Machine-readable metadata#

Declaring [[project.citations]] also makes the site’s citation machine-readable: every page’s <head> carries the self citation as Highwire and Dublin Core meta tags, along with a schema.org JSON-LD description of the site. Each citation’s type sets the schema.org type it’s described as, so type = "dataset" is what makes a data release indexable by Google Dataset Search. A citation’s version appears on the two types that define a property for it, softwareVersion on a SoftwareSourceCode and version on a Dataset, and on no other node.

Highwire meta tags are what Google Scholar’s inclusion guidelines specify and what Zotero’s embedded-metadata translator reads, so a reader on a page that carries them gets a one-click “Save to Zotero” with the right title, creators, date, and DOI. These are the tags emitted, in this order:

Tag

Value

citation_title

title

citation_author

one per author, family name first (an organization’s name is written whole)

citation_author_institution

that author’s affiliation, when set

citation_author_orcid

that author’s orcid as a resolvable https://orcid.org/ URL, when set

citation_publication_date

date as YYYY/MM/DD, or as YYYY when the entry dates the work to the year or the month

citation_doi

doi, bare

citation_publisher

publisher

citation_fulltext_html_url

For the self entry, the site’s root URL, which is the URL that DOI is registered against. For an entry that sets page, the claimed page’s own URL, with the claim’s fragment when it names one. Both need base_url, and the tag isn’t emitted without it.

DC.identifier

the DOI as a resolvable https://doi.org/ URL, the Dublin Core complement DataCite’s landing-page guidance asks for

A field that the entry doesn’t state emits no tag, so an entry located by url rather than by a DOI carries no citation_doi and no DC.identifier. Every value is HTML-escaped, so a title containing quotes or angle brackets reaches the tag intact.

Note

Documenteer writes the date tag as citation_publication_date, the spelling Google Scholar documents. A Rubin technote built with technote 0.11.0 or later writes the same tag. Earlier releases spelled it citation_date, and consumers read either.

Google Scholar’s inclusion guidelines describe only two forms for the value — a full date, or a year alone — so an entry dated to the month states only its year in this tag, rather than a 2025/06 form that no guideline covers and that Scholar may not parse. The month isn’t lost: the schema.org datePublished in the page’s JSON-LD block carries the entry’s date at the precision it states, 2025-06 included.

The JSON-LD block is about the self entry, and describes every other entry by its relation to it:

  • An entry that names a page is a part of the site’s own work (a data product of a release, for example), not something the site cites. The site-wide block names it under hasPart by reference alone: its schema.org type, its DOI, and its title, and nothing more. The full record lives on the page the entry claims, whose own block points back at the site with an isPartOf reference of the same shape.

  • An entry with no page is a work the site cites, and the site-wide block describes it in full. The preferred entry is always included, whether or not the footer shows it, because it’s the citation the site asks readers to use, which is what schema.org citation states. Any other entry is included only when the site displays it in the footer, that is, when in_footer is true.

  • An entry that’s not a part, not preferred, and not shown in the footer doesn’t appear in the site-wide block at all, because the site doesn’t display it by default. It still renders wherever a citation-card names it.

So in_footer controls a visible surface and, for the additional entries only, whether the metadata includes them. A site that sets in_footer = false on its preferred citation keeps the whole block.

A site that marks no entry self isn’t any DOI’s landing page, so it emits none of the meta tags above on any page. That includes the title and authors, since stating them would tell a harvester that this site is the full text of a work published somewhere else. Marking an entry preferred doesn’t change that: preferred says which citation the site asks readers to use, which affects the visible surfaces, and only self claims that this site is where a DOI resolves. The site’s JSON-LD block still has a subject: the site itself, a schema.org WebSite carrying the site’s title and URL and no identifier. The same entries appear in it under the same two relations, in the same form. Such a site can still be the landing page of a DOI on one of its own pages, and that page’s block names this WebSite node under isPartOf, so both ends of the relation are stated there too. The reference carries the site’s title and URL in place of a DOI, since those are what identify a site that has no DOI. Only the subject differs, so a consumer reads the block the same way whether or not the site publishes a DOI.

This site is an example. It declares two citations and marks neither self, so the source of this page has no citation_title or DC.identifier tag in its <head>. It does have one application/ld+json block, whose subject is a WebSite named Documenteer, with the Software entry (the preferred citation) described in full under citation as a SoftwareSourceCode. The Paper entry isn’t in the block: it’s not preferred and not shown in the footer, so it appears only on pages that render it in a card.

Describing parts by reference lets a site publish a DOI per data product without adding a full record to every page: a release with forty product DOIs states forty short references, not forty full records, and each product’s own landing page carries the record that describes it.

See [[project.citations]] for the full field reference.