CiteOwl

Blog/Finding and reading sources

The paper I need is behind a paywall

The CiteOwl Team· Updated 24 August 2026· 14 min read

Before you pay, check whether the paper is already legally free, because for a recent one it usually is. When we checked 120 journal articles published since 2021 on 24 August 2026, Unpaywall found a legal free copy of 89 of them. Everything on this page is a legal route, every tool named here was tested the day it was written, and the first rung takes about ten seconds if you have the DOI. Work down the list in order, and file the interlibrary loan request before you start, because it costs nothing and it is the one rung that almost always ends in the actual paper.

It is late, you are most of the way through a bibliography, and one article stands between you and finishing. That is a bad moment to take advice from a page nobody has re-checked in four years, which is most of what is written about this. One tool the top-ranking guides still tell you to install shut down in November 2025. A result Google served us for this exact question points at a university subdomain that no longer resolves at all.

So here is the ladder, fastest first, with an honest note on how often each rung really works. Some of it will not work, and the page says so where that is true.

Check whether it is already free

The single fastest move is to ask whether a legal free copy already exists. Take the DOI and put it in your address bar: https://api.unpaywall.org/v2/ plus the DOI plus ?email= and your own address. You get back every legally deposited free copy Unpaywall knows about. We explain that API and its vocabulary in full in how to find peer-reviewed articles, so this page will not repeat it.

How often does it work? Nobody publishing a guide seems to have measured it, so we did. On 24 August 2026 we searched CrossRef for journal articles on eight topics, climate and urban heat, battery recycling, offshore wind, machine learning, ocean plastics, remote work, public transport and education outcomes, then ran every DOI through the Unpaywall API. Of 120 articles published in 2021 or later, 89 had a legal free copy, which is 74 percent.

Read that as an upper bound rather than a constant. Relevance ranking favours work that is well indexed, and well-indexed work skews open, so the true figure for a paper you were assigned is lower. The direction is the part that holds, and the direction is steep.

We ran the same test on older work. For articles published between 2005 and 2010, 18 of 64 were free, which is 28 percent. For 1995 to 2004, 10 of 64, or 16 percent. If your paper is from the last five years, the fee is frequently optional. If it is from 1998, expect to work for it. Those two situations deserve different amounts of your evening.

One result reframes the rest of this page. Of the 89 free recent copies, 77 were hosted at the publisher's own site and only 7 sat in a repository. For a recent article the free version is usually not hidden away in some archive, it is on the same domain that just asked you for money, on a different page. Be careful with one flavour of that: an article that is free to read on the publisher's site without an open licence attached is free by the publisher's ongoing choice, so download the PDF now rather than trusting the link to still work the week you are marked.

The reference figure from the literature is lower and older. Piwowar and colleagues, sampling 100,000 articles for PeerJ in February 2018, put the share of the whole scholarly record at at least 28 percent open access, and found that 47 percent of the articles Unpaywall users actually looked at were open. Both numbers point the same way as ours: what you are likely to want is more open than the record as a whole.

Unpaywall also ships a browser extension that answers the question on the page itself, and this needs stating precisely. The data and the API are current, and records we pulled on 24 August 2026 carried an update stamp three days old. The Firefox extension is a different thing: its last release was 28 July 2021, five years ago, despite 66,820 daily users. We could not read the Chrome listing, so treat its status as unknown.

What Google Scholar's all versions link is for

Unpaywall works from the DOI, which means it finds copies that have been properly registered against that DOI and misses the ones that have not. Google Scholar has the opposite failure mode. It crawls, so it picks up an author's own homepage, a PDF sitting on a lecture page, a conference version under a different title, and repository deposits with no identifier of their own.

Search the exact title in quotation marks. If there is a PDF link on the right of the result, that is the fastest thing on this page. If there is not, click All versions underneath the result, which is Google's own advice: it tells you to check out the alternative sources listed there. Scholar's access links cover, in its words, articles your library subscribes to, open access articles, free-to-read articles from publishers, preprints and articles in repositories.

Nobody has published a hit rate for this, and we are not going to invent one. What we can say is why it belongs on the ladder as its own rung rather than as a substitute for the first: it searches a different set, not a larger one, so a miss on rung one tells you nothing about rung two.

Your library, entered the right way

Here is the most common reason a student pays for something they already have. Library subscriptions are recognised by network address. Arrive at an article from a Google result and the publisher's server sees a home broadband connection and asks for a card. Arrive at the same article through the library, and it sees the university.

The mechanism is usually a proxy. EZproxy connects on your behalf with an authorised address so the publisher permits access, and OCLC says thousands of libraries in over 100 countries run it. Google's own help page tells you the same thing from the other side, that you may need to search from a computer on campus or configure your browser to use a library proxy. So start at your library's search box, or sign into the proxy first, or switch on the VPN, and then go back to the article.

Extensions do this automatically, and they are worth ten minutes tonight because they save the whole problem tomorrow. Since several tools in this category have quietly stopped being maintained, here is what we found when we queried the Firefox add-ons registry on 24 August 2026. These are Firefox figures and they do not transfer to Chrome.

The pattern worth noticing is that the actively maintained tools in this space are now the library-side ones, plus one small independent project. If you have a library, install whichever of the first two it supports. If you do not, Open Access Helper is the one still shipping.

Where the free copy actually lives

When the DOI lookup and Scholar both come up empty, go to the repositories directly and search the exact title rather than the identifier. Deposited copies frequently carry no DOI of their own, which is exactly why the first rung missed them.

Two real examples, both checked on the day this was written. A 2026 machine learning paper on interpretability in the Journal of Machine Learning for Modeling and Computing is paywalled at the publisher, and the authors' submitted version sits on arXiv. A 2024 Marine Pollution Bulletin paper modelling how plastic waste fragments is paywalled at Elsevier, and a free copy sits in HAL, the French national repository. Neither copy is advertised anywhere near the paywall.

Ask the people who published it

Publishers run their own legal free-access routes and almost nobody mentions them. Elsevier gives every author a Share Link that grants 50 days of free access with no sign-up, and says most of its journals offer one. Springer Nature's SharedIt gives authors shareable links to view-only versions across its whole owned portfolio plus a thousand partner journals. Authors post these on their profile pages and in announcement posts, so searching the exact title plus the author's surname often turns one up.

ResearchGate has a Request full-text button, and it is worth knowing exactly what it does before you rely on it. The request goes only to authors who are on ResearchGate and have claimed that publication. ResearchGate's own help says plainly that it does not provide or automatically supply copies of any content, that the timing is entirely up to the authors, and that your request may never be fulfilled. You also cannot cancel it once sent.

Treat a copy already sitting on ResearchGate with the same care. A study found that over half of the papers uploaded there appear to infringe copyright because the author posted the publisher's version rather than their own manuscript, and the platform settled litigation with Elsevier and the American Chemical Society in September 2023. The peer-reviewed paper documenting that finding is itself closed access, and when we queried the Unpaywall API on 24 August 2026 it returned no free copy of it anywhere.

Then email the author, and send it before you work the rest of the list rather than after. Nobody has published a reliable response rate for students asking an author for a PDF, so any guide quoting you one is guessing. The closest real evidence is a randomised trial in BMJ Open in January 2019 that emailed corresponding authors asking for study data: 78 percent responded, 27 percent within a day and 38 percent within a week. That was a request for data rather than a PDF, so do not carry the headline number across. The transferable finding is that the short email did just as well as the long one.

So keep it to one paragraph, from your university address:

Subject: Request for a copy of your paper on urban heat

Dear Dr Lindqvist,

I am an undergraduate at Leeds writing a paper on how cool-roof coatings perform in dense urban blocks. I would like to read your 2023 article, "Rooftop albedo and street-level heat in compact city centres," in Urban Climate, but my library does not have access to it.

If you are able to send a copy, I would be grateful. If your accepted manuscript is easier to share than the published version, that is completely fine.

Thank you for your time,
Ana Reyes, BA Geography, University of Leeds

Name the exact article, say what you are writing about in one clause, attach nothing, and do not apologise twice. Offering to take the accepted manuscript matters more than it looks: many publishing agreements let an author share that version freely and the final one not at all, so you are removing the reason to say no.

Interlibrary loan, which you should have started already

This is the rung the competing guides skip, and it is free. Your library will buy a copy of the article from another library on your behalf and email it to you, and at most universities enrolled students are not charged. UCSB states flatly that it does not charge currently enrolled students for interlibrary loans.

The turnaround is better than its reputation. Penn Libraries puts articles at two to five business days, and East Carolina says articles sometimes arrive within a few hours. Where the library runs RapidILL, the vendor reports a 95 percent fulfilment rate and an average turnaround of 11.2 hours across 1.5 million requests a year. Books are a different matter and will take one to two weeks, but a book is not your problem tonight.

None of that is guaranteed to land before your deadline, which is exactly why it goes first. It costs two minutes to file, it runs in the background while you work every other rung, and it is the only one on this page that reliably ends with the actual published article in your hands. If it arrives at nine tomorrow morning, you have your source. If it does not, you have lost nothing.

The version you found may not be the version you cite

Most of the free copies above are not the published article. There are three versions of any paper and they are separate objects: the submitted version, which is the preprint before review, the accepted version, which has been through review but not through the publisher's typesetting, and the published version of record. Repository deposits are usually one of the first two.

This matters less than people fear and in a different way than they expect. Klein and colleagues compared preprints against their published counterparts and found that the text generally changed very little between the two. The risk is not the prose, it is the numbers. A sample size, a coefficient or a hedge on a conclusion can change in review, and quoting the preprint's version of a claim as though it came from the published paper is a real error rather than a technicality.

So cite what you actually read. If you read the arXiv copy, cite it as a preprint with its repository and identifier, and if you can see the published abstract, check that the specific sentence you are leaning on survived into it. Presenting a preprint as the published article is the one move here that can actually cost you marks, and it is entirely avoidable by labelling it.

And when nothing on this page works, that is a real outcome rather than a failure of technique. Roughly a quarter of recent papers and most pre-2005 papers are genuinely closed, and no legal route opens them tonight. You have two honest options. Swap the source for one that says something close enough and that you can actually open, which is usually easy once you go back to the search with a clear head. Or cite the preprint openly and say what version you read. What you should not do is cite a paper from its abstract and hope, because the claim you need is exactly the kind that lives in the part you could not see.

This is the shape of the problem CiteOwl is built around. It searches the same open academic infrastructure described above, reads the papers it can actually retrieve, and cites only sources it has genuinely opened, with the supporting quote attached to each one. A source it could not read does not quietly become a citation. You still decide what belongs in your argument. What goes away is the reference that looks solid because nobody could check it.