I'm pretty sure of it, and I think that Google uses Location Services on android phones to identify bike trips and flag new routes for human reviews when the "bot" sees some threshold of instances.

Around '09 or '10 I lived in an apartment complex, which was isolated except for a single road out. Way in the back, where I lived, I had a short-cut climbing up a bank, through about 30 feet of brush/forest that I eventually cleared out, to an access road leading directly a parking lot to the Greenway. After a few months my shortcut appeared as a bike route on Google Maps! I was the only one there who got around by bike (because of the killer hill on the exit road I think), so it's unlikely that some resident added it in manually as a suggestion. I am convinced that Google saw my phone moving across, day after day on obvious bike rides, and added it on that basis. I'm also pretty confident that Google has refined their algorithms since then, and bike routing is now more dynamic based on utilization.

They also know it's just you. A way used by one person isn't likely to be considered as public.

They likely have people review the data that they require. That is, there no reason to expect that it's all automated.

Google may get ideas of ways from that sort of data but the data they got initially wasn't created that way.

My "bad" is less about preferences and more about giving a clear indication of the route that was picked.

To illustrate, let me start with an example that predates Google Maps. I was cycle touring through Western Kansas and saw parts of "old-40" that paralleled I-70. It seemed like a reasonable choice, though my paper maps disagreed on how much of the route was gravel. I (mistakenly) assumed that roads generally went from being unpaved to getting paved and hence assumed that the map that showed ~8 miles of gravel was more likely correct.

Turns out, Grove and Trego counties had found maintenance of the old pavement was too much trouble and hence had torn up old US 40 in their counties and converted it back to a gravel road. Rather than 8 miles of gravel, I ended up with ~18 miles of good hard-packed gravel and ~12 miles of soft sandy gravel road. If I had known I might have made a different choice.

Similar idea with Google Maps bike instructions. I don't necessarily need it to avoid gravel roads, but instead give a clearer indication of the routes that it has picked. A similar example from my current travels: go from "Heroica Mulege, Mexico to Ciudad Constitution, Mexico". Auto route is ~282km of paved roads. Bike route is ~274km of which I'm guessing much less than half is paved and water re-supply points are very limited.

Last week I cycled the auto route in four not too difficult days of riding (18.5 hours riding according to Strava). I expect the off-road riding through Baja would have been much tougher and certainly longer than the "15h 25m" that Google Maps bike route gives as a time estimate. So the "bad" isn't that there may not be people with preferences to go bikepack back roads in Baja, but that Google Maps bike instructions doesn't give clear indications what parts are bike touring and what parts are bikepacking and it can make a big difference.

My biggest beef is that there is not enough contrast between the road and forest.