How to Promote a Research Dataset So People Reuse It
A step-by-step guide to getting a dataset you built found, understood and cited. It covers identifiers, documentation, citation files, data papers and a checklist of the places where other researchers decide whether your data fits their work.
You spent a year building the dataset. You cleaned it, documented the edge cases and uploaded it with the paper. Six months later the repository page shows a handful of downloads, and the one group that should be using it has built their own, smaller version.
Datasets rarely fail because they are bad. They fail because the people who need them cannot find them, cannot tell quickly whether they fit, or cannot see how to give credit.
Make It Findable First
Start with the plumbing, because every later step points back to it. The FAIR Guiding Principles begin with findability: data are assigned a globally unique and persistent identifier, described with rich metadata, and registered or indexed in a searchable resource.
In practice that means a repository rather than a lab server or a cloud folder link. Zenodo, for example, says it does not impose requirements on format or access restrictions, that it can embargo content and open it automatically when the associated paper is published, and that it makes research results citable, passing citation information to DataCite. At the time of writing (October 2026), its policies set a total file size limit of 50GB per record, with higher quotas available on request, and require a licence for all publicly available files. Your field may have a discipline repository that its researchers already search; if it does, use it.
Write the Description a Stranger Needs
A file listing is not a description. In Datasheets for Datasets, Gebru and colleagues propose that every dataset come with a datasheet documenting its motivation, composition, collection process, recommended uses and more. They describe the aim as better communication between dataset creators and dataset consumers.
If you work in machine learning, the Hugging Face Hub's dataset card does similar work in a place people already browse. The card is the README rendered on the dataset's main page, and its YAML metadata can record license, language, size and tags that help users discover it. If the card links your arXiv paper, the Hub adds an arxiv tag that lets visitors reach the paper page and find models that cite the same paper.
Write the first paragraph for someone deciding in thirty seconds. What is in it, how much, collected how, and what question it lets someone answer.
Tell People How to Cite It
Credit is the part most dataset authors leave implicit, and it is the part that turns reuse into citations. If your data or code lives on GitHub, GitHub's documentation on CITATION files explains that adding a CITATION.cff file to the default branch adds a "Cite this repository" link to the repository page. A preferred-citation override can point that link at a data set or at your journal article instead of the software.
Put the same citation, word for word, in the repository record, the README and the paper's data availability statement. Consistency is what lets the credit add up in one place.
Consider a Data Paper
A substantial dataset can carry its own paper. Scientific Data's submission guidelines describe Data Descriptors as descriptions of datasets shared in repositories, reviewed alongside the data. They say a Data Descriptor should not contain results, discussion or analyses, because its fundamental goal is to describe the technical contents of a dataset to encourage others to use it, and they include an optional Usage Notes section for information that helps others reuse the data.
Even if you never submit one, those headings make a useful outline for your README.
A Checklist for Where Reuse Is Decided
Say you built a labelled dataset of field recordings for a bird-call classifier. A potential user meets it in one of five places. Each one should answer the same three questions quickly: what the data is, what it is good for and how to cite it.
| Where they meet it | What they need to see | What to put there |
|---|---|---|
| Repository record | Is this the right data, and can I use it? | Persistent identifier, clear title, license, a one-paragraph description, the citation |
| Dataset card or README | Does it fit my task? | Motivation, composition, collection process, recommended and discouraged uses |
| Code repository | Can I load it today? | Loading example, version, CITATION.cff with the preferred citation |
| The paper | Why should I trust it? | Methods, validation, data availability statement with the same citation |
| Your lab or project page | Who made this, and what else is there? | A plain explanation, links to all of the above, the paper |
The last row is the easiest to skip. A project page for your paper is where a curious visitor gets the story in one place, and the guide to promoting a machine learning paper beyond arXiv covers the channels that send people there.
Where Amplify Fits
Most datasets have a paper behind them, either the study that introduced the data or a data paper. That paper is what Amplify works from. Upload the PDF and Amplify makes a video abstract, a vertical short, an audio brief and a one-page infographic from it and nothing else, and every artifact carries the title, the authors and a link back to the paper of record.
The embeddable widget puts all four formats beside the paper on your project page or lab site with one script tag, so a visitor can read, listen and chat with the paper without leaving the page. The analytics show reads, listens, questions asked and which format moved people. The questions readers ask can also hint at what potential users want to know before they go looking for the data.
A dataset is reused when someone can find it, understand it in a minute and see how to credit you. None of that is extra work once it is written down in the right places.
Quick answers
How do I make my research dataset citable?
Deposit it in a repository that gives it a persistent identifier, then tell people how to cite it. Zenodo says it makes research results citable and passes citation information to DataCite. On GitHub, a CITATION.cff file in the default branch adds a "Cite this repository" link, and its preferred-citation field can point to a data set or to your paper.
What should a dataset description include?
Enough for a stranger to decide whether the data fits their work. Datasheets for Datasets proposes documenting motivation, composition, collection process and recommended uses. On the Hugging Face Hub, the dataset card is the README rendered on the dataset page, and its metadata can record license, language, size and tags that help people discover it.
Should I write a data paper for my dataset?
It is worth considering if the dataset is substantial. Scientific Data publishes Data Descriptors, which describe datasets shared in repositories and are reviewed alongside the data. Its guidelines say a Data Descriptor should not contain results, discussion or analyses, because its goal is to describe the dataset's technical contents to encourage others to use it.
Where do researchers decide whether to reuse a dataset?
Usually in a few places: the repository record, the dataset card or README, the code repository, the paper that introduced the data, and your own lab or project page. Each one should answer the same three questions quickly: what the data is, what it is good for, and how to cite it.
