JSON-LD for AI Search: A Practical Structured Data Guide

Updated 2026-07-13 · 13 min read

JSON-LD gives machines an explicit graph of entities and relationships already visible on a page. It can reduce ambiguity and qualify pages for supported search features. It cannot make weak content authoritative, force an LLM citation or replace crawlable HTML. The best implementation is accurate, connected and boring.

Start with stable entities

Define one Organization and one WebSite with persistent @id values, then reference those IDs from every page. Do not create a new slightly different organization entity on each route. Use canonical absolute URLs.

{"@context":"https://schema.org","@type":"Organization","@id":"https://example.com/#org","name":"Example","url":"https://example.com/","logo":"https://example.com/logo.png"}

Choose schema by what the page is

Connect the graph

An article’s publisher should reference https://example.com/#org; its isPartOf references #website; mainEntityOfPage points to the canonical page entity. Breadcrumb items use canonical page URLs. These links tell parsers that repeated nodes are the same entity.

Model one entity graph across the site

Think beyond isolated snippets. The homepage defines Organization and WebSite. An article references those nodes instead of redefining them. A category is a CollectionPage whose ItemList links to canonical article or product pages. The AboutPage is about the same Organization ID. This connected graph is easier to maintain and less ambiguous than twenty independent blocks with slightly different names and logos.

Generate schema server-side from the same records used for titles, visible headings and canonical links. If a price, date or item count changes, one data update should change both HTML and JSON-LD. Parallel hand-maintained facts inevitably drift.

Match visible content exactly

Never add reviews, prices, FAQs or authors that a visitor cannot verify. During a free launch, do not leave paid offer schema active while hiding the price from the page. Keep dates real: dateModified changes only when content meaningfully changes, not on every server start.

JSON-LD for articles

A strong Article node includes headline, description, canonical URL, image, datePublished, dateModified, inLanguage, author, publisher, articleSection, keywords and wordCount. Add citations when the article relies on external primary sources. The structured fields should be generated from the same content object that renders the page to prevent drift.

JSON-LD for directories

Use CollectionPage as the page and ItemList as its main entity. List position, name and canonical URL for items. Do not dump thousands of entities into every category page; include the visible items or a sensible representative list and let their own pages carry detail.

Does JSON-LD make LLMs recommend a site?

No mechanism guarantees recommendation. LLM-oriented discovery benefits from the same foundations: accessible pages, clear entities, consistent facts, original evidence, internal links and external references. JSON-LD improves machine readability and disambiguation; citations are earned by useful, trustworthy information.

Validation workflow

  1. Parse every JSON-LD block as JSON in automated tests.
  2. Check Schema.org vocabulary and entity IDs.
  3. Run representative pages through Google’s Rich Results Test.
  4. Compare structured claims with visible page content.
  5. Crawl the site and flag missing canonical, duplicate IDs or inconsistent entity names.
  6. Revalidate after template changes.

A prompt for an AI coding agent

Inventory every indexable route and map it to the most specific truthful Schema.org type. Create stable Organization and WebSite @id nodes shared across routes. Generate page JSON-LD from the same server-side data as visible HTML. Add Article fields, BreadcrumbList and CollectionPage/ItemList where appropriate. Do not add invisible claims. Write tests that parse every block and assert canonical URLs, required fields and unique stable IDs.

Frequently asked questions

Does JSON-LD help AI search?

It helps machines understand entities and relationships, but does not guarantee citations or recommendations. Original content, crawlability and authority remain essential.

Which JSON-LD schema should every website have?

Organization and WebSite are useful foundations for most organizations. Page-specific schema should then match the actual page type.

Can I use FAQ schema on every article?

Only when the questions and complete answers are visible and genuinely useful. Search engines may limit FAQ rich results regardless of valid markup.

Sources & further reading

Keep reading

Vibe Coding SEO: How to Make AI-Built Sites Rank14 min readLovable SEO Guide: Make Your App Visible in Search11 min readDesign Tokens Explained: The Vocabulary of Consistent UI6 min read
Browse 350+ measured design breakdowns →