RIS, BibTeX, and CSL JSON all store the same thing: the bibliographic facts about a paper — who wrote it, what it is called, where it appeared. None of them decides how a citation looks on the page. That is the job of a citation style, and all three formats can feed any style.
So the choice between them is not about quality. It is about which tool has to read the file next. Pick the format your destination accepts natively, and you never touch it again; pick the wrong one and you spend the afternoon on a converter.
| Format | File | Shape | Best for | Reach for something else when |
|---|---|---|---|---|
| RIS | .ris | Tagged lines — a two-letter tag, then the value | Reference managers: Zotero, Mendeley, EndNote, RefWorks, Citavi, Papers | You are writing in LaTeX, where BibTeX is native |
| BibTeX | .bib | Brace-delimited entries — @article{key, field = {value}} | LaTeX and Overleaf, and any tool built around a .bib file | Your reference manager offers RIS, which carries publication types more precisely |
| CSL JSON | .json | Structured JSON — arrays and objects, not flat text | Pandoc, Quarto, R Markdown, and anything that processes references in code | You are importing into a reference manager, where support is patchier |
One thing none of them does: decide how your citations look. That is the citation style, and every one of these formats can feed every style.
Every sample below is the real export of one DOI — 10.1038/s41586-023-05881-4 (Moor M, Banerjee O, Abad ZSH, et al. Foundation models for generalist medical artificial intelligence. Nature. 2023;616(7956):259-265.) — so the differences you see are differences in the formats, not in the papers.
TY - JOUR AU - Moor, Michael AU - Banerjee, Oishi AU - Abad, Zahra Shakeri Hossein AU - Krumholz, Harlan M. AU - Leskovec, Jure AU - Topol, Eric J. AU - Rajpurkar, Pranav TI - Foundation models for generalist medical artificial intelligence JO - Nature PY - 2023 VL - 616 IS - 7956 SP - 259 EP - 265 DO - 10.1038/s41586-023-05881-4 UR - https://doi.org/10.1038/s41586-023-05881-4 ER -
RIS is a line-based format: each line opens with a two-letter tag, then the value. TY is the publication type and must come first; ER closes the record. Repeating a tag adds another value, which is how a paper with seven authors gets seven AU lines.
Its strength is reach. Zotero, Mendeley, EndNote, RefWorks, Citavi, and Papers all import RIS without a plugin, and most publisher sites offer an RIS download beside the PDF. If you do not know what the person receiving your file uses, RIS is the safest guess.
Its weakness is that the specification has been extended by different vendors over the years, so an unusual publication type may map differently between two managers. For journal articles, books, and chapters — the overwhelming majority of references — this never comes up.
@article{Moor2023s4158602,
author = {Moor, Michael and Banerjee, Oishi and Abad, Zahra Shakeri Hossein and Krumholz, Harlan M. and Leskovec, Jure and Topol, Eric J. and Rajpurkar, Pranav},
title = {Foundation models for generalist medical artificial intelligence},
journal = {Nature},
year = {2023},
volume = {616},
number = {7956},
pages = {259--265},
doi = {10.1038/s41586-023-05881-4},
url = {https://doi.org/10.1038/s41586-023-05881-4}
}BibTeX stores each reference as an entry with a type (@article, @book, @inproceedings) and a cite key — here Moor2023s4158602. That key is the point of the format: you write \cite{Moor2023s4158602} in your manuscript, and LaTeX pulls the reference in and formats it to whatever style the document declares.
If you write in LaTeX or Overleaf, this is not a preference. Your .bib file is part of the build, so BibTeX is what you want and converting away from it costs you the cite keys your manuscript refers to.
Outside LaTeX its appeal drops. BibTeX has a narrower vocabulary of publication types than RIS, and its brace syntax means special characters and non-English names sometimes need escaping. Reference managers read it, but usually as a second-class import.
[
{
"id": "10.1038/s41586-023-05881-4",
"type": "article-journal",
"title": "Foundation models for generalist medical artificial intelligence",
"author": [
{ "family": "Moor", "given": "Michael" },
{ "family": "Banerjee", "given": "Oishi" }
// … five more authors
],
"issued": { "date-parts": [[2023]] },
"container-title": "Nature",
"page": "259-265",
"volume": "616",
"issue": "7956",
"DOI": "10.1038/s41586-023-05881-4",
"URL": "https://doi.org/10.1038/s41586-023-05881-4"
}
]CSL JSON is the data half of the Citation Style Language. The other half is the style file, which holds the formatting rules — CSL deliberately separates the two, so one set of metadata can be rendered into any of 10,000+ styles without being rewritten.
Because it is real JSON, it is the format to choose when a program rather than a person handles the file next. Pandoc and Quarto read it directly, R and Python parse it with a standard library call, and structure that the other two formats flatten into text survives: an author is an object with family and given fields, not a string somebody has to split.
That structure is also why it is the most faithful of the three. A name like Zahra Shakeri Hossein Abad keeps its parts labelled instead of relying on a comma convention. The cost is import support — reference managers accept CSL JSON less consistently than RIS, so it is a poor pick for handing a colleague a bibliography.
| If you are… | Pick | Because |
|---|---|---|
| Importing into Zotero, Mendeley, EndNote, or RefWorks | RIS | Every major reference manager imports it natively, with no plugin or conversion step. |
| Writing a paper in LaTeX or Overleaf | BibTeX | Your .bib file is part of the build, and the cite keys are what \cite{} refers to. |
| Writing in Pandoc, Quarto, or R Markdown | CSL JSON | These tools read it directly and render it through any CSL style. |
| Processing references in a script or pipeline | CSL JSON | It parses with a standard JSON library, so there is no bespoke text parser to maintain. |
| Sending a bibliography to a colleague whose tools you do not know | RIS | It has the widest import support, so it is the least likely to need a conversion. |
| Submitting to a journal with a required format | Whatever they ask for | Author guidelines override every rule here. Check the submission page before you export. |
Converting between the three is usually lossless in the direction that matters, because all three encode the same underlying facts. Going from CSL JSON to RIS or BibTeX flattens some structure — a labelled author object becomes a text convention — but the fields themselves survive.
The safer habit is not to convert at all. Export from the identifier each time you need a different format, so every file is generated from the publisher's record rather than from another export. That also means a mistake never propagates: a wrong page range fixed at the source is fixed in all three.
Scholar Sidekick exports all three from an identifier: DOI to RIS and DOI to BibTeX have their own pages, and CSL JSON is one of nine formats on POST /api/export. Free, no signup, and the same record behind each.
They are three file formats for storing the same bibliographic facts. RIS uses tagged lines and is the format reference managers accept most widely. BibTeX uses brace-delimited entries with a cite key and is native to LaTeX. CSL JSON is structured JSON, so programs can parse it directly, and it is what Pandoc and Quarto read. None of the three controls how a citation is formatted - that is the job of a citation style.
RIS. Zotero imports it natively, as do Mendeley, EndNote, RefWorks, and Citavi. Zotero can also read BibTeX and CSL JSON, but RIS is the path with the fewest surprises, especially for unusual publication types.
BibTeX. Your .bib file is part of the LaTeX build, and each entry's cite key is what \cite{} in your manuscript refers to. Converting away from BibTeX costs you those keys, so there is no reason to use anything else in a LaTeX workflow.
CSL JSON. Both read it directly and render it through any CSL style. They also accept BibTeX, so an existing .bib file will work, but CSL JSON preserves more structure - author names stay split into family and given fields rather than relying on a comma convention.
No, and the distinction matters. CSL - the Citation Style Language - has two halves. CSL JSON is the data: the facts about a paper. A CSL style is an XML file holding the formatting rules for APA, Vancouver, and 10,000+ others. Separating them is the whole design, so one set of metadata renders into any style.
Yes, and it is usually lossless for the fields that matter, since all three encode the same underlying facts. Going from CSL JSON to RIS or BibTeX flattens some structure. The better habit is to export from the identifier each time rather than converting an export, so every file comes from the publisher's record.
No. The format decides how the reference is stored; the citation style decides how it is rendered. The same RIS, BibTeX, or CSL JSON record produces identical output in APA, and identical output in Vancouver. If your citations look wrong, the style is the thing to check, not the file format.
This page is a front end for the public Scholar Sidekick API — export in all three formats runs over plain HTTP, so you can call it from a script, an AI agent, or your reference manager. Light use needs no key at all. A free account issues an ssk_ key that raises your rate limit.