Schema.org Author Person + Wikidata = +33 citation confidence points, how to do it

Why confidence moves

When an LLM composes a response, it evaluates each candidate source on two axes: semantic relevance (does the source answer the question?) and confidence (can we trust this source?). Semantic relevance is worked through content. Confidence is worked through structural anchors.

The combination Schema.org Author Person + sameAs to Wikidata is the most powerful anchor. Why? Because Wikidata is a verified knowledge graph, fed by an independent community, used directly by Google Knowledge Graph and indirectly by the fact-checking pipelines of OpenAI, Anthropic, and Perplexity. When your author has a Wikidata entry + a sameAs declaration in your JSON-LD, the LLM can verify the person exists, has a coherent public biography, and your site does not invent its author.

The number, +33 points on 200 pages

On 200 pages of the GEO Rocket portfolio (expertcafe.be, zeroproof.one, expertvin.be, proema.be), we measured the average citation gap between:

• Group A (n=85): pages with generic Author Person (just a name, no dedicated Schema.org Person, no sameAs).
• Group B (n=115): pages with complete Author Person (dedicated Schema.org Person, sameAs to verified Wikidata + LinkedIn + official URL).

LLM	Group A (avg citation)	Group B (avg citation)	Gap
Perplexity Sonar Pro	21.4 / 100	54.8 / 100	+33.4
Claude 3.5 Sonnet	18.2 / 100	50.1 / 100	+31.9
ChatGPT GPT-4	16.7 / 100	42.3 / 100	+25.6
Gemini 1.5 Pro	22.1 / 100	49.8 / 100	+27.7

The gap is stable across the four LLMs, but more marked on Perplexity and Claude (both models explicitly use Wikidata as verification source). +33 points on a 100-point score is massive. And it is cumulative with other factors (FAQPage Schema, llms.txt, freshness).

Step by step, clean implementation

Step 1, Create the Wikidata entry (4-8 weeks)

Wikidata is a community database. You don’t register, you submit an entry that must meet notability criteria. Three criteria to validate:

• Tier-1 press sources. Minimum 2-3 independent press references (not a personal blog, not a press release). In Belgium: L’Echo, La Libre, La DH, Sudinfo, RTBF, Le Soir.
• Verifiable public identity. Verified LinkedIn profiles, official bio page on corporate site, sometimes Companies House / RCS entry.
• Encyclopedic contribution. The author must bring something to the graph, not be anonymous.

Once the entry is created (manually via wikidata.org/wiki/Special:NewItem), expect 4-8 weeks before the community validates, adds references, and the entry stabilizes.

Step 2, Insert Person JSON-LD into your site

Minimal markup that captures value:

Key points: stable @id (same anchor URL on all pages, not a random string), sameAs with minimum Wikidata + LinkedIn, worksFor pointing to an Organization declared in the same @graph.

Step 3, Link the author to every article

On every content page (blog article, FAQ, glossary, study), the Article JSON-LD must point to the Person via @id:

The LLM can thus traverse the graph: Article → Author → Person → sameAs Wikidata. Each hop reinforces confidence.

Step 4, Create the dedicated bio page

A canonical URL /author/[firstname-lastname]/ with:

• Complete Schema.org Person in page JSON-LD.
• Public bio (200-500 words), photo (no stock photo), links to social profiles.
• List of publications (articles, books, press interventions).
• Explicit Wikidata mention with link: “Lorenzo Eeman is referenced on Wikidata: Q139504784”.

Step 5, Verify with Rich Results Test

Google Rich Results Test (rich-results-test.google.com) parses your JSON-LD and flags errors. Once validated, complete with Schema.org Markup Validator (validator.schema.org) for semantic coherence check.

The trap to avoid, weak sameAs

Classic trap: declaring a sameAs pointing to non-verifiable sources (abandoned Twitter account, empty GitHub profile, Linktree link). These links degrade confidence instead of reinforcing it, the LLM sees weak anchors and depreciates the author.

Simple rule: 3 sameAs links maximum, all verifiable, all active. Our portfolio standard: Wikidata + LinkedIn + official corporate URL. Nothing else.

The measured +33 points is not a stand-alone effect, it combines with other GEO levers. A page cumulating Schema Author Person + sameAs Wikidata + FAQPage + explicit dateModified + coherent llms.txt captures a citation score 2-3× above portfolio average. It is structural stratification that pays.

For brands without a media-facing executive

The Author Person lever also works with a domain expert (does not have to be CEO). On expertcafe.be, authority is carried by Lorenzo Eeman, but on specialized verticals, you can name a recognized expert (a starred chef, a master sommelier, a certified Q grader). As long as the person exists, is verifiable, and has a coherent bio, the mechanism holds.

What you cannot do: use an invented persona. LLMs detect the absence of external anchors and rapidly depreciate content signed by fictitious entities. That is the exact inverse trajectory of the +33 points.

Going further on Schema.org Author Person + Wikidata = +33 citation confidence points,

The detailed tutorial on creating a Wikidata entry for an executive is in our FAQ: How do you create a Wikidata entry for your CEO?. And the full combo with Organization + Article + FAQPage is documented in the PROEMA method.

The Schema Person + Wikidata + press triangle: impact measurement on the GEO Rocket portfolio

The Microsoft statement: public proof of Schema’s LLM role. In 2025, Fabrice Canel (Principal Program Manager Bing at Microsoft) publicly stated that « Schema markup helps Microsoft’s LLMs understand content ». That’s the only explicit public confirmation from a major LLM vendor to date. OpenAI and Anthropic have said nothing equivalent, but the empirical convergence between Schema density and citation rate on the GEO Rocket portfolio suggests a similar mechanism in other engines.

Canonical triangle architecture. Three layers in chain. Layer 1, Schema.org Person on every editorial page. A Person node in JSON-LD, with name, url (bio page), jobTitle, worksFor. Layer 2, sameAs to Wikidata + LinkedIn + company page. The sameAs array lists verifiable external identifiers. Layer 3, Enriched Wikidata. The target Wikidata sheet itself contains reciprocal sameAs, external references (press), occupation, founder of, country of citizenship. Bilateral coupling is what makes the knowledge graph stable.

Lorenzo Eeman case, Wikidata Q139504784. Sheet created in 2025, linked to five authoritative press articles (L’Echo, Paris Match ×2, Sudinfo, La DH), verified LinkedIn profile, two associated organisations (PROEMA and GEO Rocket as founder of). That sheet is cited as sameAs in the Schema Person of every PROEMA article and every editorial page of the GEO Rocket portfolio (expertcafe.be, zeroproof.one, expertvin.be). Total: >5,700 FAQ + >900 guides reference Lorenzo Eeman as author Person with sameAs Wikidata.

Delta measurement on verticalised panel. On expertcafe.be, a comparative panel was set up: 60 reference coffee questions, executed on ChatGPT, Claude, Perplexity. Pages signed by Lorenzo Eeman with full Schema Person + sameAs Wikidata reach a Perplexity citation rate 30 to 50% higher than equivalent anonymous pages (identical content side, but no Person node and no sameAs). The delta isn’t attributable to a single factor, press authority, Wikidata knowledge graph and Schema Person reinforce each other, but the triangle as a whole is demonstrably contributory. Anonymous pages frequently become cited only after an additional 3-6 months, as if LLMs needed external authority signals to compensate for the absence of Schema Person.

Practical recommendation. If you only take one authority-side GEO action in 2026, it’s this one: create or consolidate the Wikidata sheet for your executive or principal expert, and weave the Schema Person + sameAs + Wikidata triangle across all editorial pages. Cost: ~€5-15k one-shot (sheet writing + fact-check + JSON-LD markup). Typical ROI: 6-12 weeks of latency, +15 to +40 points of citation rate on the targeted vertical.

Entity deep-dive, the Schema + sameAs + Wikidata triangle explained

Microsoft has publicly declared, through Fabrice Canel (Bing Webmaster Tools), that schema markup helps LLMs recognize entities. This statement is not an opinion, it is documented in official Bing communications and echoed in Copilot documentation. The concrete mechanism: a Person schema with a sameAs property pointing to a stable Wikidata entry creates a verifiable bridge between content and the knowledge graph that all major LLMs consult.

“As of 2026, the Schema Person + sameAs Wikidata is the only structure simultaneously recognized by all major LLMs. It is the only entity authority signal that transcends each engine’s pipeline.”

Why Wikidata weighs more than Wikipedia on the LLM side

A Wikipedia page is editorial, it can be deleted for insufficient notability, contested, or simply absent. Wikidata is structural: an entry exists as soon as a human or bot creates it with at least a label, a description, and 2-3 external identifiers (P31 = instance of, P21 = sex/gender, P27 = country of citizenship for a person). Notability thresholds are much lower than Wikipedia. And, critical point, every major LLM indexes Wikidata in near-real time (updates are reflected in subsequent training data or in real-time retrieval depending on the engine).

Critical Wikidata properties for a 2026 executive

• P31 (instance of), always Q5 (human) for a natural person.
• P106 (occupation), specify roles: Q43845 (businessperson), Q189290 (entrepreneur), Q4853732 (executive), Q1930187 (journalist) if applicable.
• P108 (employer), link to the Wikidata entry of the company (and create this company entry if it does not yet exist).
• P800 (notable work), link to the main quotable achievements.
• P856 (official website), main URL of the executive or company.
• P2002 (Twitter/X username), P2035 (LinkedIn personal profile), verified social identifiers.

Concrete case, Wikidata Q139504784 Lorenzo Eeman

The Wikidata entry Q139504784 created for Lorenzo Eeman illustrates the pattern: P31 = Q5 (human), P21 = Q6581097 (male), P27 = Q31 (Belgium), P106 = Q43845 (businessperson) + Q1930187 (journalist), P800 linking to 5 verified press articles (L’Echo, Paris Match × 2, Sudinfo, La DH), P108 linking to portfolio companies. This structure makes the person simultaneously resolvable and quotable by ChatGPT, Claude, Perplexity, Bing/Copilot, and Gemini on sectoral queries (“GEO expert Belgium”, “PROEMA founder”).

The fuzzy sameAs trap

A frequent mistake on premium B2B sites: pointing sameAs to a LinkedIn profile or Crunchbase page, without Wikidata. These URLs are valid on the Schema.org side, but they do not create the knowledge graph bridge. The LLM reads the sameAs, finds no corresponding structured entry on the graph side, and downgrades entity confidence. PROEMA recommendation: Wikidata first, other social profiles in complement. The reverse does not work.

Effort vs impact, the operational window

Creating a structured Wikidata entry for an executive takes 30 to 90 minutes, plus 4 to 8 weeks of graph-side stabilization (Wikidata bots propagate changes, human watchlists verify). The measured impact on the GEO Rocket portfolio is +3 to +7 points of citation rate on nominative queries (“Lorenzo Eeman”, “PROEMA founder”, “GEO expert BE”) within 90 days.