Home / Guides

How to check that your references are real

Every reference should point to a work someone can find. A method for checking yours before a reviewer does.

DeskReady · Published 16 October 2026

A reference list is assembled over months, from BibTeX exports, copied citations, a co-author's library and, increasingly, suggestions from AI tools. Most entries are fine. The few that are not are hard to spot, because a wrong reference looks exactly like a right one: authors, a plausible title, a venue, a year.

Reviewers and editors do check. A reference that leads nowhere makes them wonder what else was not checked, and a citation that does not exist at all is now one of the clearest signs of unchecked AI-written text.

What goes wrong in reference lists

Works that do not exist

Chat assistants are good at producing citations that look real: real author names, a title in the style of the field, a well-known venue. Some of them were never published. If any part of your reference list came from an AI suggestion, every one of those entries needs checking against a real database, not against the assistant.

Wrong or drifting details

More common than invented works are real works cited wrongly: a title reworded from memory, the preprint year next to the journal volume, a missing venue, a conference paper cited as its arXiv version. Each one makes the work harder to find.

Things that are not papers

Datasets, software, standards, agency notices and web pages are legitimate references, but they need what makes them findable: a DOI or URL, a version, an access date. A dataset cited like a journal article, with no repository and no version, is a dead end.

How to check every reference

1. Start with the DOI. If an entry has one, open https://doi.org/ plus the DOI. It should land on the work you meant, with the same title and authors.
2. No DOI: search the title in Crossref (search.crossref.org), which covers most journals and conference proceedings. Paste the whole citation; it matches on the full text, not just the title.
3. Use the database of your field when Crossref has nothing: arXiv for preprints (check the ID resolves to that title), DBLP for computer science, NASA ADS for astronomy and physics, OpenAlex across fields.
4. Compare the details, not just the title. The record you find should have the same first author, a year that agrees and the same venue. A year that is off by one is often just the online date against the print issue; anything else that differs, correct.
5. Check what the entry is missing. Venue, year, volume and pages for articles; DOI or URL and version for datasets and software.
6. Treat anything you cannot find as unverified. Either find the real work and correct the entry, or remove the citation.

How often references fail

When we checked the full reference lists of ten recent arXiv preprints, 418 references in all, every scholarly reference but three was found in one of the databases above. The three were a dataset, a government notice and a journal article whose author name had been garbled by the PDF (an umlaut printed as a separate character). None of the ten invented a work, which is what you would hope for published preprints. In a manuscript that has not yet been checked, the share is unknown, and that is the point of checking.

Doing it with DeskReady

DeskReady looks up every reference in your PDF in Crossref, PubMed (via Europe PMC), arXiv, DBLP, NASA ADS, INSPIRE-HEP, OpenAlex and DataCite, checks every DOI for a retraction notice, and lists the ones no database knows so you can check them, next to the rest of its findings: contradictions, numbers that don't add up, figures that disagree with the text. Web pages, books and reports are not flagged just for being absent from these databases.

To check a reference list on its own, use the free reference checker. More in the pre-submission checklist.