I’d also be interested in seeing the rate of hallucinated citations from the pre-ChatGPT era.

I’m not entirely convinced this is purely an LLM-related issue. I’ve definitely come across countless misattributed citations in big four reports long before generative AI became widespread.