It can make sense sometimes, but I think it can often be awkward. If the thing being cited is highly known (which will often be the case, if it is still being c

It can make sense sometimes, but I think it can often be awkward. If the thing being cited is highly known (which will often be the case, if it is still being cited despite the original document being lost in time) it will probably be cited in many secondary sources, and the citing author might very well not have learned it from any of them (but instead from a university course, a blog, etc.). Sure, one can randomly pick a textbook that they like and that cites that content, but this looks rather arbitrary to me. Personally it's not worth introducing that arbitrariness to solve this problem of "ghost" citations which is not especially common nor harmful. Maybe this is just a biased view... I'm from a country where they obsessively count metrics to evaluate researchers (not that I agree with that, but it's what they do), so being cited or not can be a big deal. I think the best solution might be to use the original citation but mark it somehow as inaccessible, that way it would be explicit that one is just citing for credit and did not really read the document. (I don't think I have seen this ever being done, though).