Check Citations

One pass, in the order of the list

A finished reference list is already a sequence. The first item is the first item, and the last item is the last item. Checking it should not turn that sequence into a search box you have to visit thirty times. Paste the list. The result is the same sequence, with a status on each row.

That is the whole job of this page. You are not asked to decide which line looks suspicious before you start. You are not asked to retype a title into a publisher site, then a second title into another. The list you already have is the input, and the list you get back is the input with a record beside it.

Order matters because a correction has to land on the right item. If the result were sorted by status, a mismatch on item 12 would be sitting next to other mismatches, and you would have to hunt for it in the manuscript. Here item 12 stays item 12. You can move down the file and down the result together.

The free run covers the first 30 items of that sequence. A paid report continues through the rest of the same list, up to 1,000, and keeps the order. The result says how many items were left. It does not drop them and pretend the list was short.

How to read the three rows in the sample

The sample on this page is a real shape of output, not a diagram of the product. Three rows, three different endings. Read them as you would read your own list: top to bottom.

The first row is a match. The DOI 10.1038/nature12373 resolves, the year in the list is 2013, and the venue is Nature. The title is Nanometre-scale thermometry in a living cell. Nothing on that row asks you to open the record, unless you want to. A verified row means the fields we could read agree with the Crossref record. It does not mean the sentence above the citation is fair to the paper.

The second row is the quieter failure. The DOI resolves, so the paper exists, and the year in the list does not. The sample uses Attention is all you need with the year 1999 against a record that says 2017. A year can be copied from a preprint, from a slide, or from an earlier version of the same work. The row does not delete the citation. It puts the year you wrote next to the year on the record so you can change one of them.

The third row has no Crossref record. The sample DOI 10.1234/not-a-real-doi is not registered. An empty lookup can be a made-up citation. It can also be a book, a dataset, a preprint that never received a Crossref DOI, or a real paper whose publisher has not registered it. The row stops at the empty lookup. It does not upgrade a miss into an accusation, and it does not offer a different paper to cite instead.

A line with a DOI, and a line without one

If a line contains a DOI, that DOI is the lookup. Crossref is asked for that exact record. The title, the year, the first author, and the venue in your line are then set beside the fields on the record. A DOI that does not resolve is the third kind of row, even if the title sounds plausible.

If a line has no DOI, the lookup is a search built from the title, the first author, and the year. The result shows the record that came back, including its DOI, so you can see whether the search landed on the paper you meant or on a neighbor with a similar title. A search is not an exact lookup. The row is there so you can reject a neighbor.

You can paste the list in the style you already used. A year in parentheses, a title after it, a DOI printed at the end, or a numbered line are enough for those two paths. A .bib file still works if you have one. Drop it under the box. This page does not ask you to convert the list before you can see a result.

The manuscript is not part of the input. A citation command inside a .tex file can be used to notice a key that the list does not contain, but the sentences are not read, and the file is not uploaded. The check runs in the browser. Crossref receives a DOI, or a short bibliographic query, not the paper.

What to do with each row before you submit

Work down the result in order, and only stop on the rows that are not verified. A verified row can stay as it is. A mismatch is a field to correct: usually the year, the venue, or a title that has drifted from the DOI it was pasted onto. Open the record, change the line, and leave the rest of the list alone.

A missing row is a decision, not an automatic deletion. If you have the book or the PDF in front of you, the citation can be real and still absent from Crossref. If you cannot find the work at all, that line should not stay in the list. The result will not make that call for you. It only tells you that Crossref did not.

When you are done, the screen is the check. A file you can attach or send is separate, and it is the same rows: status, the fields you wrote, and the fields on the record. The sample at the top of this page is what that comparison looks like before you paste anything of your own.

Where this pass stops

This pass does not search the literature for a claim you have not sourced. It does not draft a sentence, soften a paragraph, or replace a citation with a different one. If a row is empty, the next step is yours: find the work, or remove the line.

It also does not read the paper the citation points at. A verified row can sit under a sentence that misstates the result. Comparing a quotation with a PDF is a different job, and it is not this one.

Crossref is the only registry asked. Google Scholar is not queried. A work that exists only there, or only in a library catalogue, can come back as no record. That limit is on every row, which is why the result says so in plain language instead of hiding it under the status.

Use the other pages when the question is different. A single DOI you were handed, with no list around it, belongs on the source checker. A definition of what a citation checker is belongs on that page. This page is the list, in order, from the first item to the last item you included.