answer needs: Organization with logo and sameAs links, Product with Offer for price and availability, LocalBusiness for address and opening hours, and Article with author and dates. No AI operator has named a preferred type, so choose by what your site is and what Google and Bing already use for rich results.

Can AI crawlers read JSON-LD in the page?

They can fetch it when it is in the HTML the server sends, because it is plain text inside a script element. Whether an AI system keeps or discards that block when it prepares the page for the model is undocumented and varies. JSON-LD injected by JavaScript is never seen by crawlers that do not render. Keep markup server-side and repeat its facts in visible text.