Interestingly, research shows LLMs are better than humans at persuading people, and one of the reported reasons is that their arguments are deemed to be more evidence-based. (Findings even show that people are more likely to be persuaded when they KNOW the other side is an AI... maybe the underlying dynamic is that they assume people have vested interests?)

So maybe evidence does help, but what detracts from its effectiveness is how it is presented. Like we see every day on various public forums including this one, arguments are made in a rather heated manner, which are probably not conducive to making the other side more receptive to ideas. I'm not sure if this might be the confounding in these previous studies.

LLMs have the advantage of infinite patience and RLHF'd dispassionate mannerisms and the presumption of no vested interests. But that also means your friend could be technically right.