It's not quite automated, but my starting point would be a (custom sql) search like this.
select t_object.ea_guid AS CLASSGUID, t_object.Object_Type AS CLASSTYPE, t_object.Name, t_object.Object_Type as BaseType, t_object.Stereotype, t_xref.Description
from t_object, t_xref
where t_object.ea_guid = t_xref.Client and t_xref.Name='Stereotypes'
That will return every stereotyped object in the model. It can be extended to return attributes/operations/connectors as well if needed.
If you drag the Description column to the group area, you can collapse the groups for stereotypes you actually want, then do something with the elements using stereotypes you don't want. Alternatively, you can use the filter bar to search for particular stereotyped objects. eg. filter the description column for 'GUID=', to find all of the stereotypes that don't come from a profile. Or filter the same column for Name=MyStereotype to find all usages of that stereotype to make sure they all come from the same profile.
You can even use Repository.CurrentSelection in a script to bulk process your selection within the search.