The export is going to look to the text layer (if there is one) of the PDF, so if there’s an issue with the text layer then you’ll get artifacts like this. For example, when a scanned document is OCRd there are frequently issues with the text layer. It’s easy to check in DEVONthink. Select any PDF whose kind is “PDF+Text” and then Data > Convert > to Rich Text. What you see in the resultant RTF is the best you can get from the OPML export or anything else that copies text from a PDF.
Check the script here for another approach.