Educational article · July 2026

JSON-LD for AI discovery: an evidence-first approach

Structured data works best when it is a compact, testable representation of information people can already verify on the page.

JSON-LD is a way to express entities and relationships in a machine-readable form. It can help a consuming system identify a page’s subject, its author, a service, an organization, or a set of visible questions and answers. It does not replace clear writing, reliable source material, or an accessible page.

Good markup clarifies an already-supported fact. It does not manufacture one.

Begin with the page, not the schema type

It is tempting to select an attractive vocabulary first and then fill its properties. A safer workflow starts by collecting the page’s visible claims and its authoritative identity facts. Only then should a reviewer choose a schema type that accurately describes the material.

For example, a resource with a byline, publication date, explanatory body, and publisher information may support an article-oriented annotation. A page that merely lists service capabilities should not be labelled as a product with invented offers, ratings, or availability.

Keep the delivery boundary narrow

When markup is delivered through a CDN or another intermediary, it should be treated as a governed content change. The delivery system should apply only a reviewed document to an approved URL, preserve the origin’s visible content, and return unmodified responses whenever the eligibility checks are not met.

  • Generate and validate annotations before a visitor request arrives.
  • Version each approved annotation and retain its supporting evidence.
  • Inject one identifiable managed block, without silently replacing customer-owned markup.
  • Verify the public response independently after activation and after rollback.

Measure what the prototype actually proves

A successful technical delivery proves that an approved annotation can appear safely and consistently in the delivered HTML. It does not prove that a search engine will show a rich result, that an AI system will cite the site, or that traffic will improve. Those are separate outcomes with many contributing factors.

For this demo: the original static page deliberately contains no managed structured-data block. A later edge-delivery phase will make the difference observable in the public response.