Although we have a blacklist for the major legacy publishers and a check to ignore publisher locations from Unpaywall, OAbot suggestions still contain things like:
327 http://www.biodiversitylibrary.org 298 http://www.app.pan.pl 259 http://www.scielo.br
Adding such URLs is not a mistake but can be considered pointless, if they go to the same place as the DOI. Citation bot users have been busy removing such redundant URLs and we should not reverse them. Given the long queue of suggestions available, it's best to focus the tool's users' time on those with most added value.
A simple solution might be to ignore the PDF URLs associated to a doi.org URL on Dissemin paper records. A better solution might be to check whether the URL really means "the DOI is gold/hybrid/bronze OA" and if so just set |doi-access=free, avoid touching |url=. This is the logical continuation of T228632 and might be considered a return to the original functionality.