I think it is more about an exploration-exploitation problem. I think you're focusing on the exploitation aspect too much. Algorithms like TikTok's do well with respect to exploration because there is a low cost for a bad recommendation. The quick content allows the algorithm to update quickly to the users current mood. This isn't going to be true for articles, where you are going to have spend more time and energy on the content being provided. This means that you have to more quickly model a person's mood/energy levels when they are using the app (you can determine this faster with quick content because you are iterating faster). Exploration is an important part of the overall reinforcement strategy because as humans our cost function is constantly changing, we do crave novelty (just in different levels). If you don't have the exploration side, and only have exploitation, then you just end up in a bubble (a bit can be good, but life needs spice). Worse, you'll end up bored.