CiteOwl

Blog/Getting citations right

My reference has no DOI

The CiteOwl Team· Updated 29 August 2026· 13 min read

An empty DOI field is correct. APA, MLA and Chicago each publish a rule for a source that has none, and not one of the three asks you to produce an identifier from somewhere else. Two things come before that, though. Check whether the paper really has no DOI, because a great many have one that was added years after the copy you are holding was printed, including papers from 1925. And once you are sure, do not fill the blank with something that looks close: we changed one digit in each of 600 real DOIs and 59% of the time the altered string was still a registered article in the same journal, so a borrowed identifier survives every check a marker would run.

Almost every guide on this question starts at step two. It tells you what to put in the reference once you know there is no DOI, and it never asks how you know.

First, check whether it really has none

You looked at the PDF and there was no identifier on it. That is weaker evidence than it feels, because the DOI printed on a PDF is a snapshot of what the publisher put there on the day the file was made. Registration for older articles happened afterwards, sometimes decades afterwards, when publishers deposited their archives in bulk.

The dates make this concrete. Crossref's own history records that a working group met in December 1999, that the organisation was incorporated in January 2000, and that "the Crossref system, the first collaborative reference linking system, went live in June 2000". Nothing published before that carried a DOI when it was printed. Everything you now see on an older paper was added later.

Three examples, each found in under a minute by searching the title:

• Shannon's 1948 "A Mathematical Theory of Communication", in the Bell System Technical Journal, carries 10.1002/j.1538-7305.1948.tb01338.x.
• Manabe and Wetherald's 1967 "Thermal Equilibrium of the Atmosphere with a Given Distribution of Relative Humidity", in the Journal of the Atmospheric Sciences, carries 10.1175/1520-0469(1967)024<0241:TEOTAW>2.0.CO;2.
• Hardin's 1968 "The Tragedy of the Commons", in Science, carries 10.1126/science.162.3859.1243.

The step itself is short. Search the title, not the identifier, at CrossRef search, or paste the whole reference into the API as api.crossref.org/works?query.bibliographic= followed by your reference text, which returns the registered records as JSON and takes about as long as loading a page.

Then read the result instead of taking the top one. For two of those three examples the first hit was the wrong object. The Shannon search returned a 2021 anthology chapter that reprints the paper, with a DOI of its own. The Hardin search returned a 2022 article revisiting Hardin before it returned Hardin. Reprints, translations, anthology chapters and later commentary all carry their own identifiers, so check the first author, the year and the journal against the paper in front of you before you copy anything.

Why "CrossRef has no record" is not "there is no DOI"

Here is where most advice stops, and it stops one layer too early. CrossRef is a DOI registration agency. It is not the DOI system.

The International DOI Foundation lists the agencies, eleven of them, and CrossRef is the largest rather than the only one. DataCite sits beside it, along with Airiti, Chinese DOI, CNKI, the Japan Link Center, the Korean DOI Center and the Publications Office of the European Union, plus three that identify films, construction products and performers rather than scholarship. CrossRef's entry on that page describes it as holding "~180 million DOI records representing more than 90% of all DOI usage", which is dominant but is not all of it. DataCite carries most of the remainder that a student runs into: datasets, software, theses deposited in institutional repositories, and preprints.

Preprints are the cleanest demonstration. arXiv states that "all articles submitted to arXiv are automatically assigned DOIs that correspond to their arXiv ID", using the prefix 10.48550 and registering them through DataCite. Ask CrossRef's API for 10.48550/arXiv.1706.03762 and it answers 404. Ask DataCite and the record is right there, titled "Attention Is All You Need", publisher arXiv, year 2017. Paste it into a browser and it opens the paper. The identifier is real and permanent. The wrong registry was asked.

In practice that means two attempts, not one. Search the title in CrossRef. If nothing comes back and the source is a preprint, a thesis, a dataset or a report from a repository, look for the identifier on the landing page instead, then confirm it resolves. Our free DOI lookup does the resolving half and shows you the record an identifier is registered to, which is also the check that catches the failure in the next section.

Never fill the blank with something that looks close

An unfinished-looking reference is uncomfortable, and there is usually a plausible identifier within reach: the one on the article next to yours in the same issue, or the one attached to the version of the paper you did not read. Do not use it. This is the one part of the page that is not a formatting question.

We measured why. Taking 600 journal articles drawn at random from CrossRef's index, we changed the last digit of each DOI by one in each direction and asked whether the altered string was still a registered work. For 352 of the 600, or 58.7%, at least one neighbour resolved, and in 350 of those 352 cases the neighbour was an article in the same journal. The dataset is public, and the mechanism is covered in full in the DOI links to the wrong paper.

The consequence is what matters here. A near-miss identifier does not produce an error page. It produces a real article, by real authors, in the right journal, published within weeks of the one you meant. It passes the resolve test, it passes the prefix test, and the only way anyone catches it is by reading the record at the other end. A blank field is visibly incomplete and costs you a mark. A wrong identifier is invisibly incorrect and reads as a source you never opened.

The same warning covers four everyday versions of the borrowed DOI:

• The issue instead of the article. Front matter, editorials and whole issues get their own identifiers in the same sequence.
• The review instead of the book. A published review of a monograph has a DOI; the monograph may not.
• The preprint instead of the published version. These are different objects. The 2015 batch normalisation paper from ICML has no CrossRef record for the proceedings version, while the arXiv preprint has 10.48550/arXiv.1502.03167. Citing one and linking the other is a version mismatch, not a fix.
• An identifier a chatbot supplied. If you cannot find the DOI yourself, a model's version of it is a guess wearing a uniform. A DOI that does not work covers the dead kind, and how to check if a citation is real covers the rest.

What APA, MLA and Chicago actually say

All three permit a reference with no DOI. They disagree about what goes in its place, and averaging them produces something none of them accepts, so the table below quotes each one.

Take one source and the disagreement is stark. A journal article you read in a library database that has no DOI ends, in APA, exactly where the page range ends, with nothing after it. In Chicago it ends with the database name. In MLA it ends with a permalink, and MLA says explicitly that a gated link is still acceptable. Three styles, three different endings, all correct. Use the one your department asks for and do not blend them, which is the same discipline that applies when you are citing an AI tool across the three styles.

Note what none of them contains: a fallback that involves supplying an identifier you did not find. APA's table of missing reference elements works the same way throughout. An unknown date becomes n.d., an unknown author moves the title forward. The style marks the gap or closes it. It never fills it with a substitute.

Which sources genuinely have none

Most lists of DOI-less source types were written before publishers finished depositing their back-files, and they have aged badly. These are the categories where a blank is still the normal outcome:

• Conference proceedings published outside the DOI system. Machine learning is the clearest case: papers in the Proceedings of Machine Learning Research, which publishes ICML and AISTATS, carry no CrossRef DOI. IEEE and ACM proceedings do, which is why the 2016 CVPR residual networks paper has 10.1109/cvpr.2016.90 and the ICML paper beside it on your reading list has nothing.
• Theses and dissertations, unless the university repository registers DataCite DOIs. A growing number do, so check the repository page before assuming.
• Newspaper and magazine articles, which have URLs rather than identifiers.
• Archival material, interviews, lecture slides, working papers on a personal site, and most ordinary web pages.
• Journals whose back-files were never deposited, which is where a lot of older humanities scholarship sits.

And three categories where the folk advice is simply wrong, so check before you leave the field empty. Book chapters in academic collections routinely have DOIs, and the MLA Style Center's own worked examples include chapter identifiers from Oxford Handbooks Online and Cambridge Core. Government and intergovernmental reports often have them too: the International Energy Agency's World Energy Outlook is registered under the OECD's 10.1787 prefix, and the UN Environment Programme's Emissions Gap Report under 10.18356. Preprints always have one.

One thing nobody knows, and you should distrust anyone who quotes it: the share of published scholarship that has a DOI. Every index large enough to count is itself assembled from the registries, so works without identifiers are missing from the count by construction, and any percentage you get describes the index rather than the literature. What you can establish is the status of one reference at a time, which is all your bibliography needs.

Writing the reference without one

Chicago's free quick guide gives a bibliography entry for a 1925 article in the Sociological Review that ends, in place of a link, with the single word EBSCOhost. It is a clean demonstration of the rule, and it is also this whole page in one line, because that article does have a DOI. Wiley deposited the journal's back-file, and 10.1111/j.1467-954x.1925.tb01552.x opens B. T. Hebert's "The Island of Bolsö" on the publisher's site today. The rule the example teaches is right. The example happens to be a case where the identifier turned up later, which is the ordinary situation for anything printed before 2000 and the reason step one is a title search rather than a look at the PDF.

When the search really does come back empty, the shapes are these:

• APA. The reference stops after the page range or the publisher. Nothing marks the absence, and nothing needs to.
• MLA. The location slot takes the next thing down the list: permalink, then URL. For a print source the page numbers already are the location.
• Chicago. The name of the database in place of the link, or nothing at all for a source you read in print.

Two practical notes. If your reference manager insists on a DOI field, leave it blank rather than typing a placeholder such as n/a, because a placeholder becomes visible text in the exported bibliography. And do not compensate for a missing DOI by padding the entry with an accession number, a database record ID or a shelf mark unless your style asks for one. Our citation generator renders the entry with the field empty in each of the eight styles it supports, which is also the fastest way to see what the ending is supposed to look like. If you are annotating as you go, how to write an annotated bibliography covers the note that sits under the entry.

A reference is a record, not a string

The anxiety behind this question is worth naming, because it is not really about a field. A reference list is a set of claims that you found and read particular things, and an empty slot feels like a hole in that claim. It is not. The identifier is a convenience for the reader, added by an industry that has existed since 2000, and scholarship worked without it for three centuries before that. An author, a title, a year and a venue locate a paper perfectly well.

What does break the claim is a reference assembled as text rather than taken from a record. Once an entry exists only as a formatted string, nothing in it can be checked without going and finding the paper again, and the string gets copied onward by everyone who quotes you. That is how a borrowed DOI outlives the person who borrowed it. The habit that prevents all of this is unremarkable: when you decide to use a source, open it, and take whatever identifier it has from the source itself rather than from the thing that handed you the reference.

That pairing is what CiteOwl is built around. It searches OpenAlex, CrossRef, Unpaywall and the web, reads the papers it retrieves, and writes each claim from a passage it keeps beside the sentence, so a reference in your list carries the identifier the index returned or no identifier at all, never one assembled to look right. Every edit arrives as a diff you accept or reject, which means the reference list you hand in is one you have seen.