During my observation of the recommendation systems, I’ve somehow found them to have a completely different impact, depending on the sphere I am moving in, and applicability of the, for example, item I am trying to buy, or look for.
In the case where I am looking for an air pump, I keep getting recommendations to buy an air mattress, because those are the items most often bought together. In my particular case, it totally misses the point, since I am buying an air pump to interact with balloons.
But in the situation where I am buying a book, recommendation on the books that usually go together with the one I am buying have a much more precision to it, of course considering also various applications of the certain content of a certain book, but yet the chances are much higher that the recommendations actually match  your interest.
There is also a compromise I think that you have to be ready to make in order to actually make the recommendations system to work, and I think that social networks have a great part in that, by enabling the insight into your own interests communicated more precisely if you yourself state them then if algorithm tries to track them trough your movement on web, I guess at the end it comes to the amount of engagement you want to have in order to make this system work.
I have also found that recommendation systems which work on the basis of collaborative filtering are much more precise, although it does depend on the as I said area you are moving trough and applicability of the same.
I rarely almost never agree to take part in processes of explicit data collection,since it is time consuming, and somehow makes me think that machine will anyway never know what is best for me.

I’ve also found that recommendation systems are of much use actually when you are looking trough some things or areas you are not familiar with since that you have reference points of navigation, trough the materia, or possible things that can be connected and whose connections perhaps you wouldn’t be able to see, in which case you are using other peoples experience directly, but on the other hand I can see how this can also limit your movement in a certain way or even become anoying if the system is severely wrong.

The best reccommendations of course I’ve found to be the ones that are directly derived from the point of my interest, if we take for example certain topic I am researching or certain article, in which case recommendations based on the actual reference list for that article, somehow end up being the most precious for me, which then somehow excludes the whole algorithm tactic of recommendations, returning it practically to analog indexing principle.

The thing that is for sure universal for all recommenadation systems is tat it doesn’t actually allow individual approach to the subject, trying to generalize interests, and possible directions based on the information derived from the mase, practically.
As an orientation points, in unknown field I think it works very well but in a field with much more specificity, or precision there is not much weight to it actually.
I’ve found that recommendation systems work the best when combined with couple of tehniqes of collecting data , from the side of the user as well as from the side of the.
For example when I am looking for certain categories of experimental videos on Vimeo, recommenations based on the most popular videos seen align with the one I was already looking for , doesn’t actually give ma good enough source, because there is a big possibility that I will find a video which doesn’t have that big ratings from the side of the users, or viewers, which can in a certain case indicate totally opposite, that that video is not easy to look at or like with thumbs up which comes as a positive information in this kind of search, when I guess searching trough taggs self-referenced from the side of author are the most precious data sources.