This depends on what navigation software you’re using, but I have some experience editing the Waze map. The way it works on Waze is that your phone sends the server your desired start and end points, and the server responds with a list of all the intersections you need to traverse in order. (This is actually a series of road segment junctions, wherever the map editors joined two road pieces together). These intersections can contain metadata on how to announce specific turns, but generally don’t because there’s an algorithm that looks at the angle the segments meet at and automatically decides how to describe the turn. The places I’ve seen it manually overridden include intersections where two divided highways meet at an angle far enough from 90° that it gets confused about how to announce a left vs a u-turn. I’ve also seen forks in the road where the side road requires less of a turn than continuing on the main road and the algorithm gives ambiguous instructions, like “continue straight” meaning turn onto the side road.
Edit: On your point about non-visually noticeable “blips”. This is also pretty common when roads change width right at an intersection (e.g. adding turn lanes). The Waze map doesn’t include road width in its data, so editors usually draw it down the centerline of the road. If the road changes width suddenly, you have to choose between keeping the line straight-ish, or faithfully following the centerline, which can mean that if you were to zoom way in there can be weird jumps and sharp angles that get smoothed out by the visual renderer
The cars already have decent GPUs to process the camera data for driving assistance features, so someone at the company probably just thought it would be neat to do something with that computing power when it’s not being used for driving.