How to Check Whether a Paper You Cited Has Been Retracted (2026 Guide)

This is one of those checks nobody teaches you and everybody assumes you already do. It takes about five minutes for a single paper and maybe half an hour for a full reference list. I want to walk you through exactly how to do it, because the alternative is trusting that none of the forty or sixty papers you gathered over eighteen months has been withdrawn in the meantime, and that is a bigger bet than most people realise.

A Retraction Is Not a Deletion

When a journal retracts an article, it is saying the findings should not be relied on. That is the whole meaning of it. The paper does not vanish from the internet, and it does not vanish from Google Scholar. It usually stays exactly where it was, sometimes behind a notice, sometimes not, and anyone who downloaded the PDF before the retraction has a copy that looks completely normal.

The reasons vary more than you might expect. Fabricated or manipulated data, duplicated images, plagiarism, undisclosed conflicts of interest, and paper-mill output are all in there. So is honest error. Plenty of retractions happen because the authors themselves found a mistake in their own analysis and asked for the paper to be pulled, which is the system working rather than failing. That distinction matters if you ever need to discuss a retracted study in your writing, because “retracted” and “fraudulent” are not synonyms and you should not imply they are.

Retraction Watch keeps the largest list of these that exists. By late 2025 the database held more than 63,000 entries, and Crossref acquired it in September 2023 and made the whole thing freely available, which is the reason all the checks below are free. You do not need an institutional subscription for any of this.

Why Almost Nobody Catches One by Accident

Think about the actual shape of a research project. You find a paper in early 2025, save the PDF into a folder, write your notes, export the citation into your reference manager, and move on. The retraction notice is published in late 2025. Your folder still contains the same PDF. Your reference manager still contains the same entry. Your notes still say what they said. Nothing anywhere in that chain pings you.

There is no natural moment in a normal workflow where anyone checks, which is why retracted papers keep collecting citations after they have been withdrawn, frequently in papers that make no mention of the retraction at all. A study published in June 2026 looked at how quickly Wikipedia corrects citations to retracted papers and found a median correction time of 3.68 years. Wikipedia has thousands of active volunteer editors and a culture built entirely around sourcing. If it takes them the better part of four years, one postgraduate student working alone at eleven at night is not going to spot it on instinct.

So the check has to be deliberate. It has to be a step you do, not a thing you hope you would notice.

The Five-Minute Check on a Single Paper

If you have one paper you are unsure about, here is the order I would work through. Stop as soon as you get a clear answer.

1. Go to the DOI. Open the paper on the publisher’s own site rather than the PDF on your hard drive. The publisher page is the authoritative record, and a retraction will normally show as a banner, a linked retraction notice, or “RETRACTED:” in front of the title. This single step catches most cases.
2. Search the Retraction Watch database. It is free and searchable by DOI, title, or author. This is the one to trust when the publisher page is ambiguous, and it also tells you the reason for the retraction and the date, which the publisher page sometimes buries.
3. Check PubMed if you are in health or life sciences. PubMed tags retracted publications and lets you filter by publication type, so a quick title search there gives you a clear yes or no.
4. Look at Google Scholar, but do not rely on it. Scholar does label some retracted work, and it is convenient because it is probably already open in another tab. It is also inconsistent, so treat a clean Scholar result as weak evidence rather than an all-clear.
5. Glance at PubPeer. Comments there are not retractions, and a flagged paper is not automatically a bad paper. But if a study is under sustained methodological criticism, that is worth knowing before you build a chapter on it.

For most papers you will be done at step one or two.

Scanning a Whole Reference List Without Losing Your Afternoon

A full bibliography is a different job, and doing forty individual DOI lookups by hand is grim. Two things make it manageable.

First, collect your DOIs in one place. If your references live in a reference manager you can export them, and if they live in a Word document you can pull the DOIs out into a plain list. Either way, having them all in one column turns a vague worry into a finite task.

Then triage rather than treating every entry equally. The papers that deserve careful checking are the ones doing load-bearing work in your argument, the studies you cite for a headline finding or a number you quote. After that, prioritise older papers, since retraction usually lags publication by a couple of years, and anything you picked up secondhand from another paper’s reference list rather than reading yourself. Those secondhand citations are where retracted work travels, because nobody in the chain ever opened the original.

Papers you cite once, in passing, as one of three examples of a general trend, matter less. If one of those turns out to be retracted you cut it and the sentence survives.

Let Your Reference Manager Watch for You

This is the part that genuinely changes your odds, and it is the reason I keep recommending Zotero to students even though the three big reference managers are close on most other features.

Zotero has been integrated with the Retraction Watch data since 2019. It checks your library automatically, flags retracted items in the item list, and shows you a warning with the retraction details when you click one. Better than that, if you try to cite a flagged paper through the word processor plugin it will stop and ask whether you really want to, and if a paper you already cited gets retracted later it warns you the next time you update the document’s citations. That last behaviour is the one that saves people, because it catches the retraction that happened after you wrote the sentence.

Two caveats worth knowing. The feature works on items that have a DOI or a PMID recorded, which covers roughly three quarters of the Retraction Watch data, so an entry you typed in by hand without a DOI is invisible to it. And you do not need to sync your library or upload anything for this to work, which I mention because a few people avoid the feature thinking it sends their reading list somewhere.

If you are already deep into Mendeley or EndNote, that is not a reason to migrate mid-thesis. It does mean the manual check matters more for you. Our comparison of Zotero, Mendeley, and EndNote goes through the wider trade-offs if you are still deciding.

You Found One in Your Draft. Now What?

Do not panic, and do not quietly hope it is fine. There are really three situations.

If the retracted paper is carrying a claim your argument depends on, you have to replace it. Go back to the source of that claim and find out whether it holds up independently. Sometimes it does, and two other studies support the same point, in which case you cite those and move on. Sometimes it does not, and the honest outcome is that the claim comes out of your work. That is uncomfortable, and it is also the correct result, because the alternative is building a chapter on a finding the field has withdrawn.

If it is a minor supporting citation, cut it. The sentence will usually stand on its own or work with one of your other sources.

And occasionally you will want to keep it on purpose. Discussing a retraction is legitimate, especially if the paper shaped the field before it was withdrawn or if the retraction is itself part of your story. In that case cite it explicitly as retracted, name the retraction in the sentence, and cite the retraction notice alongside the original. Something like: “An early and widely cited study reported a strong association, though the paper was retracted in 2024 following concerns about the underlying data.” Your marker now knows you know.

If the work has already been submitted or published, tell your supervisor or the editor rather than sitting on it. Nobody enjoys that email. It is still much smaller than the version where somebody else finds it.

Where AI Tools Will Not Save You

Retraction status is metadata that lives outside the paper. It is published after the fact, in a separate notice, and the text of the article does not change. That is why language models are poor at this: a model that has read a paper’s contents has no way of knowing that a notice appeared eighteen months later unless it goes and looks the record up. A 2026 study in JMIR tested freely available AI tools on exactly this and found they cannot reliably identify retracted literature.

Which gives you two distinct reference problems to guard against rather than one. The first is references that do not exist, the confident, plausible, entirely invented citation, and we covered how to catch those in how to spot fake AI citations before you submit. The second is references that exist perfectly well and have been withdrawn. The checks are different, and passing one tells you nothing about the other.

Worth saying plainly, since Litrevu is an AI tool: working from PDFs you gathered yourself removes the invented-citation problem, because every source is a real paper you chose and can open. It does not remove the retraction problem. No tool that reads your PDFs can tell you what a journal published about them afterwards. That check stays yours, and it belongs in your process alongside the reading itself. If you want the fuller picture of what these tools can and cannot do, can AI write a literature review is our honest attempt at it.

Make It a Step, Not a Heroic Act

The reason this gets skipped is that it feels like extra work at the exact moment you have the least energy. So build it in at two points where it costs you almost nothing.

Check when you add a paper. You are already on the publisher page downloading the PDF, so noticing a banner takes no additional time. If you are recording sources properly as you go, whether in a matrix or an annotated bibliography, add a column or a line for “retraction checked” with the date. It looks fussy. It is the difference between checking once and re-checking everything in a panic later.

Then check again before you submit, on your load-bearing sources only. Twenty minutes, maybe thirty. Put it in the same pass where you verify that every citation actually says what you claim it says, which is a habit worth having anyway.

If you are still assembling your sources rather than writing them up, this fits naturally alongside your search strategy and your inclusion and exclusion criteria. A retraction check is really just one more eligibility rule: the paper has to still stand.