Why does my DOI open a different article?

Almost always because one character is wrong, and DOI suffixes at most publishers run in sequence rather than at random. We sampled 600 journal articles from CrossRef and changed the last digit of each DOI by one: 59% of the time the altered string was still a registered article, and in 350 of those 352 cases it was an article in the same journal. So a single mistyped or mis-generated digit usually lands on a real paper rather than on an error page.

Is a wrong DOI plagiarism?

No. A wrong identifier is a citation error, and APA states directly that incorrect citations are not plagiarism when the mistake is a minor editorial oversight rather than an attempt to take credit for someone's ideas. It can still cost you marks, and it becomes a different problem if the reference was produced by a tool and never opened, because then the error is not in the identifier but in the claim that you read the source.

How do I find the correct DOI for a paper?

Search the exact title rather than guessing at the identifier. CrossRef's public search takes a full reference and returns the registered record with its DOI, and the article's own landing page prints the DOI the publisher assigned. Take the identifier from the record, never from a formatted citation string, and confirm that the first author, year and journal on the record match the paper you actually read.

Does a working DOI mean the citation is real?

No, and this is the check that misleads people most. In our own August 2026 test of current chat models, a bad DOI from the strongest model was almost twice as likely to resolve as to be dead: 12% opened a real but different paper against 7% that opened nothing. A study of citations that slipped through peer review at a major conference named the same pattern identifier hijacking, where a valid identifier creates false verifiability. The check is not whether it resolves, it is whether it resolves to the paper you meant.

The DOI links to the wrong paper

The CiteOwl Team· Updated 29 August 2026· 12 min read

You clicked the DOI in your own reference list and a different article opened. Right journal, right publisher, wrong paper. This is the citation error that survives every check students are taught, because the identifier works, the prefix belongs to the correct publisher, and the page that loads is a genuine piece of scholarship. It happens because DOI suffixes are usually sequential, not random: we changed one digit in each of 600 real DOIs and 59% of the time the altered string was still a registered article, almost always in the same journal. Below is how to find the paper you actually meant, and how to tell an honest slip from the version that gets you in trouble.

Contents

Why this error survives the standard checks
We changed one digit in 600 DOIs
Where a wrong DOI comes from when no AI is involved
And when AI is involved, it has a name
Finding the right DOI, in about a minute
Fixing it in a draft, a submission or a published paper
Citing from the record instead of from a string

First, the reassuring part. A DOI that opens the wrong article is usually an ordinary mistake, and every checking guide on the internet, ours included, has been telling you to run a test that cannot catch it.

Why this error survives the standard checks

There are two checks in common circulation. The first is to paste the DOI into doi.org and see whether it resolves. The second, which we recommend in the DOI that does not work, is to check whether the prefix belongs to the publisher of the journal you cited, since an invented identifier often borrows a familiar prefix and bolts an invented suffix onto it.

A wrong-paper DOI passes both. It resolves, because it is a registered identifier for a real article. The prefix is right, because the article it belongs to is usually published by the same publisher, in the same journal, often in the same issue. Everything you were told to look at looks correct. The only thing that is wrong is the one thing neither check examines, which is whether the record at the other end is the work you cited.

The check is not "does this DOI resolve". It is "does this DOI resolve to the paper I read". Those are different questions, and only the second one is worth your time, because a dead DOI announces itself and a live wrong one does not.

We changed one digit in 600 DOIs

Why is landing on a different real paper so easy? Because the space around a DOI is crowded. A suffix like 10.1016/j.rser.2009.07.037 is not a random string. It is a journal code, a year, a month and a sequence number, and the article next to it in that issue is .038. Change one digit and you have not produced nonsense, you have produced the neighbouring article.

We wanted to know how general that is, so we measured it. We drew a random sample of 600 journal articles from CrossRef's own sampling endpoint, with no query terms, so the sample is a slice of the registered literature rather than of any topic we chose. For each DOI we took the last run of digits, added one and subtracted one, keeping the number of digits the same, and asked CrossRef whether each of those 1,186 neighbouring strings is a registered work.

Result	Count	Share
DOIs where a one-digit change still lands on a registered article	352 of 600	58.7%
Of those, where the article is in the same journal	350 of 352	99.4%
Individual neighbouring strings that resolved	646 of 1,186	54.5%

So for a clear majority of published articles, a single wrong digit does not produce an error page. It produces a different real paper, in the same journal, published within weeks of the one you meant. The raw dataset is public under CC BY, with the method, the counts and forty worked examples.

Here is one of them, from the sample. The real article is "Energy and renewable energy scenario of Pakistan", DOI 10.1016/j.rser.2009.07.037. Add one to the last digit and 10.1016/j.rser.2009.07.038 opens "Application of geothermal energy for heating and fresh water production". Subtract one and .036 opens "State of the art on high-temperature thermal energy storage". All three are in Renewable and Sustainable Energy Reviews, all three are real, and a reference list containing any of them looks equally correct.

The density is not the same everywhere, and that is worth knowing when you decide how carefully to check. These are the publishers with enough articles in our sample to say anything about.

Publisher	Articles sampled	A one-digit slip lands on a real paper
Wiley	59	86%
SAGE Publications	13	85%
Oxford University Press	14	79%
Informa UK (Taylor & Francis)	25	60%
Elsevier	89	55%
American Chemical Society	13	46%
Springer Nature	56	36%
Cambridge University Press	9	22%

Read the small samples with appropriate suspicion: nine Cambridge articles cannot tell you much about Cambridge. The spread is real, though, and it follows from how each publisher builds suffixes. The ones at the top run tight sequences per issue. The ones at the bottom leave gaps or mix in identifiers that are not simply counted upward.

Two things this measurement is not. It does not say how often people mistype DOIs, only what happens when they do. And a neighbour that resolves is not evidence that anyone made that particular error. Read it as the reason a wrong-paper DOI fails silently rather than loudly, which is the property that matters when you are deciding what to check.

Where a wrong DOI comes from when no AI is involved

This error long predates chatbots, and assuming otherwise will make you look in the wrong place. The common origins:

A line break in a PDF. Copying a DOI out of a reference list in a PDF often drops or duplicates a character where the string wrapped. One digit is all it takes.
A reference manager importing from the wrong page. Grabbing metadata from a search result, a table of contents or an abstract aggregator rather than the article page can attach the identifier of a neighbouring item, and the entry then looks complete, which is why nobody re-checks it.
Front matter and back matter. Editorials, mastheads, corrections and book reviews get their own DOIs in the same sequence as articles. One entry in our sample resolved to "Masthead" in Advanced Energy Materials. A reference to a real journal issue's front matter is still a wrong reference.
Copying a citation from another paper's reference list. If their DOI was wrong, yours now is too, and the error propagates through everyone who copies it. This is also why a reference you never opened is a risk even when it came from a published source.
Preprint versus version of record. These are different objects with different identifiers, and quoting a page number from one while citing the other is a version mismatch rather than a typo, but it produces the same "that is not the paper" reaction from a marker.

None of that is misconduct, and it is worth being clear about this because the panic is common. APA states it plainly: using incorrect citations, including "forgetting or mistyping an element in a reference list entry", is not considered plagiarism when the error is minor and attributable to an editorial oversight rather than an intentional attempt to steal someone's ideas. It adds, correctly, that such errors "may still result in deductions on an academic assignment". A wrong DOI is a marks problem, not an integrity problem.

With one exception, which is the next section.

And when AI is involved, it has a name

If the reference came out of a chatbot and you never opened it, the wrong DOI is not a typo. It is the visible edge of a reference that may not correspond to anything.

We measured this directly in August 2026, asking three current models for a hundred references each on ten ordinary essay topics with no web access, then resolving every DOI against CrossRef. For the strongest model, 19% of DOI-bearing references were unusable, and the split is the finding: 7% were dead and 12% resolved cleanly to a different real paper. A bad DOI from that model was almost twice as likely to work as to fail. The smaller model produced 25% dead and 16% wrong-paper. Full numbers, method and the wrong-paper examples are in why AI makes up citations and in the published dataset.

The pattern has a name in the research literature now. A February 2026 preprint by Samar Ansari at the University of Chester takes 100 fabricated citations that appeared in papers accepted at NeurIPS 2025, each having passed review by three to five expert reviewers, and sorts them into five failure modes. One of them is identifier hijacking: "the citation uses a valid scholarly identifier (arXiv ID, DOI, or URL) that points to a real paper, but the authors, title, or other metadata do not match the paper at that identifier". It was the primary failure in 4% of those citations and a secondary characteristic of 29% of them. The term the paper gives to what that buys is exact and worth stealing: false verifiability. The citation passes the "the link works" test while failing the "the content matches" test.

A 2025 evaluation of GPT-4o's citation output found the same thing at a much higher rate: when the model attached a DOI to a fabricated citation, 64% of those DOIs resolved to real but unrelated papers. For a fabricated reference, in other words, a working link is close to meaningless as evidence.

Finding the right DOI, in about a minute

Do not try to repair the identifier by guessing. Find the paper, then take its identifier from the record.

Search the title, not the DOI. Put the article title in quotation marks into CrossRef search or Google Scholar. If a paper by that title exists, this finds it in seconds along with its registered DOI.
Open the publisher's page for it. The DOI printed on the article's own landing page is the authoritative one. Take it from there, not from a formatted citation, not from a search snippet, and not from the reference list of another paper.
Match four fields before you accept it. First author, year, journal and volume. If any of the four disagrees with the paper you read, you have found a different work with a similar title, which is common in fields with standard phrasings.
Confirm it in the other direction. Paste the DOI you are about to use into our free DOI lookup, which shows the record the identifier is registered to, and read the title it returns rather than glancing at it. The failure this whole page is about is caught here and nowhere else.
If nothing has that title, stop repairing and start checking. A title that no index knows is the signature of a fabricated reference, not a mistyped one. The full routine for that is in how to check if a citation is real, and our citation checker will run a whole reference list through it at once.

One case deserves its own note. If the paper genuinely has no DOI, which is normal for older articles, book chapters and a great deal of humanities scholarship, do not borrow a nearby one to fill the field. An empty DOI field is correct and every style permits it.

Fixing it in a draft, a submission or a published paper

What you do depends on where the reference has got to.

In a draft. Fix it, then check the neighbours. If one DOI in your list came from a copied citation string or a chatbot, the others from the same source deserve the same scrutiny, and it is faster to check twenty than to argue about one.

In something you have already submitted. Tell your marker before they find it, in one sentence, without a paragraph of apology: the DOI on reference twelve is wrong, the correct one is this, here is the paper I actually used. A student who reports a citation error reads as someone who checks their work. The alternative reads as someone who did not open their sources, which is a much worse impression and the one that invites the whole list to be examined. If the problem turns out to be bigger than one digit, my professor said my sources do not exist covers that conversation.

In something published. Contact the journal. Publishers issue corrections for reference errors routinely, and CrossRef metadata can be updated by the publisher that deposited it. This is dull administration, not a scandal.

Citing from the record instead of from a string

Every version of this problem comes from the same place: a citation was assembled as text rather than pulled from a record. Once a reference exists only as a formatted string, nothing about it can be verified without going and finding the paper again, and the string will be copied onward by anyone who quotes you.

The habit that prevents it is unglamorous. When you decide to use a source, open it, and take the identifier from the article rather than from whatever handed you the reference. Everything else, formatting, style, ordering, is mechanical and a tool can do it. Knowing that the record and the paper are the same object is the part that requires you.

That is the pairing CiteOwl is built around. When the agent uses a source it retrieved the record from CrossRef, OpenAlex or Unpaywall, kept the passage the claim came from, and attached both to the sentence, so the identifier in your reference list is the one the index returned rather than a string that was written to look like one. It also means the check in this article is already done for every reference before you see it, which is the only version of this advice that survives a deadline.