As you note, it's going to be based on what the LLM knows about the user, and as any good predictor knows, rich people like to buy expensive stuff.

Whether or not the article has examples, it would be surprising (an LLM prediction failure) if this expected behavior did not extend to different prices for different users, e.g. recommending the rich guy the Whole Foods bananas and the poor guy the Walmart bananas.