Yes, people often use the substrate PNP (because it happens
to be in the right electrical orientation for a ground-referred
bandgap design.
In standard Psub/Nwell CMOS, the P+/Nwell (which you want
for a groiund-referred design) is really a substrate PNP, no
getting around that (absent SOI or deep Nwell with enough
of a buried layer or retrograde doping that the PNP base is
effectively dead).
What Dick-freebird said about substrate BJTs as bandgaps appears in the one called "Bandgap reference" on pages/slides 15 and 17 (17 is the parasitic PNP one), it has explanatory diagrams so I was able to understand it a bit.
MT-087 shows diferent ways of making them, not all were BJT-based.