Well, that’s my experience anyway. So far out of the three publishers I tried:

InderScience was readable after creating my own PRISM module for the Rome parser to be able to get at some of the information. However this does mean I can now create complete referencing for these items for export into things like EndNote.

ScienceDirect was correct in its syntax but only included the names of the paper authors in the <description> rather than as <author>, so making it impossible to recreate any sort of citation.

The final one, IEEE Explore, did have author information, but used lots of invalid tags. Didn’t they even try to use feedvalidator.org? Look at the results!

Not impressive, is it? I guess this is the sort of nonsense Santy and co are having to cope with over at TicTocs. (API please?)