LLMs are designed to make plausible-looking text, so whatever bullshit statistics they extrude will be more convincing to the untrained eye.

Dice are more random, too random for believable results. The nice thing with an LLM is that it knows the real-world probability of every answer to a question, or can make a realistic inference of what that probability will be.

It won't give actual feedback about your thing, but it will give feedback that appears real.