Citing a retracted paper — and why no reference checker will tell you

A retracted paper does not disappear. Its DOI still resolves, its title is unchanged, its authors are still its authors, and every automated check you run on it comes back clean. That is precisely the problem.

This is the defect class we are least able to help with directly, so it is worth being exact about what a reference checker can and cannot see — and about the one thing our tool gives you that makes the manual check quick.

What a retraction looks like to software

Take an invented but entirely typical case. A candidate cites Halloran & Vidmar (2018), a methods paper, in her literature review. Two years after publication the journal retracts it — an error in the analysis, discovered by a reader, acknowledged by the authors. The journal does the correct thing: it leaves the paper online, stamps it RETRACTED, and publishes a notice explaining why.

Now run every check a citation tool performs.

  does the entry parse?          yes — author, year, title,
                                 journal, volume, pages

  does the citation pair with
  the reference?                 yes — exact author and year

  does the DOI resolve?          yes — 200 OK, publisher page

  does the record exist in
  the databases?                 yes — Crossref has it,
                                 the metadata is complete

  is the reference recent?       2018 — flagged as ageing,
                                 like any other 2018 paper

  VERDICT                        matched, valid, unremarkable

Every signal is green, and every signal is correct. The paper exists, the metadata is accurate, the link works. Nothing in the mechanics of reference checking asks the question that matters, which is not "does this paper exist" but "does the field still stand behind it".

Retraction is not a property of the reference. It is a property of the literature's opinion of the reference, and it arrives after the entry was written.

Why we do not claim to detect it

We could add a retraction check. We have not, and the reason is worth stating plainly rather than dressing up as a roadmap item.

Retraction data is uneven. It lives in publisher notices, in a specialist database, and in Crossref's update metadata — and coverage across those varies by field, by publisher and by year. A check that finds most retractions is worse than no check at all in one specific way: it teaches you that a clean result means clean. The first time you rely on that and a retraction is missed, the confidence the tool gave you is the thing that hurt you.

This is the same reasoning that keeps the extractor generous and the screening visible. We would rather tell you a question is yours to answer than answer it unreliably on your behalf.

What the tool does give you

One thing, and it is more useful than it sounds: every reference is followed out to its source. The verification pass resolves each DOI and confirms it lands on the publisher's own page for that paper.

Which means the retraction notice — if there is one — is exactly one click away from each entry in your report, on the publisher's page, in the publisher's own words. You are not searching for the paper. You are opening a link that has already been checked to land in the right place.

  1  open the report's link for the entry
     (already resolved and verified)

  2  look at the publisher page for a banner:
     RETRACTED · WITHDRAWN · EXPRESSION OF CONCERN
     · CORRECTION

  3  if there is one, read the notice — the
     reason decides what you do next

  worth doing for: the load-bearing citations
  in your argument, not all 300

That last line is the practical part. Checking three hundred references by hand for retraction is not a realistic use of anyone's final month. Checking the eight or ten that your central claims rest on is an hour, and those are the ones where a retraction would actually damage the argument.

What to do when you find one

The instinct is to delete the citation. That is usually wrong, and occasionally it is the worst available option.

What matters is what the paper was carrying in your argument. If it was one of several sources supporting an established point, replace it — the point survives, the evidence for it changes. If it was the sole support for a claim, the claim itself now needs revisiting, and quietly swapping in a different reference conceals a real problem rather than fixing it.

And there is a third case that people miss: sometimes a retracted paper is still the correct citation. If you are writing about the retraction, or about the episode, or using it as an example of a methodological failure, then the retracted paper is the source and citing it is right. What the reader needs is for you to say so, in the text.

None of these is a decision software should make. Which is, again, why the tool resolves the link and stops.

Why doctoral work is unusually exposed

The timing is what makes this a thesis problem rather than a general one.

A journal article is written and submitted inside a year or so. A doctorate runs four, sometimes six. The literature review is often drafted early — it is the part that exists first — and then sits, largely untouched, while the empirical chapters are written. By submission, some of those citations are five years old and were last looked at four years ago.

  year 1   literature review drafted
           paper read, entry added, sound

  year 3   the retraction notice appears
           you are writing chapter 5 and have
           no reason to revisit chapter 2

  year 4   submission
           the entry is unchanged, and nothing
           in your process would have told you

Nothing in that sequence involves carelessness. The candidate did the reading, recorded it properly, and moved on to the next chapter — which is what they were supposed to do. The literature simply changed its mind behind them while they were looking elsewhere.

It is also why the check is worth doing once, late, rather than continuously. A sweep of your load-bearing citations in the month before submission catches the whole four-year accumulation in one sitting. Doing it earlier means doing it again.

The near-miss that is more common

Far more frequent than retraction, and much easier to fix: the paper was not retracted but it was corrected, or it was superseded by a later version from the same authors. The DOI resolves, the notice on the page says Correction, and the figure you cited may have changed.

That case the tool does help with, because a superseding version is a newer record for the same work and the update pass looks for exactly that. It is the same mechanism as refreshing an ageing bibliography, arriving at a more urgent version of the same answer.

What this does not claim

Citing a paper that was later retracted is not misconduct and this post should not be read as suggesting otherwise. In the overwhelming majority of cases the candidate read the paper in good faith, years before the notice existed, and had no mechanism for learning that it had changed status. The literature moved after the citation was written — which is the same ordinary problem as a bibliography ageing underneath you, in a sharper form.

And CitationLab does not check retraction status. It resolves your references to their sources so that you can, in the small number of places where it matters.