Why This Property Exists

Health content makes claims that should be verifiable. Schema.org provides citation for linking articles to their references, but this only works on CreativeWork types—it describes what sources an article used, not what evidence supports a specific claim within that article.

Health Schema types like CausalStep and ContributingFactor inherit from Thing, not CreativeWork. They need a way to link individual mechanistic claims to their supporting evidence.

The evidenceSource property enables:

• Claim-level verification: Link specific mechanisms or statistics to the research that established them
• Transparency: Show readers exactly which claims are evidence-based and which are inference
• Knowledge graph traversal: Connect health content to the broader scientific literature
• LLM grounding: Provide language models with verifiable sources for factual claims
• Trust signals: Distinguish well-sourced content from unsupported assertions

This is distinct from article-level citations—evidenceSource operates at the granularity of individual claims within a schema.

Domain and Range

Used On (Domain)

Expected Values (Range)

Distinction From Related Properties

Several properties relate to evidence and citations. Here's how they differ:

The Key Distinction

evidenceSource answers: "What research supports this specific claim?"

citation answers: "What sources did this article reference?"

evidenceLevel answers: "How strong is the evidence overall?"

These can be used together—a CausalPathway might have an evidenceLevel of "Strong" and multiple evidenceSource links on its individual steps, whilst the article containing the pathway uses citation for its reference list.

Usage Guidance

Do

• Use persistent identifiers—DOIs, PubMed IDs (PMIDs), or stable URLs that won't break
• Be specific about type—use ScholarlyArticle for peer-reviewed papers, MedicalStudy for clinical trials, WebPage for guidelines
• Link at the appropriate granularity—link individual CausalSteps to specific papers, not entire pathways to vague "research"
• Include identifying information—at minimum, provide @id (DOI/URL) and name (article title)
• Use arrays for multiple sources—when a claim is supported by several studies, list them all

Don't

• Link to paywalled content without alternatives—prefer PubMed (which shows abstracts) over direct journal links when possible
• Use evidenceSource for general background—this is for specific supporting evidence, not "further reading"
• Omit the property when evidence exists—if you have sources, link them; transparency builds trust
• Link to low-quality sources—blog posts, social media, and press releases are not appropriate evidence sources

Identifier Best Practices

Examples

Example 1: CausalStep With PubMed Source

Linking a mechanistic claim to peer-reviewed research:

{
  "@type": "CausalStep",
  "stepNumber": 2,
  "mechanism": "Growth factors bind to receptor tyrosine kinases on target cell surfaces",
  "result": "Receptor dimerisation and autophosphorylation, activating intracellular signaling cascades",
  "evidenceSource": {
    "@type": "ScholarlyArticle",
    "@id": "https://pubmed.ncbi.nlm.nih.gov/19176360/",
    "name": "Platelet-rich plasma: Growth factor enhancement for bone grafts",
    "author": {
      "@type": "Person",
      "name": "Marx RE"
    },
    "datePublished": "2004",
    "isPartOf": {
      "@type": "Periodical",
      "name": "Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology, and Endodontology"
    }
  }
}

Why this works: The source is a peer-reviewed article with a stable PubMed identifier. The @id serves as both identifier and link. Author, date, and journal provide context for evaluation.

Example 2: ContributingFactor With DOI

Linking a risk factor claim to epidemiological research:

{
  "@type": "ContributingFactor",
  "factorName": "Female gender",
  "importance": "Major",
  "mechanism": "Hormonal complexity with less margin for error in sex hormone balance",
  "prevalence": "7.1% of females vs 0.5% of males; 14-fold increased risk",
  "modifiable": false,
  "evidenceSource": {
    "@type": "ScholarlyArticle",
    "@id": "https://doi.org/10.1001/jamadermatol.2023.4112",
    "name": "Association of GLP-1 Receptor Agonists With Alopecia",
    "datePublished": "2023",
    "isPartOf": {
      "@type": "Periodical",
      "name": "JAMA Dermatology"
    }
  }
}

Why DOI: DOIs are persistent identifiers that resolve regardless of where the article is hosted. They're the gold standard for academic citations.

Example 3: Recommendation With Clinical Guideline

Linking a treatment recommendation to an authoritative guideline:

{
  "@type": "PrioritizedRecommendation",
  "priority": 1,
  "category": "Nutritional Foundation",
  "actions": [
    {
      "@type": "MedicalTherapy",
      "name": "Protein optimisation",
      "description": "1.2-2.0 g/kg daily during weight loss"
    }
  ],
  "rationale": "Prevents muscle loss and supports tissue maintenance during caloric restriction",
  "evidenceSource": {
    "@type": "WebPage",
    "@id": "https://www.nice.org.uk/guidance/ph53",
    "name": "Weight management: lifestyle services for overweight or obese adults",
    "publisher": {
      "@type": "Organization",
      "name": "National Institute for Health and Care Excellence",
      "alternateName": "NICE"
    },
    "datePublished": "2014"
  }
}

Why WebPage: Clinical guidelines aren't scholarly articles—they're authoritative documents published by professional bodies. WebPage is the appropriate type.

Example 4: Multiple Evidence Sources

When a claim is supported by several studies:

{
  "@type": "Observation",
  "name": "Hair density improvement with PRP",
  "value": {
    "@type": "QuantitativeValue",
    "minValue": 14,
    "maxValue": 25,
    "unitText": "hairs per square centimetre"
  },
  "description": "Mean increase over 3-6 months in androgenetic alopecia",
  "evidenceSource": [
    {
      "@type": "ScholarlyArticle",
      "@id": "https://doi.org/10.1111/dth.12369",
      "name": "Platelet-rich plasma for androgenetic alopecia: A review",
      "datePublished": "2016"
    },
    {
      "@type": "CreativeWork",
      "@id": "https://doi.org/10.1002/14651858.CD014174",
      "name": "Platelet-rich plasma for androgenetic alopecia",
      "additionalType": "Systematic Review",
      "publisher": {
        "@type": "Organization",
        "name": "Cochrane Database of Systematic Reviews"
      },
      "datePublished": "2024"
    }
  ]
}

Why an array: Quantitative claims are stronger when supported by multiple independent sources. The array shows this is a well-established finding, not a single study's result.

Example 5: Minimal Valid Reference

When you have limited metadata but a valid source:

{
  "@type": "CausalStep",
  "mechanism": "Vitamin D receptors in hair follicles regulate follicle cycling",
  "result": "Deficiency disrupts normal growth-rest phase transitions",
  "evidenceSource": {
    "@type": "ScholarlyArticle",
    "@id": "https://pubmed.ncbi.nlm.nih.gov/20178699/",
    "name": "Vitamin D and the hair follicle"
  }
}

Why this is acceptable: At minimum, @id (a resolvable URL) and name (article title) make the source verifiable. More metadata is better, but this is sufficient for basic attribution.

Related Properties

Complementary Properties

• evidenceLevel — Grades the quality/strength of evidence (use alongside evidenceSource to indicate both what the evidence is and how strong it is)
• limitations — Describes caveats about the evidence or claims

Schema.org Properties

• schema:citation — For article-level references (on CreativeWork types)
• schema:isBasedOn — For derivative works

Source Types

• schema:ScholarlyArticle — Peer-reviewed research
• schema:MedicalStudy — Clinical trials and studies
• schema:MedicalGuideline — Clinical practice guidelines