If instead you "score" the popularity of a road based on "relative" popularity (i.e., correct for road popularity in a 5km radius), the automatic planner might avoid super busy urban areas!
I really like the popularity routing feature ("follow most popular" in desktop, I assume it's switched on by default in the mobile version). It allows me to get from point A to B using safe and nice roads.
I'm curious how this feature technically works, but I guess some of this is your companies secret, which is fine! I bet pieces of road are given a value given how many Strava users cycled on those roads (also visible in the "global heatmap"). A bunch of relatively short routes between point A and B are then compared based on the summed value of the pieces of road it uses. (I'm assuming it works this way!)
If I'm right, then I might have a nice suggestion. When I plan a long segment that crosses a couple of urban areas, sometimes it's much nicer to have a 10% longer route that actually goes around those urban areas (safer, more quiet, etc.). However, the "baseline popularity" of the roads in these urban areas will be relatively high; not because people like cycling there, but because they start and end their rides there.