Bernard Williams begins an influential essay  by defining internalism about reasons as follows:
Internalism There is a reason for person A to φ only if A has some motive which would be furthered by his or her φ-ing.
Plenty of philosophers have found something intuitive about this idea, but there has also been no shortage of disagreement over the exact sense in which A must “have some motive” which φ-ing must further. In the introduction to a recent anthology of literature on internal reasons, Kieran Setiya  seems to think that the most attractive versions of internalism are those which satisfy the explanatory constraint. Bernard Williams gives it best:
EX If something can be a reason for action, then it can be someone’s reason for acting on a particular occasion, and then it would figure in an explanation of that action. (p. 106)
There are at least a few reasons for adhering to EX. You might think that there is a unified account of explanatory and normative reasons, and that EX is a link in that unification. You might think that what it is for A to have a motivation which would be furthered by A’s φ-ing is just for there to be some p such that A is disposed to make p A‘s motivating reason for φ-ing. If you’re inclined to believe either of these, you’ll probably think with Setiya that the broadest plausible version of internalism is IR:
IR The fact that p is a reason for A to φ only if A is capable of being moved to φ by the belief that p. (p. 4)
However, I don’t see how EX could possibly be true, as I don’t think it can overcome the kinds of cases which motivate so-called “advice models” of reasons. I argue that an agent can have a reason for action which, qua reason, could not possibly motivate them. This undercuts the motivation for thinking that IR is the correct way of understanding internalism.
First, some clarifications. (a) The reasons under consideration here are objective, holding in virtue of the facts and not in virtue of what one believes the facts to be. (b) As they stand EX and IR require interpretation. If I’m sufficiently irrational, any proposition or fact I can cognize has the potential to motivate me to do anything. Under that reading any proposition could be a reason for action, and the two requirements fail to rule anything out. I think a better way of capturing the intuition is IR`:
IR` If p is a reason for A to φ, then A is capable of being rationally motivated to φ by the belief that p.
Now we can see where the supposed explanatory constraint may lead: reasons are to be explained in terms of rational motivation. If we have a good handle on what rational motivation looks like, we’ll have an attractive account of reasons.
And now for the negative phase. The key to the argument is that one can have reasons to act which stem from one’s own ignorance and which cannot be cognized qua reasons.
Suppose I’m driving and, unbeknownst to me, I’m about to hit a patch of black ice. A curious fact about me is that I can’t handle black ice if it surprises me – I always freak out and swerve wildly. However if I’m aware that there will be black ice on the road ahead, I’m always able to collect my senses and navigate it just fine.
Let’s say that at t1 I will hit this ice if I go straight. But it’s now t0 and I have two options: go straight, or take a detour. I think the intuitive thing to say is that b, the fact that there’s black ice on the road straight ahead, is a reason for me to take the detour at t0 (ψ).
However if I were to come to believe b it would no longer be a reason, for I’d then have sufficient time to prepare myself for the coming black ice. As soon as I believe b it ceases to be a reason to ψ, and hence my being motivated to ψ by the belief that b would not be rational. So b is a reason for me to ψ, but it isn’t possible for me to be rationally motivated to ψ by the belief that b. And if it isn’t possible for me to be motivated in that way, then I’m not capable of being motivated in that way. So, IR` is false.
I think this kind of example provides stronger evidence for the advice model than those which have generally been advanced. Take Gary Watson’s example of the irrationally angry squash player: that an attempt to shake hands would most likely lead to an assault is a reason not to attempt a hand-shaking. Still, you might think that even an irrational squash player is capable of being rationally moved not shake hands by the realization of this fact, so this case does not count against IR or IR`. However things are worse than they seemed for those who favor the explanatory constraint, since some reasons derived from one’s own ignorance cannot be accomodated by the model, even though the advice model can accomodate them. (I haven’t shown this latter, but I hope it’s intuitive.)
 Williams, Bernard. (1979). “Internal and External Reasons.” In Rational Action, ed. Ross Harrison. Cambridge UP, pp. 17-28. Reprinted in Williams, Bernard. (1981). Moral Luck. Cambridge UP: pp. 101-113.
 Setiya, Kieran. (2012). “Introduction.” In Internal Reasons, ed. Kieran Setiya and Hille Pakkunainen. MIT Press, pp. 1-34.