Can I cite a paper I cannot get hold of?

The situation is common and the fix is usually administrative rather than intellectual. Your institution does not subscribe, the article costs more than a week of food, and the claim matters to your argument. Before treating this as a citation problem, treat it as a retrieval problem, because it almost always is.

Interlibrary loan is the underused option. Most doctoral students either do not know it exists or assume it is slow and difficult. At most universities it is a form, it is free to the requester, and articles arrive in days rather than weeks. There is usually a generous annual allowance and nobody checks whether you used it well.

Emailing the author works more often than people expect. Researchers are permitted to share their own work with individuals, most are glad to hear their paper is being read, and a two-line request gets a reply within a few days a surprising share of the time. Say who you are, name the paper, and say what you are working on.

The repository route is worth checking first because it takes seconds. Many publishers permit the accepted manuscript to be deposited, so the version of record is paywalled while a legally identical text sits in an institutional repository or on a preprint server.

Google Scholar is good at the job you need here and bad at the job you might mistake it for. Haddaway and colleagues found it useful for locating a known citation or a full text, but it missed 31.5% of the important evidence in their case study, so they recommend it as a supplement to traditional databases rather than a replacement (Haddaway et al., 2015, pp. 13-14). The underlying studies they review disagree on the size of the gap. Giustini and Boulos could not identify 5% of the studies from one systematic review through specific Scholar searches, while Boeker and colleagues found up to 34% missed across 14 reviews (Haddaway et al., 2015, p. 13). Treat a failed Scholar search as one negative result, not proof the paper is unobtainable, and go back to your library catalogue, the citation databases, and interlibrary loan.

What are the legitimate ways to get a copy?

Your library's interlibrary loan service, which is free to you at most institutions. The corresponding author, who is normally pleased to send it. Institutional repositories and preprint servers, which often hold an accepted version. And open-access finders that locate legal copies the publisher has permitted.

If you use a repository or preprint copy, check whether it is the accepted manuscript or an earlier draft, since pagination and occasionally content differ. When the text differs from the published version, cite the version you actually read and say which it was.

Conference proceedings and dissertations are a fifth route worth remembering. A finding published in a journal has often appeared first in a thesis or a proceedings paper that is freely available.

How do I cite it if I genuinely cannot read it?

Cite it as reported in the source where you encountered it, using your style's indirect citation format, or leave the claim out. Do not cite from an abstract as though you had read the article, since abstracts routinely omit the conditions that make a finding true.

The standard you are being held to is direct consultation. Penders argues that a reference exists so readers can retrace your steps and judge whether your use of the source is appropriate, which means citation frequency or an abstract is not a substitute for reading the content (Penders, 2018, p. 2). Penders also asks that each source receive credit for its exact contribution and that you make clear why and where you are citing it (Penders, 2018, p. 2). So do not write "Smith (2012) found" unless you have read Smith. Write that Jones reports Smith's finding, cite Jones, and let the reader see the route.

Wording is where an unread paper does the most damage. Penders notes that an argumentative citation should signal whether the source supports or opposes your claim, and that writing "established" where the authors only "suggested" inflates the contribution. Penders classifies the deliberate version as deception and the accidental version as careless writing (Penders, 2018, p. 3). You cannot calibrate that verb from an abstract, because the abstract is where the hedges were removed first.

Greenberg shows what happens when a field cites without checking. In a citation network built around a disputed claim about β-amyloid in inclusion body myositis, 218 papers produced 636 supportive citations against only 21 critical ones, and four reviews from a single research group carried 95% of all citation paths even though they reported no primary data (Greenberg, 2009, pp. 2, 4). That is not proof any individual unread paper is wrong. It is proof that repetition manufactures authority when nobody returns to the source, and your citation is one of the repetitions.

Abstracts are the specific trap here. They are freely available, they read like a summary of the findings, and they systematically drop the qualifications. A claim that holds for one subgroup under one condition appears in an abstract as a general result, and citing it from there imports an overstatement into your work.

If the claim is important enough to matter and the paper is genuinely unobtainable, say so in the text. "Reported by Alvarez as showing X, though the original could not be consulted" is honest, and it tells a reader exactly how much weight to place on it.