@fell that comes down to the system prompt really, loosely guided you'll receive ANYTHING in relation to the tokens you've passed to it, not necessarily correct information. It's important to remember the model you're dealing with, it's training set (for a lot of the general english ones it's based off commoncrawl primarily including reddit posts, updated or live info has to come in from data sources and how that's configured is a bit of a black box if your'e using one of the online services.
@raptor85 Maybe it's the way I prompt them. I usually tell them upfront what I know and what I think and then ask a specific question.
6 out of 10 times they just parrot what my assumption was.