I think you are right that ChatGPT and other LLMs can be useful in prompting us (provided we prompt it well first) to consider what we might otherwise have neglected, without making or even recommending a decision. But even its own selection of what to bring to our attention is subjective, and may not map onto our own value set and preferences. Now this is the same as if another person would do so - but we will not normally blindly trust another person (unless they have earned that trust and we know them really well). As long as we treat AI as a stranger who does not know us and who has not demonstrated that they are competent and trustworthy, and take its output in that vein, it can be helpful. If we stray, it might not.