If one is a Christian deontologist, one thinks everyone ought to follow a certain set of rules, but one needn't prefer that - one might be rather pleased that only oneself will get into heaven by the following the rules. One might believe things, events or people are morally "good" or "bad" without preferring or preferring not that thing, event or person. For instance, one might think that a person is bad without preferring that person didn't exist. One can believe one ought to do something, without wanting to do it. This is seen very often in most people.

I'm talking about personal morals here, i.e. "what should I do", which are the only ones that matter for my own decision making. For my own actions, the theorem shows that there must be some utility function that captures my decision-making, or I am irrational in some way.

Even if preferences are distinct from morals, each will still be expressible by a utility function or fail some axiom.

And one can obviously have preferences which aren't morals. For instance, I can prefer to eat a chocolate now without thinking I ought to do so.

That example is one where the errors are so low that it doesn't make sense to spend time thinking about it. If you value your happiness and consider it good, then you ought to eat the chocolate, but it may represent so little utility that it uses more just to figure that out.

We should also be wary of equivocating on what we mean by "preferences". Revealed preference theory is very popular in economics, and it equates preferences with actions, which evidently stops us having preferences about anything we don't do, and thus means most of the usages of the word "preference" above are illegitimate. I think we normally mean some psychological state when we refer to a preference. For instance, I see the word used as "concious desire" pretty often.

When I say preference I mean "what state do you want the world to be in". The problem of akrasia is well known, and it means that our actions don't always express our preferences.

Preferences should be over outcomes, while actions are not. An imbalance can be akrasia, or the result of a misprediction.

Regardless of how you define preference, if it meets the axioms then it can be expressed as a utility function. So every form of preference corresponds to different utility functions, whether it's revealed, actual, or some other thing.