My guess is there just isn’t enough content covering this in the training data. And then there’s the reinforcement learning step, which is editorial to a large degree. The latter is influenced by the former. So maybe we just need to get out more and power past the rookie numbers as far as crawlable content on these topics?