In context of SO, garbage in garbage out (GIGO) applies. That is, many ChatGPT-generated answers look "almost" correct but actually they aren't. While many of SO answers which might became data model for the bot training are high-quality and correct, there are many other answers which are questionable (low-quality posts). Also, bots (ChatGPT included) don't take and understand surrounding context like humans do. So, without context, no matter how good something is, people with subject-matter expertise can (and will) see the fishy behind it.