Skip to content

Dataset schema for a research page with CSV downloads

Describe a published dataset and its downloadable files accurately. Test landing-page identity, DataDownload URLs, versions, and the bytes readers receive.

By Crawlito · 5 min read

In short

  • A dataset's landing page and its downloadable files have distinct URL roles.
  • DataDownload describes where a distribution can be retrieved and its format.
  • A research page should explain scope, units, provenance, and version so readers can use the data responsibly.
On this page

Use Dataset structured data on the page that explains a published dataset, and describe each available file with a DataDownload distribution. Keep the landing-page URL distinct from the CSV or spreadsheet address. Then download the file as a new visitor and confirm that its version, columns, and scope match the page. Google’s Dataset structured data documentation

Identify the dataset behind the chart

A research article can contain several charts derived from one dataset, multiple datasets, or a proprietary analysis with no public download. Start by deciding what you are actually publishing. Give the dataset a name that distinguishes its subject, coverage, and relevant period. A chart image alone does not supply the underlying rows or the methodology necessary to reproduce it.

Write down the observation unit. One row might represent a page checked, a business surveyed, a monthly aggregate, or an individual measurement. Define the columns, units, missing values, and inclusion criteria on the landing page or in an accessible data dictionary. Without those details, a correctly downloaded CSV can still be easy to misinterpret.

Keep the creator, collection method, and version visible. A publisher compiling licensed source material should describe the actual provenance and link to the applicable terms supplied by the rights owner. Use confirmed licensing information. A developer adding schema should not invent a reuse permission to fill an optional field. The original research guide discusses how to make evidence understandable.

Map the page and files to their proper fields

The Dataset URL describes the resource's public landing page. Google documents distribution for the ways someone can obtain the data, including a downloadable address and its encoding format. The following illustrative fragment shows that relationship for a fictional measurement series. It omits details that would need to come from the publisher's actual records.

json
{
  "@context": "https://schema.org",
  "@type": "Dataset",
  "name": "Example public-page response observations, October 2026",
  "description": "Published observations of public-page response status for the stated October sample.",
  "url": "https://example.com/research/october-responses",
  "distribution": {
    "@type": "DataDownload",
    "contentUrl": "https://example.com/data/october-responses-v1.csv",
    "encodingFormat": "text/csv"
  }
}

If the publisher offers CSV and spreadsheet versions, describe the separate available distributions and verify both. They should represent the documented dataset or clearly state any differences. A formatted workbook that excludes incomplete rows can have a different scope from the raw CSV. Explain that distinction rather than letting a shared filename imply equivalence.

Inspect the final emitted markup, especially when a general article plugin already creates a BlogPosting entity for the research page. A page can legitimately describe an article and a dataset, but each entity should identify the thing it represents. The structured data guide explains how to keep machine-readable descriptions tied to visible content.

Test the download without an editor session

Follow the public download link from a clean browser. Save the final URL, response type, and the file itself. A CMS attachment link can open a preview page or sign-in screen instead of returning the advertised file. An export button might require a request parameter available only after the editor has opened the dashboard.

bash
curl -sSL --max-time 30 -D download-headers.txt   -o dataset-download.csv https://example.com/data/october-responses-v1.csv

Read a few records with a CSV parser. Check the header row, delimiter, character encoding, and expected units. A semicolon-delimited file with a .csv suffix can be valid for its intended audience, but document the format so an automated consumer can interpret it. A response beginning with an HTML error message needs a delivery fix before any data analysis.

Compare the downloaded row count and key fields with a trusted publication manifest. Treat the manifest and file as parts of the same release. Counting against a live database that has changed since export can produce a false mismatch. Record which version was tested and when. The content type guide helps diagnose a file whose bytes and headers disagree.

Publish corrections without rewriting history silently

Suppose an editor discovers that a unit conversion was wrong for one column. Correct the data, explain what changed, and decide how the new version should be identified. A versioned download address allows people to reproduce an earlier analysis while directing new readers to the corrected release. A stable “latest” link can be useful too, provided the page makes its current version explicit.

Update the download link, distribution markup, and visible change note together. Check any chart generated from the old data. A corrected CSV alongside an unchanged chart produces contradictory evidence on the same page. The correction should cover the published outputs that depend on the changed values.

Preserve legitimate access controls and publication permissions. A report may summarize a dataset whose raw records cannot be shared publicly. In that case, describe what is available and how authorized access works without advertising an unrestricted download that returns a login page. The schema should follow the actual publication model.

Verify discovery and reader comprehension

Validate the public markup using the supported tools and inspect the rendered page. Confirm that the dataset's name, description, and distributions refer to the same release. Link the landing page from the research article, methodology page, or relevant resource hub where a reader would look for the source data.

Ask a colleague to answer three questions from the page: what does one row represent, which period does the file cover, and what would make a comparison with another dataset invalid? Missing answers indicate documentation work that a valid JSON block cannot resolve. Add examples using clearly labeled fictional rows if the real data needs more explanation.

Keep discovery observations separate from publication checks. A page and file can be accessible and accurately described before a search service surfaces them. Record the live test and any later discovery result with their dates. When a reader reports a broken citation, use the versioned landing and file addresses to determine whether the problem concerns the dataset identity, a replaced download, or a stale link.

A research dashboard may generate signed download URLs for its users. If an editor copies one of those addresses into a public dataset page, the link can work during review and fail later. Inspect the URL for an expiration mechanism and confirm the hosting service's intended public distribution method. Use a durable public release address when the dataset is meant to be openly downloadable.

Test the link from a new session after the relevant expiration period in a controlled environment. Confirm that the returned file is the same documented release and that the landing page still points to it. If controlled access is intentional, describe that process clearly and avoid representing a temporary authenticated link as an enduring public file.

Keep a file checksum or version manifest in the normal publication process when reproducibility matters. It lets a researcher distinguish a changed download from a broken link even if the filename remains the same. Compare that identifier with the correction note whenever a revised dataset replaces an earlier distribution.

Frequently asked questions

Can a CSV file be the only public page?

A descriptive landing page gives readers context, methodology, and a stable place to understand the release. Link the actual file as a distribution.

Does Dataset markup make unsupported findings credible?

The markup describes the resource. The methods, provenance, and published data provide the evidence a reader needs to evaluate a conclusion.

Should the description repeat the article headline?

Describe the dataset's actual contents and scope. An article about what the data suggests can have a different purpose and title.

Inspect the research landing page

Run the free page check on the page describing the dataset, then verify the downloadable file separately. Together those checks cover both the explanation and the resource readers use.

Filed under Structure and schema. Topics: structured-data, research, google