Sorry but about 4 weeks ago, I did some testing for my ex-employers since I have time on my hands after retirement. Contrary to the puffery coming from the AI companies, I found AI while lacking in accuracy in the tests I ran was more than willing to feed me garbage. Anything that gives ~40% accuracy is not ready for prime time. The testing was on student papers where the AI generated citations were often garbage (total garbage, pointing to the wrong links, pointing to information that had nothing to do with the student paper, etc.). The policy now is that using AI to create citations is a non-starter and automatically results in a 0 on the paper if any inaccurate citations are found. The students can still use AI to generate the citations but if they don't check each one manually for accuracy ... life can be tough.
Oddly, this article
AI Search Has A Citation Problem seems to show pretty much the same class of results as I saw.