A short list to finish before upload
Paste the reference list that will actually go in the file, not an earlier draft. For each line, keep it only if the verdict is real, or if you have fixed a “real with errors” line from the corrected citation. Remove “likely fabricated” lines. Do not submit “unconfirmed” lines that you have not opened yourself.
Read every retraction flag. A paper can match perfectly and still be the wrong thing to cite as support. The notice, when RefCheck has a DOI for it, is linked from Crossref (Retraction Watch), PubMed, or Europe PMC.
Then do the step the tool cannot do: for any sentence that reports a number, a causal claim, or a “studies show,” open the matched paper and confirm it says that. RefCheck stops at the bibliographic record.
Where invented references enter a manuscript
They enter when a chat tool drafts the related work, when a citation manager is fed a made-up DOI, and when a line is copied from another paper that was itself unchecked. The style of the journal, APA or IEEE or Vancouver, does not protect you. Run the checker on the list you will submit.
ICLR’s program chairs have described desk-rejecting submissions with confirmed references to documents that do not exist. That episode is written up under citation incidents. The practical lesson is to check before review, not after a desk reject.
Limits to tell a coauthor
RefCheck queries public scholarly databases: Crossref, OpenAlex, Semantic Scholar, and, by type, PubMed, Europe PMC, arXiv, DBLP, DataCite, Open Library, and Google Books. A Chinese-only journal article often stays unconfirmed and is linked out to CNKI, Wanfang, and Baidu Scholar. A missing record is not always a fake paper. A found record is not always the version you read.
3 free reference checks without an account, plus 5 more free after you sign in. Then 1 credit each. Failed lookups are refunded.