It's a tool, not a source of empirically valid data. No one's going to 'get used' to LLMs being cited as sources of factual information any more than they'd complacently accept someone's Monte Carlo simulation or a hypothetical thought experiment being cited for the same purpose.