What a Reference Manager Can and Cannot Do for a Thesis

What a reference manager does well

What does a reference manager do, in one sentence: it stores citations, organises them, and formats bibliographies in the styles your discipline uses. Zotero, Mendeley, Paperpile, EndNote, and JabRef all do this, with variation in the exact features. This is the core job the category was built for, and every modern reference manager does it competently.

Four specific jobs sit inside the category:

• Save citations from anywhere. Browser extensions pull metadata from journal pages, databases, and web sources with one click.
• Store PDFs alongside metadata. Most tools attach the full paper to the reference entry, so the source is one click away.
• Format bibliographies automatically. Choose a style (APA 7, Chicago, MHRA, Harvard, or a discipline-specific one), and the tool generates the reference list in that format.
• Integrate with word processors. Word, Google Docs, LibreOffice, and LaTeX all have integration paths, so in-text citations get inserted and updated as you write.

Every doctoral candidate should use a reference manager. Doing this work by hand across a bibliography of 200 to 400 sources is a significant time cost with no upside.

What a reference manager does not do

The category has a structural limit that becomes more visible as thesis lengths grow and AI-assisted drafting spreads. Reference managers store references. They do not verify them.

Four specific jobs sit outside the category:

• Verify that the citation exists. A reference you save (or import from an AI-assisted draft) is stored as-is. The tool does not check whether the DOI resolves or whether the paper described actually exists.
• Catch retractions after saving. A paper you save in year one and cite in year three may have been retracted in year two. Reference managers do not scan against Retraction Watch or check retraction status on a schedule.
• Verify that the source supports the claim. The tool does not read the paper. It cannot know whether the sentence you cited actually says what you claim it says.
• Cross-check metadata as journals update. Some journals reformat their metadata, change page numbers, or restructure volumes. Reference managers save what was there at the time.

Each of these gaps was small when retraction rates were lower and AI-assisted drafting was rare. The gaps are now large enough that ignoring them produces thesis bibliographies with unverified, potentially fabricated, or retracted entries.

Why the gaps matter more now

Two trends have made the gaps consequential.

Retraction rates have risen. Retraction Watch tracks more than 65,000 retractions across the academic record, with the 2023 annual count exceeding 14,000 (Retraction Watch database, 2026). A biomedical audit found 69% of dated citations to retracted works were made after the retraction had already happened. A reference list built in year one and not re-checked contains sources that have moved.

AI-assisted drafting introduces fabricated citations. Walters and Wilder (2023) found 55% of GPT-3.5 references and 18% of GPT-4 references were fabricated across 636 citations in 84 literature reviews. Bhattacharyya and colleagues (2023) found 47% of ChatGPT-generated medical references were completely fabricated. A candidate who imported AI-drafted citations into their library has a live integrity issue that no reference manager surfaces.

The result: a thesis bibliography built with a good reference manager and no additional verification is more exposed than the same bibliography would have been a decade ago.

The missing half of reference management

The gap the category leaves is verification. Existence, correctness, current status, and match to the claim being made. This is a machine task at bibliography scale. Doing it by hand for 200 to 400 sources is slow; a program does it in one pass.

Thesisroom’s citation check runs the verification in one pass, drawing on eight registries. Every entry returns as verified, flagged, or unverifiable, with a specific reason for each flag. The reference watch runs the check continuously, so retractions and corrections that happen after you save the source get flagged while you can still act.

The tools are complementary, not competing. Zotero, Mendeley, or Paperpile stores your library. Citation Check verifies it. Reference Watch keeps the verification running.

What examiners now do with bibliographies

Examiners spot-check reference lists during their read of the thesis (Park, 2003). Since 2023, that spot-check has become more common and more thorough. Panels routinely resolve DOIs of citations that look important. A citation that fails a Crossref lookup surfaces at exactly the moment the examiner tries to verify the source.

The candidate who ran continuous verification arrives at the viva without those specific surprises. The candidate who relied on the reference manager alone is exposed to whatever has happened to the field’s evidence base in the years the thesis was being written.

What Thesisroom does with this

Thesisroom is not a reference manager. It does not compete with Zotero, Mendeley, Paperpile, EndNote, or JabRef. It runs the verification layer that all of them lack. The Monthly plan is $12 per month and covers continuous re-checking across the whole degree.

The reference list you build in your preferred manager stays where you built it. What Thesisroom adds is the read against the academic record.

Frequently asked questions

Do I still need a reference manager if I use Thesisroom?

Yes. Thesisroom does not store references or format bibliographies. It verifies references that you have stored elsewhere. Use Zotero, Mendeley, Paperpile, or your preferred manager to store; use Thesisroom to verify.

Which reference manager works best with Thesisroom?

Any reference manager that exports BibTeX or RIS works. Zotero, Mendeley, Paperpile, EndNote, and JabRef all support both formats. Export your library, upload the file, and Citation Check runs the verification pass in one go.

Does Zotero not check citations at all?

Zotero can refresh metadata from Crossref on manual command, but it does not automatically re-check for retractions or verify that a saved citation resolves. The automatic re-checking is the specific gap.

How much does it matter if a citation is retracted?

For contextual or historical citation, less. For citations that support a load-bearing claim in your argument, more. A retracted paper whose finding your argument depends on may require you to reframe the claim. Silent citation of retracted sources is one of the specific things examiners now look for (Park, 2003).

Can Thesisroom generate references I do not have?

No. Thesisroom does not generate citations. Find Sources returns verified papers that support a claim you specify, so you can decide whether to cite them. The generation step (writing sentences with new citations) stays with the writer.

What is the difference between a reference manager and citation software?

The terms overlap. “Reference manager” usually refers to the storage and organisation tool (Zotero, Mendeley). “Citation software” sometimes refers to word-processor plug-ins (Mendeley Cite, Zotero for Word). Both categories overlap and are often bundled in the same product.

One takeaway before you close this tab

Reference managers do a specific job well. They do not do the verification job at all, and that gap has become more consequential in recent years. When you want the verification layer added on top of whichever manager you use, Thesisroom’s citation check runs the check in one pass, drawing on eight registries.

References

Bhattacharyya, M., Miller, V. M., Bhattacharyya, D., & Miller, L. E. (2023). High rates of fabricated and inaccurate references in ChatGPT-generated medical content. Cureus, 15(5).

Park, C. (2003). Levelling the Playing Field. Higher Education Review, 36(1), 47-67.

Retraction Watch. (2026). Retraction Watch Database. https://retractionwatch.com

Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13.