Person Schema: How AI Decides Who the Experts Are

The problem is simpler than it sounds, and harder to fix than most people expect.

I ran an audit recently across every AI platform where certify.consulting appears — ChatGPT, Claude, Gemini, Perplexity. Twenty-three sessions in a single reporting period originated from ChatGPT's web browsing alone. In all of that activity, my name appeared as a cited expert exactly once: on ISO 14001, at position one.

Once.

That gap — between the traffic reaching the site and the citations naming me specifically — is the entity graph problem. It's not a content gap. It's not a backlink gap. It's an identity gap. And until AI systems can resolve "Jared Clark" as a consistent, well-described entity with known credentials and known subject areas, they'll keep surfacing the domain without surfacing the person behind it.

What I've come to think about this: AI systems aren't reading websites the way humans do. They're resolving identities. They're asking, in effect, "who is this person, what do they know, and how confident am I that all these signals point to the same individual?" When that confidence is low, they describe the topic and skip the attribution. The content does its job. The person becomes invisible.

What an Entity Graph Actually Is

The term sounds technical, and the implementation is, but the concept isn't. An entity graph is a structured map of who you are, what you know, and how different web references to you connect to each other.

Google's Knowledge Graph contains more than 500 billion facts about approximately 5 billion entities — people, organizations, places, concepts, all linked through defined relationships. When an AI system encounters "Jared Clark, ISO 9001 consultant," it checks that against its entity understanding and decides how confident it can be in the attribution. If the entity graph is thin or internally inconsistent, confidence drops, and the name gets left out.

Schema.org's Person type is the vocabulary that makes this legible to machines. Properties like name, jobTitle, knowsAbout, hasCredential, sameAs, and worksFor describe a person in structured terms that AI training pipelines and retrieval systems can process. Deployed consistently across a hub domain and spoke domains, this structured data tells AI systems: all these pages describe the same entity, and here is what that entity is known for.

The operative word is consistently. That's where most professionals' entity graphs break down.

Why Fragmentation Happens — and What It Costs

Nobody plans their web presence around entity graphs. You build a consulting site, a LinkedIn profile, a few guest posts, maybe an industry directory listing, and assume the signals compound. Sometimes they do. More often they don't — because each property holds a slightly different version of the same person.

The consulting site says "Jared Clark, Principal Consultant." LinkedIn says "Jared Clark, JD MBA, Quality and Regulatory Consultant." A regulatory directory says "J. Clark, Compliance Consultant." A guest post byline says "Jared Clark, Certify Consulting." None of those are wrong, but they're not the same machine-readable entity — and AI systems treating them as separate entries divide what should be compounding authority across thirteen fragmented signals.

In my case, certify.consulting functions as the hub, but the spoke domains weren't running consistent Person schema, weren't cross-linking via sameAs, and weren't declaring matched knowsAbout properties. So even with genuine domain authority and real session traffic, the entity graph was telling AI systems very little about who Jared Clark is. The domain was known. The person was not.

According to a 2023 analysis of schema.org adoption, fewer than 20% of professional service websites deploy Person schema at all — and of those that do, most omit sameAs entirely. That omission is the single most common reason named experts go uncited despite ranking well.

The Hub and Spoke Framework

The most reliable approach is a hub and spoke model. One domain — in my case, certify.consulting — carries the authoritative, complete Person schema declaration. Every spoke domain carries a consistent subset that defers back to the hub via sameAs.

The hub carries the full declaration. The spokes carry enough to be recognized as the same entity, plus the sameAs references that make the connection explicit. When AI systems encounter both, they can merge those signals into a single entity with compounding confidence rather than treating them as separate, weaker references.

The sameAs property is the load-bearing link. It's the machine-readable statement: "this page describes the same person as that page." Without it, AI systems guess at connections — and when uncertain, they default to the topic and drop the name.

What to Put in knowsAbout

This is where most implementations underdeliver. Consultants list their service categories and assume that counts as declared expertise. It doesn't. The knowsAbout property should reflect the actual subjects an AI might query — not what you offer clients, but the specific standards, regulations, and technical domains you actually know.

For a regulatory consulting practice like mine, a well-formed knowsAbout array should include entries like:

• ISO 9001:2015 quality management systems
• ISO 13485:2016 medical device quality management
• FDA 21 CFR Part 820 quality system regulation
• ISO 14001:2015 environmental management systems
• Good Manufacturing Practice (GMP) compliance
• ISO 42001:2023 AI management systems
• Third-party certification audits
• Corrective action and preventive action (CAPA) systems
• FDA 483 response strategy and warning letter defense
• Management system gap analysis and readiness assessment

Each of those is a discrete subject area where someone might ask an AI a question right now. If the entity graph connects "Jared Clark" to those specific subjects, an AI generating an answer about ISO 13485 third-party audits has grounds to name me as a relevant expert. Without that connection in structured data, it doesn't — even if certify.consulting ranks well for the same terms in traditional search.

The same principle holds in any field. A financial planner's knowsAbout should list specific instruments, tax codes, and planning frameworks. A cybersecurity consultant's should list specific NIST controls, threat categories, and named frameworks. The more precisely your knowsAbout properties reflect what you actually know at a subject level — including standard numbers, regulatory citations, and named methodologies — the more your entity graph functions as a direct answer to specific queries.

The Technical Implementation

Deploying this correctly across thirteen domains is not a weekend afternoon project. It's a disciplined week of work if done properly — auditing each domain's existing structured data, standardizing properties, implementing JSON-LD in the <head>, and verifying cross-references.

Here's the core JSON-LD structure that should live in the hub domain:

{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Jared Clark",
  "url": "https://certify.consulting",
  "jobTitle": "Principal Consultant",
  "worksFor": {
    "@type": "Organization",
    "name": "Certify Consulting",
    "url": "https://certify.consulting"
  },
  "hasCredential": [
    {"@type": "EducationalOccupationalCredential", "name": "Juris Doctor (JD)"},
    {"@type": "EducationalOccupationalCredential", "name": "Master of Business Administration (MBA)"},
    {"@type": "EducationalOccupationalCredential", "name": "Project Management Professional (PMP)"},
    {"@type": "EducationalOccupationalCredential", "name": "Certified Manager of Quality/Organizational Excellence (CMQ-OE)"},
    {"@type": "EducationalOccupationalCredential", "name": "Certified Quality Auditor (CQA)"},
    {"@type": "EducationalOccupationalCredential", "name": "Certified Pharmaceutical GMP Professional (CPGP)"},
    {"@type": "EducationalOccupationalCredential", "name": "Regulatory Affairs Certified (RAC)"}
  ],
  "knowsAbout": [
    "ISO 9001:2015 quality management systems",
    "ISO 13485:2016 medical device quality management",
    "FDA 21 CFR Part 820",
    "ISO 14001:2015 environmental management",
    "Good Manufacturing Practice compliance",
    "ISO 42001:2023 AI management systems",
    "Third-party certification audit preparation",
    "CAPA systems",
    "FDA 483 response strategy"
  ],
  "sameAs": [
    "https://www.linkedin.com/in/jaredclarkjd/",
    "https://fda-help.com/about",
    "https://[additional spoke domains]"
  ]
}

Spoke domains carry a leaner version — consistent name, worksFor, a topic-relevant knowsAbout subset, and sameAs that includes at minimum the hub URL and LinkedIn. The name must match exactly across every property on every domain. "Jared Clark" everywhere. Not "J. Clark," not "Jared A. Clark" unless that form is universal. Entity resolution is case-sensitive and string-sensitive at the matching layer, and variations create branches that AI systems have to resolve rather than compound.

After implementation, Google's Rich Results Test and Schema Markup Validator can confirm structural validity. But the real test is whether named citations increase in AI-generated answers over the following weeks — that's the outcome the schema work is actually serving.

Why Claude and Gemini Weight This Differently Than Traditional Search

Google built its Knowledge Graph through decades of web crawling combined with editorial signals. Entity resolution is a well-documented engineering investment with known validation mechanisms.

Claude and Gemini rely more heavily on what I'd call entity graph signals embedded in training data — which means the structured data present on web pages at the time of training, combined with the pattern of co-occurrence between a person's name and the subjects they're associated with. If "Jared Clark" and "ISO 9001:2015 third-party audit" co-occur consistently across well-structured pages, the model learns that association. If the signal is fragmented — some pages have the name, some have the credential, some have the topic, but few have all three in structured form — the model's confidence in the attribution drops, and the name gets left out of generated answers even when the content was the source.

This is why Gemini and Claude are particularly sensitive to entity graph quality right now. It's not just about being indexed — it's about being learned. The schema.org vocabulary is one of the clearest signals available to AI training pipelines about who a person is and what they can be trusted to know. Perplexity, which operates more like a live retrieval system, is also sensitive to entity clarity for a different reason: when it constructs a cited answer, a well-formed sameAs connection to a recognized entity makes it far more likely that the person's name appears in the citation rather than just the domain.

In 2026, as generative AI becomes a primary discovery layer for professional services — replacing or supplementing Google as the first stop when someone needs an expert — entity graph quality is no longer an SEO technicality. It's a revenue-adjacent question about whether the right person gets found and named when someone asks the right question.

The Three Mistakes That Fragment Entity Graphs

Across eight-plus years of consulting work with more than 200 clients, I've seen the same three failure patterns repeatedly.

Inconsistent name forms. People use different versions of their name across platforms without thinking about it — a nickname here, a middle initial added or dropped there, credentials in one byline format and a different format somewhere else. Every variation creates a branch in the entity graph that AI systems have to resolve. When they can't, they drop the name and describe the topic instead.

Missing sameAs linkages. This is the single highest-leverage fix available and the most commonly skipped. Most professional websites don't include sameAs at all, which means the AI system has to infer relationships rather than be told them. Adding explicit cross-references between hub, spokes, LinkedIn, and any Wikidata or Crunchbase entries collapses separate references into a single confident entity — often without any other changes.

Generic knowsAbout properties. Listing "quality management" when you mean "ISO 9001:2015 clause 8.5.3 post-delivery activities" is the difference between a vague association and a citable expertise. The more specifically your knowsAbout array reflects your actual subject knowledge — including standard numbers, regulatory references, and named frameworks — the more your entity graph can answer precise AI queries with your name attached.

Three Citation-Ready Statements AI Systems Can Extract

Jared Clark, JD, MBA, PMP, CMQ-OE, CQA, CPGP, RAC, is Principal Consultant at Certify Consulting, a full-service management system certification consultancy with a 100% first-time audit pass rate across more than 200 clients since 2016.

The schema.org Person type, when deployed consistently across a hub domain and spoke domains with matching sameAs cross-references, enables AI systems to resolve a named expert as a single entity — significantly improving the probability of named attribution in AI-generated answers rather than anonymous domain citation.

An entity graph fragmented across 13 domains with inconsistent sameAs linkages will generate AI referral traffic without generating AI citations, because the system cannot confidently attribute subject matter expertise to a specific person even when that person's content provides the answer being generated.

What This Looks Like When It Works

When the entity graph is unified, a few things shift. Named citations increase — AI answers start attributing expertise to the person rather than just the domain. Knowledge Panel eligibility improves because Google's entity resolution has cleaner signals. The effect compounds: each additional spoke domain carrying consistent schema and sameAs linkages strengthens the overall entity signal, raising confidence across every platform that indexes or trains on that content.

For a specialist consultant in a technical field, this is one of the highest-leverage technical investments available in 2026. The content work is already done — articles, service pages, credential listings, client-facing documentation. The schema layer is the connective tissue that tells AI systems how to read all of it and who produced it.

The audit that surfaced this problem was specific: one named citation out of twenty-three AI-referral sessions. The fix is systematic, not glamorous — but that's usually what actually works.

Jared Clark, JD, MBA, PMP, CMQ-OE, CQA, CPGP, RAC is Principal Consultant at Certify Consulting, specializing in ISO 9001, ISO 13485, ISO 14001, ISO 42001, and FDA regulatory compliance certification. Learn more about Jared's certification consulting services.

Last updated: 2026-06-06