Do it by using both:

Content filtering (content-based filtering): a user rates games and/or adds games to their personal library or lists, and then is shown stuff that has similar metadata to the games that that user has rated highly and/or added to their library. For example, games within the same genres, games from the same developers, games released in the same year, etc.

Collaborative filtering (user-library-based filtering): a user rates games and/or adds games to their personal library, and then is shown stuff that other users who highly rated those same games and/or have those same games in their libraries also tend to rate highly and/or have in their libraries.

As far as I know, advanced recommendation engines use more complicated algorithms than the two over-simplified techniques I mentioned.