Well, the question-asker should be able to tell if ChatGPT's answer is incorrect. Most obviously, for software, they try it and it doesn't work, or for scientific questions, they look up the references and they don't exist. It is probably even possible to automate such checking to some extent. But similar issues apply to checking human answers - at some point, the answer is correct because it sounds plausible, not because it has been verified.
If people are posting answers that they don't know are correct, putting the burden on the asker to do that vetting, then the entire point of the site is moot.
News flash, but a lot of stackoverflow answers are garbage - incorrect, out of date, etc.. Does this make it useless? No, you just have to sift through.
