What are the FAIR data principles?

FAIR is easier to satisfy than its reputation suggests, because most of it is handled by depositing in a decent repository. A good repository gives you a persistent identifier, a metadata schema, and a licence field, which covers a large part of the requirement before you have done anything unusual.

What remains is yours. Metadata quality is the first part: a description that says what the data is, how it was collected, what the variables mean, and under what conditions it may be used. Repositories provide the fields and cannot fill them in usefully for you.

Formats are the second. Data stored in a proprietary format that requires a licensed application is a barrier to reuse and often to your own future access, since software versions stop opening old files. Exporting to open formats such as CSV, plain text, or an open image standard costs little at the point of deposit and is the difference between preserved and merely stored.

Vocabularies are the part most people skip. Interoperable means using the standard terms and identifiers your field already has, so a machine can relate your dataset to others. If your discipline has an ontology or a controlled vocabulary, using it is what turns your dataset from a file into something combinable with others.

The reusable component is mostly licensing and provenance. Without a licence, a legally careful researcher cannot use your data at all, because silence is not permission.

The original formulation is more specific than the acronym suggests. Findable requires a globally unique and persistent identifier, rich metadata, an explicit link from the metadata back to the data it describes, and registration in a searchable resource. Accessible requires retrieval over a standardised protocol that is open, free, and broadly implementable, with authentication allowed where it is needed, and it requires the metadata to stay available even after the data themselves are removed. Reusable requires accurate description, an accessible usage licence, detailed provenance, and compliance with community standards (Wilkinson et al., 2016, p. 4).

What does making data FAIR involve in practice?

Four concrete actions. Deposit in a repository that issues a persistent identifier. Describe the data with metadata that includes what it is, who made it, and under what conditions. Use open, standard file formats. Attach a licence saying what others may do with it.

Do these at deposit rather than planning to improve the record later. Metadata written while you still remember the project is accurate, and metadata reconstructed two years afterwards is thin.

Ask your library which repository suits your data type. Subject repositories often provide better metadata schemas and more relevant discovery than a general institutional one. Harvard Dataverse is the example the 2016 paper works through, with more than 60,000 datasets, persistent identifiers, searchable landing pages, metadata that survives when files are restricted or withdrawn, and machine-readable interfaces (Wilkinson et al., 2016, p. 5).

If your field has no metadata standard and no subject repository, you are not doing it wrong. Belliard and colleagues report data and code sharing rates between 1% and 20% across disciplines, higher in areas such as genomics, and they attribute the gap to missing standards, missing repositories, and disagreements over which community holds the authority to pick a standard. Their answer is a ten-step sequence that starts with defining the community and building partnerships, then moves to setting minimum metadata requirements, documentation standards, and infrastructure, and ends with planning for long-term sustainability (Belliard et al., 2023, pp. 1-2). Decide your publication and access scenario before or during collection rather than at deposit, since whether the data will be public, closed, or controlled-access changes what consent and documentation you need (Hassenstein and Jung, 2025, p. 2).

Does FAIR mean open?

No, and this is the most common misunderstanding. Accessible means the access conditions are clearly stated and the metadata is findable, not that anyone can download the data. Sensitive data can be fully FAIR while remaining restricted, provided the record and the access process are public.

The formulation people use is that data should be as open as possible and as closed as necessary. A controlled-access dataset with a public record, a stated process for requesting access, and a named contact satisfies the principles.

The authors of the principles say this directly. FAIR is not a standard or a specification that mandates a particular technology, it is high-level guidance for judging whether an implementation makes a research object findable, accessible, interoperable, and reusable, and it can be adopted incrementally and in different combinations. Sensitive and personally identifiable data can meet it through rich public metadata plus a stated access procedure. The same expectations apply to workflows, software, and other research objects, which should also carry identifiers, descriptions, and reuse terms (Wilkinson et al., 2016, pp. 4-5).

This matters for anyone working with human participants. FAIR compliance and confidentiality are compatible, and a plan that treats them as opposed usually ends up sharing nothing at all, including the metadata that would have let people find the study.