What it is for
A PDF is a picture of a page. It is excellent at being read by a person and nearly useless to anything else: the words are there, but nothing in the file says which of them are the title, and an index that wants to know who wrote the paper has to guess from position and type size.
JATS exists so that nothing has to guess. Once an article is written down this way, a great many things become possible that were not:
- An index can file the article correctly - right authors, right affiliations, right dates - without a human retyping the metadata.
- A reader can be sent to the article in a form that reflows on a phone, because the text is text and the structure is known.
- A reference can be matched to the work it points at, so citations can be counted and followed.
- An archive can preserve the article in a form that will still be readable when today's software is gone, which is the thing a PDF is worst at.
- A search can find a phrase in the discussion and know it was in the discussion.
What is inside one
An article in this form has two halves. The front matter carries everything about the article: journal title and identifiers, the article's own identifiers including its DOI, the title, the authors and their affiliations and ORCIDs, the abstract, keywords, funding, the licence, and the dates it was received, revised and accepted.
The body carries the article: sections and subsections at their proper levels, paragraphs, figures and tables with captions and numbering, equations as mathematics, footnotes, and the cross-references that tie a mention of "Figure 2" to the actual figure. After the body comes the reference list, each entry broken into its parts rather than left as a line of text.
That last point is the one most often skipped, and it is the one that makes citation linking possible. A reference stored as a single string is a string; a reference stored as author, year, title, journal, volume, pages and DOI is data.
Valid is not the same as correct
A file in this format can satisfy every rule of the standard and still be wrong, and the two kinds of error are worth keeping apart in your head.
A structural error breaks the rules: a section with no title where one is required, a figure outside any section, a date that is not a date. These are caught mechanically, and anything producing JATS should catch them before you ever see the file.
An error of fact breaks nothing: an affiliation attached to the wrong author, a figure in the wrong section, a reference whose year does not match the paper it names, an abstract with a letter missing from the first word. Every one of those produces a perfectly valid file, and no validator will ever mention them. They are found by a person reading the article, which is why a step where a person reads the article is not optional.
Who asks for it
In practice: PubMed Central requires it for deposit and will not take a PDF alone. Several large archives take it in preference to anything else. Crossref takes its own metadata format for DOI registration but the content of that deposit comes from the same place. Many funders now require that the accepted article be deposited somewhere that wants structured full text.
A journal that does not produce it is not failing a rule, but it is invisible to a growing number of the places readers actually look.
Last reviewed
