I don't rely on GPT output as canonical, I only use it to help surface big rocks I can refine with other engines and find primary sources. It's fancy autocomplete after all, but PageRank was just weighting links in a link web as well.
It frustrates me to no end... and then I go back for more punishment naively believing this time it'll actually be right.
With that said, the newer voice interactive version (using the app), I've found to be pretty slick for debating yourself on technical merits of designs/architecture, etc.