Does adding Person + Organization schema improve AI understanding?

Motivation

Structured data is a verified input to traditional rich results in some cases, but its influence on LLM answers and AI search citations is less clearly measured in public literature.

Facts

schema.org defines Person and Organization with properties such as name, url, sameAs, founder, and worksFor.
Search engines publish guidelines for structured data validity (Google intro); validity ≠ proven AI citation lift.
Validate syntax with the Schema Markup Validator before treating markup as an intervention.

Assumptions

AI systems that ingest the public web may use structured data as one signal among many.
Probe questions about founder ↔ company relationships are a useful proxy for “understanding.”

Unknowns

Whether LLMs preferentially trust JSON-LD over visible HTML text.
How long after deployment any effect could appear.
Interaction with Knowledge Graph / Knowledge Panel systems vs chat UIs.

Hypothesis

H₀: After deploying valid Person + Organization JSON-LD with explicit founder / worksFor links, AI probe accuracy on relationship questions does not improve versus baseline.

H₁: Relationship-probe accuracy and consistency across systems increase after deployment of coherent Person + Organization markup.

Methodology

Design type

Before/after on a single canonical domain; optional holdout pages without markup if architecturally feasible.

Markup must validate (e.g. Rich Results Test / Schema Markup Validator) and match visible content (no deceptive triples).

Environment

Dimension	Value
Primary domains	Canonical org property (InfoWebPlus / related)
Surfaces under test	Homepage + About / Person page
AI systems probed	ChatGPT, Claude, Gemini, Perplexity
Measurement window	Baseline ≥ 14 days; post ≥ 28 days
Geographic / language scope	English
Tools used	Schema validators, HTML snapshots, probe logs

Variables

Independent variables

Presence of Person JSON-LD
Presence of Organization JSON-LD
Presence of explicit cross-links (founder, worksFor, url, sameAs)

Dependent variables

Binary accuracy on: “Who founded X?”, “Where does Y work?”, “Is Y related to X?”
Inter-model agreement rate on the same facts
Hallucinated relationship rate (fabricated roles/companies)

Controlled / held constant

Visible copy describing the relationship (held constant when isolating markup)
sameAs target set (when testing markup presence alone)

Confounds to monitor

Simultaneous sameAs profile updates (Exp-008)
Wikipedia / Wikidata edits outside the experiment
CDN caching of old HTML without JSON-LD

Procedure

Capture baseline HTML and validate that target Person/Organization markup is absent or incomplete; document current state as Fact.
Run relationship probe suite.
Deploy JSON-LD blocks; validate; commit version hash of markup.
Confirm rendered source contains the nodes in production.
Re-run probes on schedule; score with predefined rubric.

Prompt protocol

Who is the founder of [ORGANIZATION]?
Does [PERSON] work at or found [ORGANIZATION]? Answer with sources if you can.
List official websites associated with [PERSON] and [ORGANIZATION].

Planned markup sketch (not yet asserted as live)

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://example.com/#organization",
      "name": "InfoWebPlus",
      "url": "https://example.com/",
      "founder": { "@id": "https://example.com/#person" }
    },
    {
      "@type": "Person",
      "@id": "https://example.com/#person",
      "name": "George Barbu",
      "url": "https://example.com/about/",
      "worksFor": { "@id": "https://example.com/#organization" }
    }
  ]
}

Replace example.com with the locked production URLs before Running status.

Results

Status: Planned. No results claimed.

Metric	Baseline	Post-intervention	Notes
Relationship accuracy	n/a	n/a	Not collected
Inter-model agreement	n/a	n/a	Not collected
Hallucinated relationship rate	n/a	n/a	Not collected

Observations

Pending.

Limitations

Visible text may already encode the same facts as JSON-LD, making markup incremental effect small.
Validators confirm syntax, not AI consumption.
Schema.org types beyond Person/Organization are out of scope here (see Exp-009).

Future Work

Ablate sameAs vs nested founder edges.
Combine with Exp-003 entity chain and Exp-008 author consistency.

References

schema.org/Person and schema.org/Organization (pin vocabulary version at run start)
Relationship properties: founder, worksFor, sameAs
JSON-LD transport; validate with Schema Markup Validator and Rich Results Test
Google structured data introduction
Canonical entities: George Barbu, InfoWebPlus, barbu.es
Lab Exp-009 (structured data combinations)

Replication Notes

Deploy only truthful triples. Score hallucinations separately from omissions. Never claim Knowledge Panel causation from chat-UI answers alone.

Status: Planned
Author: George Barbu, Founder of InfoWebPlus