Mark Iron drew my attention to this story: https://archive.is/N6Cfa (which cites RetractionWatch )

Briefly, there's a reference that is getting over 400 citations:

Van der Geer, J., Hanraads, J.A.J., Lupton, R.A., 2010. The art of writing a scientific article. J Sci. Commun. 163 (2) 51-59.

Except, if you go looking for it... there is no such paper. Management professor Anne-Will Harzing (author of the Publish Or Perish scientiometric software) explains on her blog that she tracked down the original to an Elsevier "Instructions for Authors" document, where the desired reference format was illustrated by a dummy reference.

Once enough low-grade papers had cited this dummy reference that it started showing up in citation databases, it took on a life of its own.

Moral of the story:

(a) Many authors cite papers they have never read: they may enter some keywords in Web of Science or Google Scholar, and copy-paste a reference that seems relevant without ever bothering to check it.

(b) If you are going to use a dummy reference to illustrate a citation format, it is best to either use an existing paper or make up an obviously tongue-in-cheek reference, such as:

Neill, S., Moore, J., Spielberg, S. et al., 2049. Mating behavior of velociraptors in captivity. J. Dinosaur Reprod. 12 (3) 45-67.

At any rate: at an increasing number of publishers, the production staff will (have their computers) run your reference list through crossref.org for DOI lookup, and flag any references they cannot find a match for on the page proofs. This will probably put an end to this problem, or at least greatly mitigate it.