Can I cite a preprint?
Usually yes. Most departments will take a preprint as long as you say it is one, and in several fields the preprint server is where the literature lives. Two conditions get skipped by nearly every guide on this question. Say what the thing is, in the reference, not just in your head. And check whether it has been published since you found it, because if it has, that version is the one to cite and the two are not always the same paper. The check takes two minutes, and the preprint's own page will often not do it for you: of the 8,145 EarthArXiv preprints registered with CrossRef, 1,539 link a published version, and for SSRN it is 2,118 out of more than 1.2 million.
- The short answer, and the two conditions
- What posting a preprint actually involves
- The two minute check for a published version
- Why the preprint's own page usually will not tell you
- The reference, in APA, MLA and Chicago
- Withdrawn preprints and the file that keeps downloading
- Whether your marker will accept it
A preprint is a manuscript its authors posted publicly before, or instead of, a journal accepting it. arXiv, SSRN, OSF Preprints and EarthArXiv all host them. The question feels risky because a preprint looks exactly like a paper: title page, abstract, figures, reference list, often a DOI. Nothing on the PDF says which side of peer review it sits on.
The short answer, and the two conditions
Cite it, and make the reference say what it is. Every major style has a slot for this, and using that slot turns a preprint citation from a risk into an ordinary one. A marker who sees a preprint labelled as a preprint reads a writer who knows what they are holding. A marker who finds an unlabelled preprint dressed as a journal article reads something else.
The second condition is the one almost nobody states. Before you cite it, check whether the same work has since appeared in a journal or in conference proceedings. If it has, that is the version of record, and it is what a reader will find when they follow your reference. The two documents can differ, and rarely in the prose: Martin Klein and colleagues compared preprints with their published counterparts using standard text similarity measures and reported that the text contents generally changed very little. What moves in review is smaller and more dangerous to quote. A number, a sample size, a hedge on a conclusion, a figure that was redrawn. Attaching the preprint's version of a result to the published paper's authority is a real error, and it is invisible to every check that only asks whether the source exists.
Two conditions, both cheap. Say it is a preprint in the reference. Search the exact title first, in case the published version now exists.
What posting a preprint actually involves
Preprints are not peer reviewed. Servers apply screening instead, and it is worth knowing how far that goes, because the gap between screening and refereeing is the whole basis of a marker's objection.
arXiv is the clearest about it. Its content moderation policy says in one sentence: "Please note that the arXiv moderation process is not a peer-review process. arXiv staff and moderators cannot give feedback on the submission." The moderators are real experts, described on the same page as "volunteer subject matter experts with terminal degrees in their field", and they do refuse things. A submission "may be declined if the moderators determine it lacks originality, novelty, significance, and/or contains falsified, plagiarized content or serious misrepresentations of data, affiliation, or content".
So the filter is genuine and it is not review. Nobody has checked the derivation, re-run the analysis or argued with the conclusion. That is the honest thing to tell a reader, and the reason the label matters more than the format.
One complication cuts the other way. Some things posted to preprint servers have already been through review. Authors routinely deposit the accepted manuscript, the version a journal agreed to publish before typesetting, and it sits on the server looking identical to a submitted draft. APA acknowledges this in its reference guidance: "Preprint versions of articles may or may not be peer reviewed or may be the author's final, peer-reviewed manuscript as accepted for publication." You cannot read a server's name and conclude anything about review status. How to find peer-reviewed articles covers the three versions and where each comes from.
The two minute check for a published version
Here is the routine, in the order that fails fastest.
- Copy the exact title. Take it from the preprint's page, not from someone else's citation, and keep the punctuation.
- Search it on CrossRef. Paste it into search.crossref.org, or use the API at
https://api.crossref.org/works?query.bibliographic=followed by the title. CrossRef holds the metadata publishers deposit, so a journal article that exists is almost always in there. - Look for the same title under a different DOI. A published version appears as a journal article or a proceedings paper with a venue name attached and a DOI that is not the preprint's. Titles change in review sometimes, so also try the first author's surname plus three distinctive words from the title.
- Match the authors before you believe it. Standard phrasings repeat, and a same-title hit with a different author list is a different paper. Same discipline that keeps you out of the trap in the DOI links to the wrong paper.
- If you find one, read it. Not all of it. Find the specific sentence, number or figure you were going to cite, and confirm it survived into the published version in the form you were going to quote. That is the step the whole exercise exists for.
If the published version is paywalled and the preprint is not, you are allowed to read the preprint and say so. Cite what you read, note that the published version exists, and quote only what you can see. The paper I need is behind a paywall works through the legitimate routes to the full text. If nothing turns up in CrossRef at all, the preprint is very likely still just a preprint, and you cite it as one.
Do not skip the check because the preprint is recent. Journals are slow and preprint pages are permanent, so a paper posted last year may be in print now and its page will look exactly as it did the day it went up.
Why the preprint's own page usually will not tell you
The obvious objection is that the server should just tell you. Some records do. A preprint deposited with CrossRef can carry a relation of type is-preprint-of pointing at the published article's DOI. We counted how often that link is actually there, from CrossRef's public metadata, on 29 August 2026.
| Preprint server | Preprint records in CrossRef | Records linking a published version |
|---|---|---|
| EarthArXiv | 8,145 | 1,539 (18.9%) |
| OSF Preprints | 105,273 | 7,977 (7.6%) |
| SSRN | 1,220,730 | 2,118 (0.2%) |
That is easy to over-read. The counts do not say 81% of EarthArXiv preprints were never published. They say the link is missing, which is a different and more useful fact: the absence of a published-version link tells you nothing at all. Nobody has measured how many unlinked preprints did reach a journal, and we are not going to guess at it.
arXiv is not in the table, because it registers its DOIs through DataCite rather than CrossRef and the published-version link is not part of that record. On arXiv the information lives in a journal reference field the author fills in by hand, and arXiv's own help page for it is candid about how optional that is: "When a article is published, the author may wish to indicate this in the abstract listing for the article." Adding one does not even create a new version of the paper. arXiv also accepts automated feeds of the same metadata from publishers, which fills the field in without the author lifting a finger, but only for publishers who send a feed.
An example, and a fair one, since the paper is about this exact problem. Klein, Broadwell, Farb and Grappone posted "Comparing Published Scientific Journal Articles to Their Pre-print Versions" to arXiv in April 2016. The record still has no journal reference and no DOI field. Its comment line says the paper was "accepted to be published at JCDL 2016", which is a promise, not a citation. Run the title through CrossRef and two published versions come back: a 2016 conference paper in the Proceedings of the 16th ACM/IEEE-CS on Joint Conference on Digital Libraries, and a 2018 journal article in the International Journal on Digital Libraries, volume 20, issue 4, pages 335 to 350. The study most often cited for the claim that preprints barely differ from published papers is itself a preprint whose page never mentions that it was published, twice.
The reference, in APA, MLA and Chicago
All three styles want the same two things: the repository standing where the journal name would go, and something in the entry saying what the document is. They put the label in different places.
| Style | What the rule says | Where it lives |
|---|---|---|
| APA 7 | Author, year, title in italics, the repository in place of the journal, then the DOI or URL. "Ideally, use and cite the final, published version of a work. However, if you used the preprint version of a work, cite that version." | Preprint article references, and Publication Manual section 10.8 |
| MLA 9 | Standard entry, with the status added as a supplemental element: "You should either indicate that the source is a working paper in your prose or use the optional-element slot at the end of the entry to do so." The ninth edition calls those slots supplemental elements, and one applying to the whole entry goes at the end. | MLA Style Center on working papers and supplemental elements |
| Chicago 18 | Covered in a section of its own, titled "Journal article preprints". We could not read it. The section text sits behind the CMOS Online subscription, and we are not going to paraphrase a rule we have not seen. Open it through your library. | CMOS 18, 14.76, with ahead-of-print and forthcoming articles at 14.75 |
Two notes. MLA's page on working papers is written against the eighth edition and says so at the top, but the mechanism survived into the ninth under a new name, and MLA's own example ends with the words "Working paper." in the final slot. The Chicago row is the honest gap: the Press publishes its chapter contents openly, which is how we know 14.76 exists and what it is called, and sells the text. Every free page telling you Chicago's preprint format is either quoting a subscription it holds or repeating another blog.
Applied to the paper from the last section, the APA entry reads: Klein, M., Broadwell, P., Farb, S. E., & Grappone, T. (2016). Comparing published scientific journal articles to their pre-print versions. arXiv. https://doi.org/10.48550/arXiv.1604.05363. The MLA entry reads: Klein, Martin, et al. "Comparing Published Scientific Journal Articles to Their Pre-print Versions." arXiv, 18 Apr. 2016, doi.org/10.48550/arXiv.1604.05363. Preprint. Either way a reader can tell what they are being sent to before they click. Had you run the check first, you would be citing the 2018 journal article instead.
Withdrawn preprints and the file that keeps downloading
A preprint can be withdrawn, and the withdrawal is quieter than a retraction. arXiv's withdrawal policy explains the mechanism: "Articles that have been announced and made public cannot be completely removed. A withdrawal creates a new version of the paper marked as withdrawn. That new version displays the reason for the withdrawal and does not link directly to the full text. Previous versions will still be accessible, including the full text."
Follow that through. The notice lands on the current version of the abstract page. The PDF you downloaded three months ago is unchanged, and the direct link to version 1 still serves the full text with nothing attached. If you keep papers in a folder and cite from the folder, a withdrawn preprint looks exactly like a live one forever. Reopen the abstract page, not the file, before you submit.
The same policy explains why so many published papers still have live preprints sitting under them: "It is not appropriate to withdraw a paper because it is published or submitted to a journal. Instead you could submit a journal-ref." Publication is not a reason to take a preprint down. That is correct as archiving, and it is precisely why the burden of noticing falls on you.
Withdrawal and retraction are different events with the same consequence for your reference list, and the routine for the published side is in is the paper I cited retracted. Run both in the same pass, since you are already looking at the record.
Whether your marker will accept it
Everything above is about doing it correctly. Whether you are allowed to do it at all is a separate question with no general answer.
In high energy physics, mathematics and machine learning, arXiv is where people read, and a reference list without preprints would look out of date rather than rigorous. In a first-year essay marked against a rubric that says "peer-reviewed sources", a preprint is an unrefereed source and will be treated as one, however carefully you labelled it. Both positions are reasonable and they do not reconcile, so do not average them into advice.
Read the brief first. It usually settles the question in one line, either by naming peer review as a requirement or by setting a source count without qualifying the type, and the second case is the common one. What counts as an acceptable source in general is in how to find credible sources.
If the brief is silent, ask, in a form that can be answered in one word. Send this: "For this assignment, is a preprint acceptable as a source if I label it as a preprint in the reference, or do you want only peer-reviewed work?" That sentence names the thing, names the safeguard, and offers a binary. It is also, quietly, evidence that you know the difference, which is why it tends to get a yes.
Nobody has measured how often preprint citations are marked down, or how many departments hold a written position at all. We looked. What exists is anecdote and habit, so the question above is not a formality, it is the only reliable instrument available.
When you do cite one, treat it like any other source in one respect: confirm it exists, in the record, before it goes in the list. How to check if a citation is real is the general version of that routine, and our DOI lookup shows the record behind an identifier in one paste.
This is the kind of work CiteOwl is built to have already done. The agent 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 what lands in your draft is the record the index returned rather than a plausible-looking string. Every edit arrives as a diff you accept or reject, which means you can see which version of a paper a claim came from before you agree to it.