Schema markup for LLMs — the 2026 spec sheet
Schema.org JSON-LD is now load-bearing infrastructure for AI search. Here are the eight schema types that move the needle and the ones that don't.
Why schema matters more for AI than it ever did for Google
Google could afford ambiguous pages because Google had decades of behavioural data — clicks, dwell time, query reformulations — to disambiguate. AI engines have no such backstop. They have to decide now, from a single fetch, whether your page is talking about a product, an article, a person, an organisation. Schema is how you tell them, machine-readably, what you are.
The eight schema types that matter
1. Organization
Ship Organization schema on your homepage and /about. Include name, url, logo, sameAs (LinkedIn, Crunchbase, Wikipedia URLs). This is what AI engines use to resolve „who is X?" — without it, your brand may merge with a similarly-named entity.
2. WebSite
Ship WebSite schema once at the root, with optional SearchAction. Used for sitelink expansion in Google AIO and increasingly for site-internal navigation in answer surfaces.
3. Article (and BlogPosting / NewsArticle)
Every blog post and guide should ship Article with headline, author (as Person with url), datePublished, dateModified, image, inLanguage. The dateModified field disproportionately matters — retrieval-grounded engines bias toward fresh.
4. FAQPage
Mark up FAQ sections at the bottom of cornerstone pages. AI engines use these as direct answer sources. Single highest ROI schema type after Organization.
5. Product
For SaaS / e-commerce: Product with name, description, brand, offers (with price, priceCurrency, availability), and ideally aggregateRating if you have legitimate reviews. AI shopping interfaces increasingly query for this.
6. HowTo
For tutorial / instructional content. Each step as a HowToStep. Powerful for Google AI Overviews, which loves to reproduce step lists.
7. Person (Author)
Author profiles need Person schema with name, jobTitle, worksFor, sameAs (LinkedIn, Twitter). E-E-A-T signal that AI engines actively check.
8. SoftwareApplication
If you sell software, this is your meta-page schema. applicationCategory, operatingSystem, featureList, offers. Helps AI tooling recommend you in „best X for Y" queries.
Schema types that don't move the needle (anymore)
Some schemas were oversold during the SEO era and have minimal AI-search impact today: BreadcrumbList (informational only), Review outside Product (heavily down-weighted due to past abuse), VideoObject without an actual hosted video, Recipe outside food contexts. Don't waste cycles.
How to validate
Use Schema.org's official validator and Google's Rich Results Test. Both will flag missing required fields and invalid types. For LLM-specific validation, an AI-search visibility platform can replay your prompt set and show whether the model actually picked up your schema as a signal — the only end-to-end test that matters.
Implementation pattern: JSON-LD in <head>
Always ship as <script type="application/ld+json"> in <head>. Microdata (itemprop attributes) and RDFa work but JSON-LD is what every major AI engine actually parses. One JSON object per type, or a @graph array if multiple types apply.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Article",
"@id": "https://yoursite.com/post#article",
"headline": "Title here",
"datePublished": "2026-04-01",
"dateModified": "2026-04-22",
"author": { "@type": "Person", "name": "Jane Doe" },
"publisher": { "@type": "Organization", "name": "YourBrand" }
},
{
"@type": "FAQPage",
"mainEntity": [
{ "@type": "Question", "name": "What is X?",
"acceptedAnswer": { "@type": "Answer", "text": "X is..." } }
]
}
]
}
</script>