Dataset Schema for Research Pages

#01What counts as a dataset?

Google's examples (Google):

• A table or CSV file with data
• An organised collection of tables
• A file in a proprietary format that contains data
• A collection of files that together form a meaningful dataset
• A structured object in another format you'd load into a special tool
• Images capturing data
• Machine-learning files, such as trained parameters or network structure definitions

Our view: a research report article isn't a dataset, but the survey results you publish alongside it can be. If you run original research, publishing the underlying data with Dataset markup makes your numbers traceable, which supports being cited.

#02Which properties matter?

Source: Google.

#03How do you add it step by step?

1. Create a landing page for the dataset that describes it in plain language and links to the files.
2. Write the description (50–5,000 characters): what was measured, how, when and by whom.
3. Add creator and licence. Use a specific licence URL, such as https://creativecommons.org/licenses/by/4.0, as in Google's recommended example.
4. Add distribution for each file format you offer.
5. Add coverage only where the data has a time or place dimension.
6. Test in the Rich Results Test, then check Dataset Search after recrawl.

Expected result: the Rich Results Test detects a Dataset with no errors; the dataset eventually appears in Dataset Search.

#04What does the JSON-LD look like?

Fictional example with hypothetical data:

<script type="application/ld+json">
{
  "@context": "https://schema.org/",
  "@type": "Dataset",
  "name": "AI assistant citations of B2B SaaS brands, Q3 2026 sample",
  "description": "Prompt-level sample of answers from AI assistants for B2B software queries, with the brands and URLs cited in each answer. Collected weekly between July and September 2026. Fictional example only.",
  "creator": {"@type": "Organization", "name": "Example Research", "url": "https://www.example.com/"},
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "temporalCoverage": "2026-07-01/2026-09-30",
  "variableMeasured": ["assistant", "prompt", "cited brand", "cited URL"],
  "distribution": [{
    "@type": "DataDownload",
    "encodingFormat": "text/csv",
    "contentUrl": "https://www.example.com/data/citations-q3-2026.csv"
  }]
}
</script>

This example is hypothetical and doesn't describe real MentionX data.

#05How do you remove a dataset from Dataset Search?

Google says to use the robots meta tag to control indexing, and that changes can take days or weeks to show depending on the crawl schedule (Google).

#06What goes wrong? (Troubleshooting)

#07How does this relate to AI citations?

Our view: original data is one of the strongest reasons for other sites and AI tools to cite you. A dataset page with a clear description, a named creator, a licence and dated coverage gives anyone quoting your figures a stable source to point at. There's no published evidence that Dataset markup itself increases AI citations; the value is in making the data findable and attributable.

#08How should you document methodology?

Our view: put a short methodology section on the landing page covering the sample, collection dates, tools, known limitations and any changes between versions. Version datasets explicitly (for example "v2, October 2026") and keep old versions downloadable so earlier citations still resolve. That honesty about limits is what makes data credible enough to be quoted.

#09Exercise

#10Should you use DataCatalog too?

Google's Dataset documentation also covers DataCatalog and DataDownload types (Google). Our view: if you publish several related datasets, such as quarterly releases of the same study, a catalog page that lists them helps both people and crawlers see the series. For a single dataset, a well-described Dataset page is enough. Either way, link each dataset page from the research article that uses it, and link back from the dataset to the article so readers can find the analysis.

#11Checklist

• Page hosts or links real data files
• name unique and descriptive
• description 50–5,000 characters
• creator, licence, distribution
• Coverage only where relevant
• Tested, and checked in Dataset Search

#12From MentionX data: which pages engines citeOriginal data

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Research pages are usually published as articles. In the MentionX sample, articles are the largest cited URL type:

Engines differ in how often they cite sources:

Our view: original data is one of the few things engines can't get elsewhere. A research page with a clear method, a downloadable dataset and accurate markup makes that data easier to find and cite.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

#13Worked example: Ledgerline publishes invoice-processing research

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline surveys finance teams and publishes a research page with illustrative findings, a downloadable CSV and a methodology section explaining the sample, dates and questions.

The developer adds Dataset markup with the name, a description, the creator, a licence, the temporal coverage and a DataDownload pointing to the CSV with its format. The research page also has Article markup for the write-up.

The team tests the markup in the Rich Results Test, checks the dataset appears in Google Dataset Search after a few weeks and adds a note to update the markup when next year's data is published.

#14How do you choose a licence?

State the licence clearly so others know how they can reuse the data. Common choices include Creative Commons licences. Our view: a licence that allows reuse with attribution makes it more likely that others will cite and link to your data.

#15How do you version a dataset?

Give each release a version or date, and keep older versions available with clear labels. Update the dateModified and version properties when the data changes. Explain what changed in a short note on the page.

#16How do you present findings for readers?

Start with the main findings in plain language, then give the detail. Use tables and charts with clear labels and text descriptions. Link each finding to the data behind it. Label any estimates or illustrative figures clearly.

#17How do you promote research data?

Share the research with journalists, analysts and communities interested in the topic. Offer the data for others to analyse. Track where the research is cited and in which AI answers.

#18How do you describe the sample?

State who was included, how they were chosen, how many took part and when. Say what the sample doesn't represent. Our view: clear limits make data more credible, because readers can judge where it applies.

#19How do you format downloadable data?

Offer data in open formats such as CSV, with a header row and clear column names. Include a short data dictionary explaining each column and its units. Avoid formats that need special software to open.

#20How do you mark up several files?

Use a DataDownload for each file, with its encodingFormat and contentUrl. If you publish several related datasets, describe each one separately and consider a DataCatalog to group them.

#21How do you handle sensitive data?

Remove personal data and anything that could identify individuals or customers. Aggregate figures where needed. Check with your legal team before publishing.

#22How do you keep the page useful after the first release?

Update the page when new data is published, keep older versions available and add a note explaining changes. Correct errors openly with a dated note.

#23How do you write a dataset description?

Say what the data covers, who collected it, when and how, in two or three sentences. Mention the number of records and the main variables. Avoid marketing language. The description should help someone decide whether the data is useful before they download it.

#24How do you get the dataset cited?

Give a suggested citation with the title, creator, year and URL. Make it easy to copy. Our view: a clear suggested citation makes it more likely that journalists and researchers credit you correctly, which in turn gives engines more sources pointing to your data.

#25How do you check Dataset Search?

Search Google Dataset Search for your dataset's name a few weeks after publishing, and check the details shown match the page. Fix any errors in the markup and re-test.

#26How do you link findings to the data?

For each finding, name the column or table it comes from and the calculation used. Where useful, show the code or formula. This lets readers check your work and makes the findings easier to cite accurately.

#27Who should own the dataset?

Name a person or team responsible for the data, the methodology and updates. Give a contact for questions. Our view: a named owner makes the data more trustworthy and easier to keep current. List the owner's name and a contact address on the page itself, so anyone with a question about the data knows who to ask.

#28Common questions

Does Dataset markup help AI answers? There's no published evidence of a direct effect. It helps the data appear in Dataset Search.

Do we need to publish raw data? Our view: where you can. Raw data makes findings checkable and more citable.

Can a dataset be behind a form? The page can be indexed, but gated files are harder to reuse and cite.

#29Related

• DefinedTerm schema for glossaries
• Schema markup for AI search