Adobe should have bought and integrated AI like pdf.ai or unriddle.ai. Pdf.ai ocr is very good at converting old fomat pdfs. Both are good at analysis 'on a good day'. I have had AI responses ranging from sublime to poor in both cases. Possibly because they couldn't afford 'precise' analysis on their servers for unpaying customers.

But inbuilt and with the wealth of Adobe they could have moved them on in leaps and bounds.

For example the refs are to pdf page numbers not the physical page numbers you see. Most documents don't count the cover as page 1. Presently using Harvard (or similar) referencing with seems to be out of scope. Unless analysis can fit with the system used by a journal or university it will never be easy to generate formal papers with AI. An amazing niche market for Adobe to exploit and make pdf still worthwhile.