An unorganized territory is analogous to a census county division but not quite the same thing. The Census Bureau designates them in different states at the same summary level:
Unlike CDPs, which are named after populated places and somehow spatially relate to them, UTs and CCDs are purely statistical boundaries whose names are descriptive but don’t play a role in wayfinding, even by officials. If we do map these statistical boundaries, we’re really setting aside the on-the-ground rule, so there should be a strong reason for them. Unlike with Alaska’s census areas as county equvialents, I haven’t seen a very strong need for comprehensive national coverage of township equivalents, at least in California where I live, but maybe it’s different in other states?
Since CDPs have been tagged as boundary=census for a long time in OSM, geocoders such as Nominatim have probably grown accustomed to this tagging. A CDP lies inside of a CCD and is subordinate to a CCD in the census hierarchy. We should definitely give CCDs an explicit tag, such as border_type=census_county_division. Existing geocoders may need to adjust their heuristics to distinguish boundary=census relations by border_type=*. Setting admin_level=* may happen to help them make this distinction if they aren’t properly distinguishing between boundary=administrative and boundary=census, but that would be a bit of a hack.