JSON-LD schema for AI search

What is JSON-LD schema?

It’s a structured layer that tells machines what a page is and who stands behind it, added as a script block that never touches the visible content. Readers see the page; crawlers and AI engines read the schema to understand it without guessing. It’s the difference between hoping a machine infers your meaning and stating it outright.

In plain terms, JSON-LD is a small block of labelled data that lives in the page’s code, written in the schema.org vocabulary every major search engine and AI model understands. Where a human reads a price and knows it’s a price, a machine has to be told, and the schema is how you tell it: this string is the product name, this number is the price, this name is the author. Because it sits separately from the visible layout, you can describe the page precisely without changing a single pixel of what the reader sees, which is exactly why it has become the default way to speak to crawlers.

Which schema types matter for AI search?

Organization and Person carry identity and the experience signals AI weighs; Product and Offer describe what you sell; FAQPage works when the answers genuinely appear on the page; BreadcrumbList and Article give structure and context. Pick the types that match what the page actually is. Identity markup especially helps AI engines decide whether to trust and cite you.

For AI search specifically, the identity types do the heaviest lifting. When a model decides whether to cite a source in a generated answer, it is weighing whether the page comes from someone it can attribute and trust, and a well-formed Person or Organization block, linked out to verified profiles, is a direct signal of that. The content types matter too, but in a supporting role: an Article block frames the page as a considered piece rather than a thin landing page, and a genuine FAQPage can surface your exact wording inside an AI answer. The mistake is reaching for every type at once; the right move is to mark up only what the page truly is and let the relevant signals stand out.

How does an AI agent generate schema reliably?

It builds the markup from the page’s real content, not a generic template, so the schema describes what’s actually there. The hard rule is never marking up something that has no visible counterpart; FAQ schema without visible FAQs is a spam signal. Reliable generation means the structured data and the page always tell the same story.

This is where an agent earns its place, because reading a page and emitting accurate markup is exactly the rule-bound, repeatable work it does well. Fed the actual page, it pulls the real title, the real author, the real questions and answers, and writes them into the correct schema fields, instead of pasting a boilerplate block that quietly claims things the page never says. The discipline that keeps it honest is a single rule applied every time: no field gets filled unless a reader could point to its source on the page. That constraint is what separates schema that helps from schema that gets a page flagged.

How do you validate structured data?

Run it through the Rich Results test and a schema validator, and confirm the JSON itself parses cleanly. Then do the check the tools skip: make sure every schema claim has a matching visible element on the page. Valid syntax that contradicts the content still fails, because the schema-to-content match is the real test.

Validation has two layers, and most people only run the first. The machine layer, parsing cleanly and passing the Rich Results test, confirms the syntax is correct and eligible for rich features. The trust layer, which no tool checks for you, confirms that every claim in the markup is backed by something a reader can actually see. A page can pass every automated test and still be marking up FAQs that do not exist or a rating no one left, and that mismatch is precisely what search engines penalise. The real test is consistency: the structured data and the visible page have to tell one story.

This guide sits inside the wider how to automate SEO and AEO with Claude workflow, and the identity markup above is what makes the difference in how to get cited by AI search. The reliable schema-from-content discipline is one of the workflows packaged in the SEO & AEO Pro Kit.