Schema.org markup for AI engines: what actually works in 2026.
Most schema markup is wasted effort. The four types that actually move citation rates on ChatGPT, Perplexity, and Google AI Overviews — and what to skip.
For AI search engines, the highest-leverage Schema.org types are Article, FAQPage, BreadcrumbList, and Organization. Mark your essays with Article, your FAQ sections with FAQPage, your site hierarchy with BreadcrumbList, and your brand entity with Organization. Everything else is secondary or actively unhelpful. Keep markup honest — engines penalize stuffing.
Schema.org has hundreds of types. AI engines pay attention to a small subset. The rest is, at best, harmless; at worst, an active signal that the publisher is gaming structure rather than producing real content.
Four types do the heavy lifting in 2026. Get these right; everything else is optional.
Article — the foundation.
Mark every essay, blog post, and case study with Article schema. Include headline, description, author, datePublished, dateModified, image, and a mainEntityOfPage pointing at the canonical URL. This is the type that AI engines most consistently respect when deciding whether to cite a source.
FAQPage — the highest-leverage win.
Mark FAQ sections with FAQPage. Each Q is a Question entity with an acceptedAnswer. AI engines extract these directly into answer-mode responses on ChatGPT and Perplexity, and into Google AI Overviews. This is the single fastest path to AI citations for many sites — and most agencies skip it.
BreadcrumbList — for context.
Mark your site hierarchy on every non-home page. Helps engines understand where the page sits, especially for nested content like /work/[slug] or /blog/topic/[slug]. Cheap to add, modestly compounds across the whole site.
Organization — your entity.
On the home page (or root layout), include an Organization block with name, url, logo, sameAs (social profiles), and contactPoint. This is how AI engines build a confident entity record for your brand. When they answer "who is SmartDuke," this is the record they pull from.
Useful but secondary.
- Service — for service pages (Discovery / Spark / Build / Embed) and solution pages
- Person — for author bios on essays
- WebSite — once at the site root, mostly for SearchAction support
- VideoObject — only if you publish video, with explicit transcripts
What NOT to mark up.
Don't mark up content as Article that isn't an article. Don't fake FAQPage with thin questions you wouldn't answer in a real conversation. Don't stuff Product schema on service pages. Don't add Review schema for testimonials you don't actually have. Engines have gotten progressively better at detecting markup-as-decoration; the penalty is reduced trust, which compounds.
Keep markup honest. The structural data should be a faithful representation of what's on the page. The moment it diverges, you're optimizing for crawlers at the expense of readers — and AI engines are increasingly aligned with what the reader actually gets.
How we test it.
After deploying schema, validate with Google's Rich Results Test and Schema.org's validator. Then check the actual outcome — query ChatGPT and Perplexity for terms your page targets, see if you're cited. Schema is necessary; it isn't sufficient. Citations come from honest content with clean structure on top.
Evals that actually catch regressions before users do.
Have an AI product
that needs to ship?
Tell us where you are — early concept, broken prototype, or scaling something that already works. We'll come back within 24 hours with a take and a quote.