Models aren't trained based on what people search though, it's usually done beforehand with a large set of training data (chatGPT might also use the thumbs up/down though for training). It's more likely they're trying to avoid liability and controversy.