Does Google penalise sites using Microdata or RDFa instead of JSON-LD?

No. Google supports all three Schema.org syntaxes for rich results, and pages will not be downranked for choosing one over another. The recommendation toward JSON-LD is operational — easier to maintain, less error-prone, better tooling.

Do AI engines like ChatGPT and Perplexity actually read JSON-LD?

Yes, but inconsistently. Independent testing suggests they treat structured data more like additional on-page text than as a validated graph. They benefit from JSON-LD because it is text-rich, factual, server-rendered, and easy to read alongside the page body — not because they execute Schema.org rules.

Can I keep my existing Microdata and add JSON-LD on top?

Technically yes, but it doubles maintenance and risks contradictions when one source updates and the other does not. Prefer migrating fully to JSON-LD and removing the Microdata once parity is verified in the Rich Results Test.

What about JSON-LD injected by tag managers?

Tag-manager-injected JSON-LD lands in the DOM after first paint. Gemini and Google AI Mode can usually still read it; ChatGPT, Claude, and Perplexity often cannot. Server-render whenever possible.

Is RDFa worth learning in 2026?

Only if you publish into Linked Data ecosystems beyond Schema.org. For SEO and AI-search outcomes alone, JSON-LD is faster to implement and produces equal or better results.

JSON-LD vs Microdata vs RDFa for AI Search: Markup Format Comparison

JSON-LD is the recommended structured data format for AI search visibility because LLM crawlers parse server-rendered, script-isolated JSON-LD blocks more reliably than the inline HTML attributes used by Microdata and RDFa. Choose JSON-LD by default; reach for Microdata or RDFa only when CMS or compliance constraints make it unavoidable.

TL;DR

JSON-LD wins for AI search. Google explicitly recommends it, server-rendered JSON-LD is the most parser-friendly format for LLM ingestion, and isolating structured data in a