# How to Increase the Citation Rate of Your Shared Dataset

A dataset saved to a lab hard drive earns nothing. A dataset with a persistent identifier, a clear license, a data dictionary, and a machine-readable citation file can earn citations for years after the associated paper stops collecting them. This guide walks through a complete workflow for research data sharing that raises the odds your dataset gets found, understood, trusted, and cited.

By the end, you will have completed eight concrete steps: choosing a repository, minting a persistent identifier, writing documentation, adding a citation file, linking the dataset to your paper in both directions, protecting privacy without destroying reuse value, versioning and announcing the release, and tracking citations to the dataset over time.

Have the following on hand before you start:

- The dataset files, plus any code needed to reproduce derived values
- A draft data dictionary (a table describing every variable, its units, and how it was measured)
- An ORCID iD, which is a persistent identifier for researchers, free to register at orcid.org
- An account at the repository you plan to use
- Either a spreadsheet or Python 3.7 or later for the citation tracking step

## What the Evidence Says About Which Datasets Get Cited

Before the mechanics, it helps to know where citation rate actually comes from. Three findings from the bibliometric literature are useful.

First, structured access drives reuse even when data are not fully open. In a cross sectional study of the Yale University Open Data Access (YODA) Project platform, which manages external sharing of individual participant level data from Johnson & Johnson sponsored clinical trials, 265 of 336 eligible trials (79%) had at least one secondary publication, producing 1,167 publications in total. Of these, 209 (17.9%) were external, meaning no original trial investigator was an author. Among external publications with a reported data access mechanism, 161 of 190 (85%) obtained the data through a data sharing platform, primarily the YODA Project itself [1]. The lesson for anyone planning research data sharing is that a clearly signposted, well governed access route is itself a citation engine. People reuse data they can find and request without friction.

Second, visibility carries a measurable citation advantage. In an analysis of the Journal of Craniofacial Surgery using Journal Citation Reports 2026 data, open access articles accounted for a disproportionately higher percentage of citations relative to their publication rate [2]. That finding comes from journal level data rather than dataset level data, so treat it as directional rather than exact, but the pattern is consistent with the broader finding that discoverable, unrestricted resources get used more.

Third, citations concentrate heavily on a small number of outputs. In a bibliometric analysis of 270 clinical studies of ginseng published between 2004 and 2026, citations were strongly concentrated (Gini coefficient 0.562, where 0 means perfectly equal and 1 means a single study holds everything), and among the factors tested, only team size predicted citation counts [3]. Most datasets will receive few citations no matter what you do. The practical goal is to move your dataset into the pool that serious reusers, who tend to work in teams and produce multiple papers, can actually find and build on.

With that framing, here is the workflow.

## How to Increase Data Citation Rate Step by Step

### Step 1: Choose the right repository

![Decision path for choosing a repository type, from field specific to generalist to institutional, with controlled access for human data.](/article-images/illustrations/increase-data-citation-rate-repository-choice.webp)

A data repository comparison starts with one question: where do reusers in your field already look? A dataset deposited somewhere your community never searches will not be cited regardless of its quality.

Four repository types cover almost every case:

| Repository type | Typical examples | Best suited for | Discovery | Access model |
|--|--|--|--|--|
| Generalist | Zenodo, Dryad, figshare | Any discipline, mixed file types, no field repository exists | Broad indexing, good general search visibility | Open, or embargoed for a defined period |
| Domain specific | GenBank, the Sequence Read Archive, the Cancer Imaging Archive (TCIA) | Field standard formats and community norms | Strong within the field | Often open, sometimes controlled |
| Institutional | University research data repositories | Funder or institutional mandate compliance | Moderate, mostly local | Varies by institution |
| Controlled access platform | The YODA Project platform | Human data requiring review of each request | Moderate | Data use agreement per project |

Two practical points from the evidence. The YODA Project shows that controlled access does not prevent external reuse, since most external secondary publications went through the platform itself [1]. So if your data involve identifiable humans and your ethics approval rules out open deposit, a governed platform is a legitimate, citation producing choice rather than a consolation prize. And for imaging data, domain specific repositories matter because they carry field standard curation. NIH [Data Management](/blog/guides/data-management-basics-principles-processes-and-best-practices) and Sharing policy updates and controlled access security requirements have increased attention to how head imaging data are handled at repositories such as the Cancer Imaging Archive [4], which means depositing there places your data inside an active policy and curation ecosystem rather than in an isolated file drop.

Rank your options this way: field specific repository first, reputable generalist second, institutional repository third, and a personal website or cloud drive never as the sole location.

### Step 2: Mint a persistent identifier and verify it

A digital object identifier (DOI) is a persistent, resolvable link to your dataset. Citations attach to DOIs. A raw URL attached to a lab server will break at the first server migration, and citation indexers cannot count what they cannot resolve.

Most repositories mint a DOI automatically at deposit. After the deposit is approved, verify that the DOI resolves:

```bash
curl -sL -o /dev/null -w "%{http_code}\n" "https://doi.org/10.5281/zenodo.1234567"
```

Replace the example DOI with your own. Expected output:

```
200
```

A final status of 200 means the DOI resolves to a live page. A 404 means the DOI exists but the target page is gone, which you should fix before doing anything else, because every future citation depends on that link.

Next, confirm that citation indexers can see the metadata, since that is what allows your dataset to be counted when other works cite it:

```bash
curl -s "https://api.crossref.org/works/10.5281/zenodo.1234567" | python3 -m json.tool | head -n 12
```

Expected output looks like this:

```json
{
    "status": "ok",
    "message": {
        "DOI": "10.5281/zenodo.1234567",
        "title": [
            "County-level air quality and asthma visits, 2010-2019"
        ],
        "publisher": "Zenodo",
        ...
    }
}
```

If the query returns a `Resource not found` message, the repository has not yet registered the DOI with Crossref. Registration is usually automatic within a day or two of publication. If it stays missing for more than a week, contact the repository's support desk.

### Step 3: Write documentation that answers the three reuse questions

Reusers ask three questions before they cite anything: what exactly is in these files, can I trust the measurements, and what am I allowed to do with it. Your README and data dictionary answer all three.

A workable README skeleton:

```markdown
# Dataset title
One paragraph: what was measured, where, when, by whom, and why.

## Files
- visits.csv (12.4 MB): one row per county and year
- codebook.txt: variable definitions, units, and missing value codes

## Methods
How data were collected and processed, with links to the protocol
and to the analysis code.

## Reuse
License name, how to acknowledge the dataset, and a contact address.
```

The data dictionary deserves the most care. For every column, record the variable name, a plain language definition, the unit (milligrams per liter, counts, years), the allowed range, and the code used for missing values. A dataset whose missing values are coded as -999 but never documented generates confused emails instead of citations.

Choose a license explicitly. For datasets, Creative Commons CC BY 4.0 is a common default because it permits reuse with attribution, and attribution is exactly the behavior you want to encourage. Leaving the license field blank is the single most common reason a reuser moves on to a better documented alternative.

### Step 4: Add a machine readable citation file

A CITATION.cff file is a small plain text file, in YAML format, that tells software and repositories exactly how your dataset should be cited. GitHub displays it with a "Cite this repository" button, and other tools parse it automatically.

```yaml
cff-version: 1.2.0
message: "If you use these data, please cite them using the metadata below."
title: "County-level air quality and asthma visits, 2010-2019"
authors:
  - family-names: Rivera
    given-names: Maria
    orcid: "https://orcid.org/0000-0002-5522-0000"
version: 1.0.0
date-released: "2026-11-11"
doi: 10.5281/zenodo.1234567
license: CC-BY-4.0
```

Place the file in the top level directory of the deposit or repository. Common errors: a missing `cff-version` line causes parsers to reject the file, an incorrect ORCID format (the iD must be a full URL) triggers validation warnings, and a DOI that does not match the deposit's actual DOI will propagate wrong citations. Validate the file with any of the public CFF validators before release, and confirm the rendered citation matches the way you cite the dataset in your own paper.

### Step 5: Link the dataset and the paper in both directions

![Three step flow showing dataset citation in the reference list, a data availability statement, and a related publication backlink.](/article-images/illustrations/increase-data-citation-rate-link-both-ways.webp)

Citation indexers connect records through explicit links, not through goodwill. Do all three of the following.

First, cite the dataset itself in the reference list of the associated paper, in the standard dataset format: authors, year, dataset title, repository name, and the DOI with a version number. This is the mechanism that makes data citations countable in the first place, because each citing paper now carries a machine readable pointer to your dataset's DOI.

Second, include a data availability statement that names the repository, the accession or DOI, and any access conditions. Statements that only say "data are available from the corresponding author on reasonable request" are uncountable, unverifiable, and, in practice, rarely fulfilled.

Third, when you deposit, fill in the repository's related publication field with the paper's DOI. This creates a backlink from the dataset record to the paper, so a reader who finds the dataset through search also finds the paper, and vice versa.

The YODA study illustrates why this plumbing matters. Secondary publications in that portfolio clustered around trials whose data were discoverable through the platform, and the platform, not word of mouth, was the dominant access mechanism [1].

### Step 6: Protect privacy without destroying reuse value

Human data raise the hardest trade off in research data sharing. Two studies quantify the risk of overcorrection.

A case study of k anonymity, a privacy method that ensures every combination of quasi identifying variables (such as age, sex, and ZIP code) appears in at least k records, applied the method at k values of 5, 10, and 15 to 719,387 inpatient encounters. The combination of record suppression and microaggregation introduced distortions in diagnosis code distributions and hospital length of stay patterns that the anonymization tool itself did not report. In other words, the tool's internal information loss metrics said little about how badly the anonymized data would reproduce real clinical patterns [5].

In medical imaging, defacing, which is algorithmic removal or replacement of facial anatomy in head scans, distorted craniofacial geometry and altered image statistics in ways that affected the performance of deep learning models across three clinically relevant tasks, evaluated on MRI and CT data from 600 subjects [4].

Neither finding argues against privacy protection. Both argue for documenting it. Three design choices raise citation rates while remaining ethical:

- Use tiered or controlled access instead of destroying information. The YODA model of review board approved requests produced a steady stream of external secondary publications [1], so a data use agreement is not a barrier to citations if the request process is transparent.
- Document every transformation. State the anonymization method, the parameters (for example, which k value), and what a reuser should check before trusting specific fields [5].
- Consider alternatives that move the analysis to the data. Federated frameworks let collaborators train models across institutions without moving patient data at all. One such framework, built on secure multi party computation for clinical single cell data, reached accuracy within 4% of a centralized baseline for cell type classification [6]. If your institution cannot release the raw data, a federated release plus openly shared metadata and code still gives the community something to cite and build on.

Trust is the substrate underneath all of this. Work on AI mediated health engagement shows that privacy, reputational, and psychosocial risks erode the willingness of individuals and communities to share data with researchers at all, and that trust, consent behavior, and participation should be evaluated as explicit outcomes rather than assumed [7]. For datasets involving communities that have historical reasons for distrust, invest in the consent documentation and governance description, because reusers read those signals too.

A related case is data that belong partly to a commercial partner. Analysis of supermarket transaction and loyalty card data for nutrition research concluded that effective data sharing depends on secure data environments, transparent governance processes, strong academic industry partnerships, and committed champions on both sides, because such collaboration is not yet routine [8]. If your dataset involves partner held data, a written governance summary in the deposit record prevents the most common failure mode, which is a reuser abandoning the project because nobody can tell them who can approve access.

### Step 7: Version, announce, and sustain the release

Datasets behave like software. Assign a version specific DOI to each release and a concept DOI that always points to the latest version, which most generalist repositories support natively. When you correct an error or add years of data, publish version 2.0 with a changelog entry rather than silently overwriting files, because silent overwrites invalidate every existing citation.

Then announce the release where your reusers actually are: relevant society mailing lists, disciplinary forums, conference talks, and the methods section of any follow up paper. Outreach does not need to be elaborate. A short post that states what question the data can answer, the sample size, and the license will outperform a generic "we shared data" announcement.

Sustained maintenance matters more than a strong launch. Answer reuse questions publicly when possible, so the answer becomes part of the documentation, and deposit an annual minor update if your data stream continues. Datasets that visibly keep growing keep appearing in searches.

### Step 8: Measure and track your citation rate

Define the metric before you track it. A simple and defensible definition:

```
annualized data citation rate = citations to the dataset DOI / years since release
```

Worked example. Suppose your dataset DOI has accumulated 14 citing works over 3 years. The annualized rate is 14 divided by 3, which is about 4.7 citations per year. Track this number annually rather than obsessing over monthly counts, because data citations arrive slowly and irregularly.

For a quick check, Crossref exposes a citation count for every registered DOI:

```python
import json, urllib.request

doi = "10.5281/zenodo.1234567"  # replace with your dataset DOI
url = f"https://api.crossref.org/works/{doi}"
record = json.load(urllib.request.urlopen(url))["message"]
print("Title:", record["title"])
print("Citations indexed by Crossref:", record.get("is-referenced-by-count"))
```

Expected output:

```
Title: County-level air quality and asthma visits, 2010-2019
Citations indexed by Crossref: 14
```

Crossref counts only works that cite the DOI directly. To capture papers that used your data but cited only the associated paper, supplement with searches in Web of Science, Scopus, and Google Scholar. Meta research projects use exactly this forward citation approach, searching multiple databases and then screening each citing record to confirm genuine reuse rather than incidental mention [9]. When you report the dataset's impact, standard bibliometric quantities translate directly: total citations (TC), citations per publication (CPP), and a relative citation index (RCI) comparing your dataset's CPP against the field average [10].

A common error in this step is double counting, where the same citing paper appears in both the direct DOI count and the paper-level search. Deduplicate by DOI before reporting. Another is expecting Crossref and Scopus to agree, since their coverage differs. Report the source of every number you cite in a promotion dossier or grant renewal.

## Quick Tips That Raise Citation Rates

These short actions compound with the steps above:

- Put the dataset DOI in your email signature, conference slides, and personal webpage.
- Register the dataset with any discipline specific data aggregator your community maintains.
- Include one worked analysis script in the deposit. Reusers cite data they can run.
- When someone reuses your data, offer the exact citation string for their methods section. You remove their last excuse for citing informally.
- If reviewers at your target journal expect a data availability statement, draft it at manuscript submission time, not at acceptance.

## Common Mistakes and Limitations

These are the errors that most reliably suppress data citation rates.

1. **"Available on request" with no deposit.** This formulation cannot be verified, cannot be indexed, and rarely survives a change in corresponding author. Deposit the data or a documented reason why you cannot.
2. **No data dictionary.** Reusers spend hours reverse engineering column meanings and often give up. The time you spend on the codebook is the highest return documentation investment you can make.
3. **Over anonymization without disclosure.** Anonymization changes distributions in ways the tool does not flag [5], and defacing changes image statistics that downstream models depend on [4]. If you transform data for privacy, record exactly what you did. Undocumented transformations are worse than documented restrictions, because reusers cannot calibrate their analyses.
4. **Choosing a repository for convenience rather than discovery.** A repository your field never searches is a storage locker, not a publication. Base the choice on where comparable datasets in your discipline are deposited.
5. **Releasing summary figures instead of underlying numbers.** Charts cannot be reanalyzed, so they generate no secondary analyses and no citations.
6. **Skipping the license.** An unlicensed dataset is legally ambiguous, and many potential reusers, especially those at companies or in regulated fields, will not touch it.
7. **Expecting immediate or proportional returns.** Citations to any scholarly output lag publication by years and concentrate on a minority of outputs [3]. A quiet first year does not mean the dataset failed.

Three limitations of the metric itself deserve honest acknowledgment. Citation counts measure reuse that left a paper trail, and they miss teaching use, policy use, and internal reuse. Field norms differ enormously, so compare your dataset only against others in your discipline. And if your dataset involves human participants, decisions about what can be shared and how belong with your institutional review board or data governance office, not with a guide like this one. Citation optimization never overrides those obligations.

## Frequently Asked Questions

### How long does it take for a shared dataset to start getting cited?

Usually one to three years, sometimes longer. Citations to a dataset arrive through papers that used the data, and those papers need time to be written, reviewed, and published. Plan for a lag and judge the dataset on its annualized rate after year two, not on its first-quarter count.

### Do I need a DOI, or is a URL enough?

A DOI is strongly recommended and, for citation tracking purposes, effectively required. A URL can be cited informally, but citation indexers count references to [persistent identifiers](/knowledge/bioinformatics/persistent-identifiers-in-bioinformatics-selection-criteria-and-implementation-workflow), and URLs attached to personal servers break. Any reputable repository mints a DOI at deposit at no cost to you.

### Which repository should I choose for my data?

Choose the domain specific repository your field already uses if one exists, then a reputable generalist repository such as Zenodo or Dryad, then your institutional repository. Check where the datasets you yourself cite are deposited, because that is where your future reusers are looking.

### Will sharing my data let other researchers scoop my papers?

Controlled access platforms exist partly to manage this concern, and the evidence from a large trial data sharing portfolio shows that external secondary publications coexisted with the original investigators' own publication program rather than replacing it [1]. An embargo period, in which the data are deposited but released after your planned papers are submitted, is a standard compromise.

### How do I cite a dataset correctly in my own paper?

Use the format authors, year, dataset title, version, repository name, and DOI. Cite it in the reference list, not only in the text, so the citation is machine readable and countable. The CITATION.cff file in the deposit gives you a ready made string.

### My data cannot be shared openly for privacy reasons. Can I still raise its citation rate?

Yes. Deposit rich, open metadata with a documented access procedure, use a controlled access platform with a transparent request process, and consider federated analysis, which lets others work with the data without the data ever moving [6]. Openly documented, governed access datasets do attract citations.

### How do I find out who is citing my dataset?

Check the citation count on the DOI's Crossref record, then search the dataset DOI and the dataset title in Web of Science, Scopus, and Google Scholar, and screen the results for genuine reuse rather than passing mentions [9]. Deduplicate by DOI across sources before reporting totals.

### Do data citations count in hiring, tenure, and grant evaluation?

Increasingly yes, but practices vary widely by institution and funder. The safest approach is to report your dataset's impact in standard bibliometric terms, including total citations and citations per year [10], alongside the paper record, so evaluators can weigh it with familiar units.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "How long does it take for a shared dataset to start getting cited?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Usually one to three years, sometimes longer. Citations arrive through papers that used the data, and those papers need time to be written, reviewed, and published. Judge the dataset on its annualized rate after year two rather than its first-quarter count."
      }
    },
    {
      "@type": "Question",
      "name": "Do I need a DOI, or is a URL enough?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "A DOI is strongly recommended and effectively required for citation tracking. Citation indexers count references to persistent identifiers, and URLs on personal servers break over time. Any reputable repository mints a DOI at no cost."
      }
    },
    {
      "@type": "Question",
      "name": "Which repository should I choose for my data?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Choose the domain-specific repository your field already uses if one exists, then a reputable generalist repository such as Zenodo or Dryad, then your institutional repository. Check where datasets you cite are deposited, because that is where your reusers look."
      }
    },
    {
      "@type": "Question",
      "name": "Will sharing my data let other researchers scoop my papers?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Controlled access platforms exist partly to manage this concern, and evidence from a large trial data sharing portfolio shows external secondary publications coexisted with the original investigators' publication program. An embargo period is a standard compromise."
      }
    },
    {
      "@type": "Question",
      "name": "How do I cite a dataset correctly in my own paper?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Use the format authors, year, dataset title, version, repository name, and DOI. Cite it in the reference list, not only in the text, so the citation is machine readable and countable. The CITATION.cff file in the deposit provides a ready-made string."
      }
    },
    {
      "@type": "Question",
      "name": "My data cannot be shared openly for privacy reasons. Can I still raise its citation rate?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Deposit rich open metadata with a documented access procedure, use a controlled access platform with a transparent request process, and consider federated analysis, which lets others work with the data without it ever moving. Governed access datasets do attract citations."
      }
    },
    {
      "@type": "Question",
      "name": "How do I find out who is citing my dataset?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Check the citation count on the DOI's Crossref record, then search the DOI and dataset title in Web of Science, Scopus, and Google Scholar, screening results for genuine reuse. Deduplicate by DOI across sources before reporting totals."
      }
    },
    {
      "@type": "Question",
      "name": "Do data citations count in hiring, tenure, and grant evaluation?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Increasingly yes, but practices vary by institution and funder. Report your dataset's impact in standard bibliometric terms, such as total citations and citations per year, alongside the paper record so evaluators can weigh it with familiar units."
      }
    }
  ]
}
</script>

## Sources

1. [Secondary use and scholarly impact of a portfolio of clinical trials with shared individual participant level data: cross sectional study.](https://pubmed.ncbi.nlm.nih.gov/42686366/)
2. [Journal of Craniofacial Surgery in the Journal Citation Reports 2026: Current Metrics and Emerging Trends.](https://pubmed.ncbi.nlm.nih.gov/42733957/)
3. [Global research trends in clinical studies of ginseng: a bibliometric and text-as-data analysis.](https://pubmed.ncbi.nlm.nih.gov/42948596/)
4. [Balancing privacy and performance: the impact of facial defacing on AI in medical imaging.](https://pubmed.ncbi.nlm.nih.gov/42667924/)
5. [Quantifying the Impact of Anonymization-Induced Clinical Data Quality Loss: Methodological Quantitative Case Study Using Primary Diagnosis Codes and Hospital Length of Stay.](https://pubmed.ncbi.nlm.nih.gov/42726783/)
6. [Clifti-GPT: privacy-preserving federated fine-tuning and transferable inference of foundation models on clinical single-cell data.](https://pubmed.ncbi.nlm.nih.gov/42557585/)
7. [Understanding Race as a Key Analytic Dimension in AI-Enabled, Health-Related Social Engagement Online and Its Role and Impact in Health Research.](https://pubmed.ncbi.nlm.nih.gov/43007696/)
8. [Conflict or interest? Using consumer purchase data to support food systems transformation.](https://pubmed.ncbi.nlm.nih.gov/42504032/)
9. [Assessing the impact, uptake and use of reporting guidelines for patient and public involvement in research: GRIPP2 - study protocol for a meta-research project.](https://pubmed.ncbi.nlm.nih.gov/42020124/)
10. [Bibliometric analysis of 50 years of publications in Skeletal Radiology (1976-2025): trends, impact, and global collaboration patterns.](https://pubmed.ncbi.nlm.nih.gov/42696170/)