Which reference management software fits a thesis?

Direct answer: None of them reliably. RefWorks made the fewest errors per citation in a 2011 test, Zotero made the fewest mistakes in a 2016 test, and every product in every test made errors. Treat any ranking as a snapshot of one version tested on one set of references.

Gilmour and Cobus-Kuo (2011) imported 14 references from 7 bibliographic databases into each tool and built bibliographies in 5 citation styles. RefWorks scored 0.8 errors per citation, Mendeley 1.08, Zotero 1.3 and CiteULike 2.88. Kratochvíl (2017) ran a larger test: 17 references for each of 15 biomedical journals, plus the NLM style, produced by EndNote, Mendeley, RefWorks and Zotero and compared with hand-written references. Zotero made 575 mistakes, Mendeley 679, EndNote 882 and RefWorks 937. The author concluded that "none of the reference managers produces all bibliographical references without mistakes" (Kratochvíl, 2017).

Ivey and Crum (2018) reviewed four of the same products as librarians. Their verdict was that "All four products made occasional small errors in citations, especially when we cited web pages", and they added that "the best choice for a given purpose may be determined by factors other than the functionality of the applications themselves" (Ivey & Crum, 2018).

Source: the three studies cited above, read in abstract or full text on 2026-10-31.

Direct answer: Mendeley led in both surveys of graduate students that we read, but most students used no tool at all in the Greek survey. Speare (2018) found 39% of users on Mendeley, and Nitsos et al. (2022) found almost two-thirds of postgraduates were non-users. Awareness and learning time, not features, explain most of the gap.

Speare (2018) surveyed graduate students at the University of Manitoba in autumn 2015. Of 3,800 students, 138 answered, and 96 of them used reference management software. Among those users, Mendeley held 39%, EndNote 20%, Zotero 16% and RefWorks 10%. The two principal reasons for non-use were that students did not know which options existed and that learning a program takes time. Users also reported a steep learning curve and inaccurate citation styles.

Nitsos et al. (2022) collected 545 responses from postgraduates at the Aristotle University of Thessaloniki. Almost two-thirds called themselves non-users, and among users more than two-thirds chose Mendeley, followed by EndNote and Zotero. The university library officially recommends Mendeley, yet more than 60% of respondents did not know that.

Read together, the two surveys point at a plain distinction that comparison pages skip. A tool your library recommends and teaches is used by far more students than a tool that merely scores well, and the students who never start are lost to ignorance, not to dislike. Before you compare features, search your library's website for "reference management software" and note which tools it supports and which workshops it runs. Our guide to AI and research tools at each stage of a project shows where this choice sits among the other decisions of a thesis.

Direct answer: Zotero gives 300 MB of free file storage and sells 2 GB for $20 a year. Mendeley gives 2 GB free. EndNote Online gives 2 GB, while EndNote Web with EndNote 2025 gives unlimited attachment storage. JabRef keeps your library in a local file, so it has no cloud allowance to run out.

Zotero (n.d.-a) lists four storage plans for individuals: 300 MB free, 2 GB at $20 a year, 6 GB at $60 and unlimited storage at $120. The Bodleian Libraries (2026) report that Mendeley offers 2 GB of free personal cloud storage, "more than Zotero's 300 MB", and that EndNote Online is limited to 2 GB of file attachments. EndNote (n.d.) lists unlimited attachment storage for EndNote Web with EndNote 2025, and 7,500 or more predefined styles for both versions. Zotero supports more than 10,000 styles, and JabRef (n.d.-a) imports over 15 reference formats.

A worked example shows why 300 MB matters. Suppose an average journal PDF is 1.5 MB (an assumption, so check your own folder). A library of 150 PDFs takes about 225 MB and fits Zotero's free allowance. A library of 250 PDFs takes about 375 MB and does not, so you pay $20 for 2 GB, use Mendeley's 2 GB, or keep the PDFs on one computer.

Source: Zotero (n.d.-a), Bodleian Libraries (2026), EndNote (n.d.), read 2026-10-31. Vendor limits and prices change, so confirm them on the vendor page.

Direct answer: Zotero works with Word, LibreOffice and Google Docs. EndNote desktop also covers all three, but the browser-based EndNote Online lacks Google Docs support for citing as you write. Mendeley Cite is described as Word only. JabRef inserts citations into LibreOffice and LaTeX editors, and reaches Word through a file import.

The plugin is the part of reference management software you touch every day, so it decides more than any feature list. Zotero (n.d.-b) states that its Google Docs support runs through the Zotero Connector for Chrome, Firefox, Edge and Safari and requires the Zotero program to be open. The Bodleian Libraries (2026) table lists Microsoft Word, Google Docs, Apache OpenOffice, LibreOffice and Apple Pages for EndNote desktop, and warns that EndNote Online "does not support Cite While You Write in Google Docs or Word Online". The same guide is inconsistent about Mendeley and LibreOffice: its feature table lists LibreOffice, while its list of drawbacks says Mendeley Cite is limited to Microsoft Word. Check Mendeley's own page before you rely on it.

JabRef works differently from the others. JabRef (n.d.-b) offers a live interface for inserting citations into an OpenOffice or LibreOffice Writer document. For Microsoft Word, JabRef (n.d.-c) describes an export: you save selected entries as an MS Office 2007 XML file, then load it in Word through References and Manage Sources. That is a file hand-off, not a live plugin, so new sources mean a new export. For LaTeX, JabRef's Cite As You Write feature inserts citation keys into editors such as TeXstudio, TeXworks and Emacs. The sources we read do not describe an Overleaf link, so test that pairing on a short document first.

Direct answer: Compare each generated entry with your style guide on four fields: title capitalisation, journal name, author count, and the URL or access date of web sources. These are exactly where the tested tools failed, and the tool outputs whatever metadata it imported.

Kratochvíl (2017) found that all four tools struggled with online resources. EndNote did not generate the URL for 120 of 135 references to online resources, and the tools made on average 2.3 to 6.9 mistakes in producing references to online resources, a large number given that an article reference carries about 9 pieces of data. Gilmour and Cobus-Kuo (2011) found that capitalisation errors were common because the tools output article titles exactly as imported, and that journal-title abbreviations were the most common error of all. Some mistakes come from the program, but others come from imported database metadata or a mis-set citation style (Kratochvíl, 2017), so a sound tool can still print a wrong entry.

A worked example follows, using an invented article that illustrates the method and is not a source.

Run this check for every ten new sources you import, so errors never pile up before your deadline. Our guide to APA 7th referencing shows what a finished entry looks like, and web pages with no author or date are covered in how to cite a website in APA 7.

Direct answer: Ask your library which reference management software it supports, match the plugin to your word processor, compare the storage arithmetic, then run a 30-minute test with 20 sources. Switch in the first week of reading, when the library is empty and the choice costs nothing.

Follow these steps in order. First, ask your library and your supervisor, because Speare (2018) and Nitsos et al. (2022) show that support and awareness drive use. Second, match the plugin to the program you write in, using the section above. Third, estimate storage: multiply your expected number of PDFs by the average size of your own files, and compare the result with the 300 MB and 2 GB limits. Fourth, test. Import 20 sources, insert 5 citations in a short document, build the list, and check three web entries against the table above.

Gilmour and Cobus-Kuo (2011) conclude that "the choice of an RM should reflect the user's needs and work habits", which matches what the surveys suggest about adoption. If you are still undecided after the test, choose the reference management software with the larger free allowance and the plugin for your word processor, because those two facts are the hardest to fix later. Our guides to how to write a literature review and to APA 7th vs Harvard referencing cover two bigger thesis decisions.