JSON-LD vs microdata for scaled structured data

JSON-LD vs microdata for scaled structured data

Both produce valid schema. For programmatic pages, one is operationally correct.

1. JSON-LD: a separate <script> block, decoupled from HTML markup. Generate it from the same data object that builds the page. Google's stated preference.
2. Microdata: attributes interleaved into HTML tags. Couples schema to presentation — every template redesign risks breaking your structured data silently.
3. Decision rule at scale: JSON-LD, full stop. Decoupling means a front-end change can't quietly invalidate your schema, and you can unit-test the JSON object independently.
4. Guardrail: validate the generated JSON-LD against the live data — schema claiming a price/rating the page doesn't show is a spam signal and a manual-action risk.
5. Gate: run a sample through the Rich Results test per template before launch; fail on any error, warnings reviewed.
6. Guardrail: only mark up content actually visible on the page.

Rollback: schema is regenerated from data; deploy-reversible.

Ship gate: don't publish until all boxes are checked.