An empirical investigation of the consistency of retrieval through Internet
search engines is reported. Thirteen engines are evaluated: AltaVista, Eur
oFerret, Excite, HotBot, InfoSeek, Lycos, MSN, NorthernLight, Snap, WebCraw
ler and three national Dutch engines: Ilse, Search.nl and Vindex. The focus
is on a characteristic related to size: the degree of consistency to which
an engine retrieves documents. Does an engine always present the same rele
vant documents that are, or were, available in its database? We observed an
d identified three types of fluctuations in the result sets of several kind
s of searches, many of them significant. These should be taken into account
by users who apply an Internet search engine, for instance to retrieve as
many relevant documents as possible, or to retrieve a document that was alr
eady found in a previous search, or to perform scientometric/bibliometric m
easurements. The fluctuations should also be considered as a complication o
f other research on the behaviour and performance of Internet search engine
s. In conclusion: in view of the increasing importance of the Internet as a
publication/communication medium, the fluctuations in the result sets of I
nternet search engines can no longer be neglected.