There is a long-running argument in open data about formats. One camp says publish a spreadsheet and get out of the way. Another says anything short of linked data is a missed opportunity. Both have a point, and neither answer is right for every dataset. The useful question is narrower: what will people do with this data, and what does it need to connect to?
This piece sets out the three common options and when each fits. The principles behind linked data are explained in the guide to linked open government data.
What the rules actually ask for
EU law sets a floor, not a format. The Open Data Directive, Directive (EU) 2019/1024, asks public sector bodies to make documents available, where possible and appropriate, in formats that are open, machine-readable, accessible, findable and reusable, together with their metadata.2 For the high-value datasets listed in Implementing Regulation (EU) 2023/138, the data must be free of charge, with limited exceptions, machine-readable, provided through APIs and, where relevant, offered as a bulk download.3
Neither text requires RDF, the Resource Description Framework behind linked data. A well-documented bulk file and an API can meet the format requirements for high-value data, alongside the rules on licences, metadata and documentation; linked data is a choice, not an obligation.3
A reminder of the five stars
Tim Berners-Lee’s five-star scheme, added to his Linked Data note in 2010 to encourage government data owners in particular, is a useful way to see the trade-offs.1 One star is any format under an open licence; two is structured, machine-readable data; three is a non-proprietary format such as CSV; four uses W3C standards such as RDF to identify things; five links to other people’s data for context.1
Three-star data is already open data in the full sense. The fourth and fifth stars add value when identity and connections matter, and they add cost too.
Three options, side by side
| Option | Best when | Watch out for |
|---|---|---|
| CSV file (bulk download) | The data is a table, changes on a schedule and users want the whole thing for analysis | Undocumented columns, local codes, no stable identifiers |
| API | Data changes often, users need one record at a time, or the dataset is large | APIs that change without warning, missing documentation, no bulk alternative |
| Linked data (RDF) | Things need stable web identifiers, records must link across publishers, or the data is reference data many systems point to | Effort to model and publish; developers less familiar with the tools |
These options aren’t exclusive. Many publishers offer a CSV download and an API from the same source, and use linked data only for the identifiers and code lists the rest depends on.
Four questions that settle most cases
- Who is the main user? Analysts tend to want the whole file. Developers building services tend to want an API. Other publishers linking to your records benefit most from linked data.
- How often does it change? Monthly statistics suit a dated file. Live availability or locations suit an API.
- Do things need identifiers others will reuse? If other datasets will point at your schools, stops or bodies, give each one a persistent web identifier, whatever format carries the rest.
- Is it reference data? Code lists and authoritative lists are where linked data has lasted best, because many systems depend on the same terms. See ontologies and core vocabularies.
Example
An education authority publishes its register of schools. Analysts want the whole list each term to compare pupil numbers, so a dated CSV download suits them. A school-finder app needs today’s opening status and contact details one school at a time, so an API suits that. Planning, transport and inspection datasets all refer to the same schools, so each school also gets a persistent web identifier, published as linked data, that the CSV file and the API both carry. One register, three routes, and the identifier ties them together.
Match the effort to the data
W3C’s Data on the Web Best Practices takes a sensible middle line: it warns that complex vocabularies take more effort to produce and understand, which can hamper reuse, and notes that there is nothing complicated about a list of bus stops.4 It also warns against going so simple that useful facts, such as where each stop actually is, get left out.4
Whatever you choose, describe it well. A dataset nobody can find or understand fails whichever format it is in, which is why catalogue metadata matters; the guide to DCAT-AP and data catalogues covers that.
Sources
- Linked Data - Design Issues, W3C (Tim Berners-Lee). Accessed 17 October 2026.
- Directive (EU) 2019/1024 of the European Parliament and of the Council of 20 June 2019 on open data and the re-use of public sector information (recast), EUR-Lex, Official Journal of the European Union. Accessed 17 October 2026.
- Commission Implementing Regulation (EU) 2023/138 of 21 December 2022 laying down a list of specific high-value datasets and the arrangements for their publication and re-use, EUR-Lex, Official Journal of the European Union. Accessed 17 October 2026.
- Data on the Web Best Practices, W3C. Accessed 17 October 2026.