How to check whether a citation is real

A method you can run by hand, and what to look for.

AI writing tools produce references that look correct in every visible way: a real journal, plausible authors, a well-formed DOI, page numbers in the right range. The work behind them was never written. Courts have sanctioned lawyers over it and journals have retracted papers over it, so reviewers and supervisors now spot-check reference lists.

The four checks, in order of strength

1. Resolve the DOI

If the citation carries a DOI, paste it after https://doi.org/ in your browser. A real DOI lands on the publisher's page for that exact work. A fabricated one returns "DOI Not Found." This is the strongest single test, because a DOI is registered centrally and cannot be guessed into existence. Watch for the near-miss: a DOI that resolves to a different paper means the citation was assembled from parts.

2. Search the exact title

Put the title in quotation marks and search CrossRef, OpenAlex, or Google Scholar. Quotation marks matter: without them the search engine helpfully finds something similar, which is exactly the failure you are trying to detect. No result from any of the three is strong evidence the work does not exist.

3. Check the author actually wrote it

Fabricated citations often attach a real, well-known author to a paper they never wrote. Look at the author's own publication list, their ORCID record, or their department page. A prolific author in the right field is exactly the kind of name a fabrication attaches itself to.

4. Check the container

Confirm the journal exists, that the volume and issue correspond to the stated year, and that the page range falls inside that issue. Invented citations frequently place a real article in the wrong volume, or cite a journal that stopped publishing before the stated date.

What a fabricated reference usually looks like

Everything is plausible and nothing is checkable. Real journal, real-sounding authors, and a DOI that fails to resolve.
The title is generic for its field. Invented titles read like the average of a literature, because that is how they were produced.
The metadata is internally inconsistent. Volume, year, and page range disagree with each other or with the journal's actual run.
It sits among real references. A list is rarely all invented. Checking a sample and stopping is how fabrications survive to submission.

Where the manual method runs out

The four checks work, and they take two to four minutes per reference. A thirty-item bibliography is therefore a two-hour job, which is why most people sample instead of checking every line. Two other gaps are worth knowing: a citation can be entirely real and still point at a paper that has since been retracted, and coverage is thinner for non-English sources, legal citations, and grey literature, where absence from a registry is weak evidence of anything.

The shortcut: paste your bibliography into CiteFact and every entry is checked against CrossRef and OpenAlex at once, retractions included. Free for 10 references, no signup.