notableReviewed Mon Nov 02

Controlled RAG study finds enhanced linked entity pages outperform plain HTML and JSON-LD alone

Volpini and colleagues tested structured linked data and enhanced entity pages in a controlled standard/agentic RAG pipeline across four domains, reporting larger accuracy gains for navigable entity pages than for JSON-LD markup alone.

Context

Volpini and colleagues tested structured linked data and enhanced entity pages in a controlled standard/agentic RAG pipeline across four domains, reporting larger accuracy gains for navigable entity pages than for JSON-LD markup alone.

What changed

The preprint compared plain HTML, HTML plus JSON-LD, enhanced agent-oriented entity pages, and richer navigational interlinking across seven experimental conditions using Vertex AI Vector Search and Google Agent Development Kit components.

Why it matters

The study directly tests whether machine-readable entities and navigable relationships can improve evidence retrieval and answer completion. It distinguishes adding schema markup from providing richer linked entity pages.

Visibility OS interpretation

Visibility OS interpretation: Entity-level information architecture may improve agentic retrieval in some systems, but the reported accuracy improvements do not demonstrate a Google Search ranking signal or cross-platform GEO citation gains.

What remains uncertain

How much of the gain comes from entity linking, agent instructions, navigation, or neural search individually? Will effects replicate on uncontrolled websites, other retrieval stacks, and real search platforms?

Evidence claims

Emerging Evidenceintroduces

Volpini and colleagues evaluated seven standard and agentic RAG conditions across editorial, legal, travel and e-commerce domains, comparing plain HTML, JSON-LD and enhanced linked entity pages.

Emerging Evidenceintroduces

In the authors' controlled RAG experiments, enhanced entity pages yielded reported accuracy improvements of 29.6% in standard RAG and 29.8% in the full agentic pipeline; the incremental Enhanced+ gain was not statistically significant.