In this context, it means whatever the individuals answering the surveys think it means. The problem is that they won't all be answering based on the same definition (hence some of the confusing results).
There is a general issue with definitions, interstation's and socio-political implications of various terms and labels in this area (and others) and a lot of normal people are actually fairly ignorant of them (we're all weird outliers here
). That makes these kind of surveys difficult to accurately run and analyse.
I'm not sure it's possible to get anything beyond very general and imperfect impressions of what any given population actually believes. You can only really know on an individual basis, by asking a person to explain and describe in some detail. Before they're asked, many people probably don't even give the question a lot of thought and may well give significantly different answers depending on time, place and circumstances.