I (conditionally) second Anders' complaint. I've only used it once, just out of curiosity, to see what would happen. I asked it about myself and my published works. And while it got a few things right, it got a number of things spectacularly (actually, inconceivably) wrong. It sure looked like it was making stuff up, although I'm sure there was some reason behind the scenes why it spit out the garbage it spit. You can read my (humorous, I hope) account of the experiment at https://medium.com/the-haven/chatgpt-is-a-rotten-liar-9f96cfef32bb.

Based on this--and the fact that other people have reported similar experiences--I wouldn't trust it for research. (Although, I suppose there may be domains in which it performs better). Since I can't fully trust it, I'd have to fact-check everything it said using reliable sources. If I have to do that, why not just go to the reliable sources in the first place?