Suppose, as evidentialists do, that the truth-conditions for claims about what you ought to believe are cashed out in terms of a particular believer's evidence. You'll likely think that:
(1) S ought to believe p if S has sufficient evidence for believing p.
Don't worry about what 'sufficient' means. With one or two exceptions, those who hold the view allow that there can be sufficient evidence for believing a proposition about some contingent matter of fact about the external world even if that proposition is false.
Now, supposing this is correct, we have a quick and dirty argument for evidentialism about absolutely everything normative. Let 'p' be a proposition that represents some action of the agent's as one that ought to be done. Let 'p', for example, denote the proposition that I ought to give to charity. While it seems from my point of view that what makes it the case that I ought to give to charity is (in part) that those in need would be helped by my charitable giving, that's not true. The truth-conditions for 'She ought to give to charity' should be cashed out in evidentialist terms. For, it seems that the following rule admits of no exceptions:
(2) S ought to believe S ought to X only if S ought to X.
Since S could have sufficient evidence for believing that she ought to X even if she did not if that fact about what she should do depended on more than just what supervenes on her evidence, we can argue from (1) and (2) that whether she ought to give to charity depends not on the needs of those she is in a position to help. It depends on what sort of evidence she has.
I think that this view is silly (to put it mildly). I think it shows what's silly about the appeal to the subject's perspective in arguments for internalism. But, that's me. I can't see how to be an evidentialist about epistemic 'ought's without being an evidentialist about them all. Since I think we have as good a reason to reject evidentialism as a thesis about all 'ought's as we have for rejecting any philosophical view, I'm happy to dispense with (1).