What is schema.org: JSON-LD Markup for AI Engines in Plain Language
Schema.org is one of those technologies everyone has heard of and almost no one has implemented. The spec has 800+ types. In practice, you need about six to get into AI answers. The rest is for search-engine enthusiasts.
This article skips the encyclopedia. We cover the schema.org minimum for AI: what it is, why it matters, which JSON-LD types to add first, and how to validate the result in a minute.
What Schema.org Really Is — in Plain Terms
The usual definition — "schema.org is a structured-data vocabulary for search engines" — is technically correct and completely useless for understanding. Let's try differently.
Imagine handing a person a business card: "Acme Flowers, 5 Forest St, +1 555 123-4567, flower delivery". In a second they understand: it's a company, here's the address, the phone, what they do.
Now imagine handing a crawler a web page. The HTML is all mixed together: menu, footer, sidebar, text, images, scripts. The bot has to guess: is this a store or a blog? Where's the address, where's the phone? Is the author a company or a person?
Schema.org is that same business card, but for bots and AI. You place a JSON-LD block in the page code that says plainly: "I'm an organization. My name is this. My site is here. My contacts are these. This page is an article in English, published on that date." And the bot no longer has to guess.
JSON-LD markup looks like ordinary JSON inside a script tag of type application/ld+json in the page head or body. It has zero effect on how the site looks — users never see it. But the AI does. And that changes everything.
Terminology: schema.org, JSON-LD, structured data
These terms are often used interchangeably. That isn't quite right:
| Term | What it actually is |
|---|---|
| Schema.org | The vocabulary — a registry of types and properties |
| JSON-LD | The format — how to write data from that vocabulary |
| Structured data | The umbrella name for the result |
| Microdata / markup | Casual synonyms for all of the above |
In conversation, all of these usually mean the same thing: "the markup that search engines and AI read".
Why AI Engines Need Schema.org in 2026
Classic SEO used schema.org for rich snippets: rating stars, breadcrumbs, FAQ accordions. That still works. But since 2025–2026 a different goal has taken priority — appearing in AI search answers (this is what AEO optimization is about).
The difference is fundamental. Markup used to affect how nicely a site looks in results. Now it affects whether it appears in those results at all.
How AI engines use Schema.org
- They identify the content type. With Article or FAQPage markup, the AI immediately knows what it's dealing with instead of classifying text by indirect cues.
- They extract facts without distortion. JSON-LD is already structured data — no need to parse HTML and guess where the date or the author is.
- They assess source trust. Organization with contacts, Article with author and date, FAQPage with real Q&A — all signal "real source, citable".
- They build a knowledge graph. Linking entities: company → products → expert articles → third-party reviews. Schema.org gives entry points into that graph.
Bottom line: without structured data a site is nearly invisible to ChatGPT, Perplexity, Gemini and Google AI Overviews — even if it ranks #1 in regular search.
JSON-LD, Microdata, RDFa — the Difference and What to Use
Schema.org can be written three ways. Two are worth knowing so you don't get confused. One is worth using.
| Format | What it looks like | Status in 2026 | Verdict |
|---|---|---|---|
| JSON-LD | A separate script block in the head | The standard, recommended by Google, works in all AI engines | ✅ Use only this |
| Microdata | itemscope, itemprop attributes inside HTML tags | Outdated, breaks on redesign, AI reads it reluctantly | ❌ Avoid |
| RDFa | property, typeof attributes inside tags | Niche, harder to maintain, not relevant for AI | ❌ Avoid |
Why JSON-LD wins
- Lives separately from the layout — a redesign doesn't break it
- Generated by AI in 30 seconds
- Easy to validate — copy the block, paste into a validator, see the errors
The format is described in the open json-ld.org specification. A minimal JSON-LD for an organization:
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "My Company",
"url": "https://example.com",
"logo": "https://example.com/logo.png",
"contactPoint": {
"@type": "ContactPoint",
"email": "info@example.com",
"contactType": "customer support"
}
}This block tells the AI: "Here's an organization. Name — this. Site — this. Email — this." Five fields, and a huge difference in visibility.
5 Schema.org Types That Actually Drive AI Visibility
Of the 800+ types, five are critical for AI search. Not eight, not twelve. Exactly five. The rest is optional.
Organization — the foundation
The first thing to mark up. Without Organization the AI doesn't know who is behind the site: a company, an individual, a fly-by-night blog. Required fields: name, url. Recommended: logo, contactPoint, sameAs (social links). Organization gives the AI an anchor for building the knowledge graph.
Article — for content
Every blog post gets Article markup with headline, datePublished, dateModified, author. The AI sees: written on this date, updated then, by this author. Three key E-E-A-T signals in machine-readable form.
FAQPage — the most cited type
In SearchPilot split tests, FAQ markup lifted traffic to commercial pages by 3–8%, and an SEO Clarity case study reported roughly 25% more monthly traffic to the tested page. The logic is simple: FAQPage is ready-made question-answer pairs — ChatGPT and AI search don't need to rephrase anything, they pull the answer straight from the markup.
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "How much is delivery?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Free delivery on orders over $50."
}
}]
}Important nuance: the markup must match the visible text on the page. If the JSON-LD says one answer and the page shows another, the AI treats it as a mismatch and ignores the source.
HowTo — for instructions
If the page is a step-by-step guide, use HowTo. AI loves instructions: they're split into steps, full of specifics, easy to cite. Fields: name, step (an array with position, name and text), totalTime.
BreadcrumbList — navigation
Breadcrumbs aren't just for a pretty snippet. For AI they signal site structure: the engine understands how pages relate and identifies topic clusters more accurately. See BreadcrumbList.
How to Add Schema.org — Three Scenarios
WordPress + an SEO plugin
The simplest path. Rank Math or Yoast SEO automatically generate Article, BreadcrumbList and Organization. FAQPage is added via Gutenberg blocks, HowTo likewise. Five minutes to set up the plugin and the base markup is ready.
No CMS / custom-built site
Add a script of type application/ld+json to the head of the relevant pages — via a code editor or the admin panel. The key rule: don't duplicate types — one page = one Organization (if it's the homepage), one Article (if it's a post).
Generate it with AI
Give ChatGPT or Claude a prompt: "Generate JSON-LD Organization for company Name, site example.com, phone, email, logo at…". The AI returns ready code. All that's left is to paste and validate it.
How to Check Schema.org — Tools You Can Trust
After implementing structured data, validation is a mandatory step. Without it you don't know whether the markup works or sits there as dead weight.
Schema Markup Validator
validator.schema.org is the official tool. It shows which types were detected, which fields are filled, and any errors. It isn't tied to Google — it validates pure Schema.org.
Google Rich Results Test
The Rich Results Test checks whether the search engine can build a rich snippet from your markup. If there are no errors but the rich result isn't supported, you'll see it. Run it after every change.
The routine: implement → copy the URL → paste into the validator → fix errors → repeat. A minute per page.
Common validation errors
- A missing or extra comma in the JSON (the most common — a trailing comma after the last element)
- A relative URL instead of an absolute one (/logo.png instead of https://example.com/logo.png)
- A date not in ISO 8601 format (2026-06-09 is correct, 06/09/2026 is not)
- FAQPage contains answers that don't appear visibly on the page
Schema.org and AI Search: What Changed in 2026
The big shift this year: markup stopped being an optional SEO nicety. It's now a prerequisite for appearing in AI answers.
AI engines used to rely mainly on page text: they read it, analyzed it, and decided whether to cite. In 2026 the algorithms shifted toward structured data — JSON-LD became the priority source of information for ChatGPT, Perplexity and Gemini.
The reason is simple: text must be interpreted (risk of error), while JSON-LD is ready facts. The AI saves compute and lowers the chance of hallucinations by taking data straight from the markup.
What this means for you: without Schema.org the AI can read your site but is less likely to cite it. With markup, the odds of citation rise sharply. The gap between "maybe" and "most likely" is exactly the work that markup does.
Frequently asked questions
Does schema.org work in Bing too, or only Google?
Both. Google uses Schema.org for rich snippets and AI Overviews; Bing reads the same markup for its results and Copilot. AI engines like ChatGPT and Perplexity rely on it heavily — especially FAQPage and Organization.
Can bad Schema.org break my site?
No. JSON-LD is a separate block of data in the HTML. If it has an error, the search engine just ignores the markup — it has no effect on the site's appearance or function. The only risk: systematically deceiving a search engine (marking up ratings you don't have) can get rich snippets disabled for the whole domain.
How long until results after adding Schema.org?
Rich snippets: 2–4 weeks after re-indexing. AI citation: first movement in 2–4 weeks, stable presence in 2–3 months. It depends on the site's authority and the niche's competitiveness.
Which Schema.org should I put on the homepage?
At minimum, Organization and WebSite. If there's an FAQ block, add FAQPage. If the homepage describes services, add Service. More than three types on one page is rarely needed.
Why is JSON-LD better than Microdata?
Three reasons: it lives separately from the layout and doesn't break on redesign, it's generated by AI in 30 seconds, and it's officially recommended by Google. Microdata is an outdated format — not used in new projects.
Do I need a developer to add Schema.org?
For a custom-built site, a developer's help is advisable for the first setup — after that you can add markup via the admin panel. For CMSs and site builders it's often said you don't need a developer, but in practice some comfort with code helps. To be sure the markup is placed correctly, it's safer to involve a developer.
Check Your Site Right Now
Free GEO / AEO audit: JSON-LD, E-E-A-T, schema.org — no sign-up, takes 30 seconds.