I've had quite good luck asking Gemini and ChatGPT to include links to research papers for every claim they make. Not only can I review at least the abstracts but I find when I do this, they'll retract some of the hallucinations they've have made in prior messages. It almost seems (and maybe they do) in their web searching tools, reread the content they include. Thus, greatly reducing errors, with minimal extra effort on my part.