We can't, and shouldn't, trust media outlets to provide accurate diffs. But for text media it would be easy enough to poll them every X minutes and commit changes to an online git repo, accessible from the origin site as annotations.

It could be organized like a decentralized Wayback Machine, with volunteers tracking sites like people choose books in Fahrenheit 451. For major sites like NYT there could be multiple trackers, reducing the need for trust.

If that doesn't fit within fair use, maybe it should.