Paste a bibliography, a brief, or any text with references. Every one is checked against real databases: not a model. You'll see which references don't exist, which exist but don't match what you claimed, and which have been retracted.
Searches Crossref · OpenAlex · Semantic Scholar · PubMed · DataCite · Google Books · OpenLibrary · Google Scholar has no public API, so it's offered as a one-click search instead.
Verified, a real record matched on title, an author and the year.
Mismatch, the work exists, but your citation has the wrong year, wrong authors, or a mangled title. Common when a model half-remembers a real paper.
Not found, nothing close in any database we searched. This is what a fabricated citation looks like, but it is not proof. Government reports, local journals, older books, non-English work and brand-new preprints are legitimately missing from these indexes. Use the search buttons before you accuse anyone of anything.
Retracted, the work exists and is real, and has been withdrawn. Citing it as valid is usually worse than citing something fake.
Couldn't check, every database failed or rate-limited. This is deliberately not "not found". A network problem must never look like a fabrication.
Mimir checks that a source exists and matches its metadata. It cannot check that the source supports the claim you attached to it. That still needs you to read it.