What Is Dataset Schema? The Complete Guide (2026)

Why Dataset Schema Matters

Google Dataset Search indexes datasets from repositories like Zenodo, Figshare, Harvard Dataverse, and government data portals. Datasets without structured markup are either not indexed or indexed with poor metadata — making them effectively invisible to data discovery systems.

AI research assistants (including those built on GPT-4, Claude, and Gemini) increasingly use Dataset Search as a retrieval source when answering questions about available research data. Properly marked-up datasets with complete metadata are far more likely to be cited in AI-generated research summaries.

Required Properties

Recommended Properties

ORCID and ROR: Persistent Researcher and Institution Identifiers

Dataset schema has specific conventions for identifying creators that differ from other content schemas. The scientific community uses:

• ORCID (Open Researcher and Contributor ID): A persistent digital identifier for individual researchers. Format: https://orcid.org/0000-0000-0000-0000.
• ROR (Research Organization Registry): A persistent identifier for research organizations. Format: https://ror.org/xxxxxxxxx.

These identifiers go in the sameAs property of the creator object. Greadme warns when a creator is missing sameAs, and the message specifies ORCID for persons and ROR for organizations.

Temporal Coverage Formats

The temporalCoverage property accepts four ISO 8601 formats. Greadme errors on anything that does not match one of these patterns:

Dataset Schema Code Example

A complete dataset schema for a climate research dataset:

{
  "@context": "https://schema.org",
  "@type": "Dataset",
  "name": "Arctic Sea Ice Extent 1979-2024",
  "description": "Monthly satellite measurements of Arctic sea ice extent from 1979 to 2024, collected by NSIDC using passive microwave remote sensing. Covers the Arctic Ocean and surrounding seas.",
  "url": "https://data.example.edu/arctic-ice",
  "identifier": "https://doi.org/10.1234/arctic-ice",
  "license": "https://creativecommons.org/licenses/by/4.0",
  "isAccessibleForFree": true,
  "keywords": ["Arctic", "sea ice", "climate", "remote sensing"],
  "temporalCoverage": "1979/2024",
  "creator": {
    "@type": "Organization",
    "name": "National Snow and Ice Data Center",
    "sameAs": "https://ror.org/02nv7yv05"
  },
  "distribution": {
    "@type": "DataDownload",
    "contentUrl": "https://data.example.edu/arctic-ice.csv",
    "encodingFormat": "CSV"
  }
}

Common Mistakes to Avoid

• Description under 50 characters: Too brief to be useful in dataset discovery. Greadme warns on short descriptions — expand with methodological or subject-matter context.
• Description over 5000 characters: Greadme warns — trim to a clear summary and use documentation pages for extended methodology.
• Generic dataset name: Names like "dataset" or "data" are flagged by Greadme. Use a specific, descriptive title.
• License as text instead of URL: Writing "license": "CC BY 4.0" instead of the actual URL causes a Greadme warning. Always use the full license URL.
• Missing distribution: Without a DataDownload, users and AI systems cannot access the data. Greadme warns and deducts 10 points.
• Invalid temporalCoverage format: Date ranges must use the ISO 8601 interval format (YYYY/YYYY or YYYY-MM-DD/YYYY-MM-DD). Using a slash with just years (e.g., "1990/2020") is valid, but"1990-2020" (hyphen instead of slash) is not.

How Greadme Validates Dataset Schema

Greadme runs a dedicated Dataset validator aligned with Google Dataset Search requirements. The score starts at 100 with the following key deductions:

Frequently Asked Questions