I'm honestly curious. Think of a fact, and then ask yourself why you know it. Out of 5 attempts how many did you actually have no idea why that fact is there.
I would expect if I were to ask people why do you think daffodil flowers need lots of water they would at least say something like, oh I heard it somewhere (assuming that the do indeed believe this). From this I would choose to shift my belief only very very slightly.
It's worth being cautious here: just because a brain can generate an answer to a question doesn't mean that it was actually storing that information. "I heard it somewhere" may just be the default response when no supporting evidence can be found.
But your examples here are valid - sometimes we really do remember X separately from the evidence in support of X. And if X is something important this is probably to be encouraged.
This is my attempt to provide examples and a summarised view of the posts on "Against Doublethink" on the page How To Actually Change Your Mind.
What You Should Believe
Lets assume I am sitting down with my friend John and we each have incomplete and potentially inaccurate maps of a local mountain. When John says "My map has a bridge at grid reference 234567", I should add a note to my map saying "John's map has a bridge at grid reference 234567" *not* actually add the bridge to my map.
The same is true of beliefs. If Sarah tells me "the sky is green" I should, assuming she is not lying, add to my set of beliefs "Sarah believes the sky is green". What happens too often is that we directly add "The sky is green" to our beliefs. It is an overactive optimisation that works in most cases but causes occasional problems.
Taking the analogy a step further we can decide to question John about why he has drawn the bridge on his map. Then, depending on the reason, we can choose to draw the bridge on our map or not.
We can give our beliefs the same treatment. Upon asking Sarah why she believed the sky is green, if she said "someone told me" and couldn't provide further information I wouldn't choose to believe it. If, however, she said "I have seen it for myself" then I may choose to believe it, depending on my priors.
I Believe You Believe
The curious case is when someone says "I believe X". This can be meant a few ways:
The first case I do not have a problem with. It means your probability density has not yet shown a clear winner but you are giving me the answer that is in the lead at the moment. In this case I should add a note saying "You believes there is a bridge here, you are not very confident in the belief".
I don't have a problem with the second case either. I can have the belief "Angelina Jolie is attractive", someone else not have that belief, and we both be rational. This is because we are using different criteria for attractive. If I were to change it to a consistent definition of attractive it wouldn't be a problem e.g. The phrase "Angelina Jolie is regularly voted in the top 100 most attractive people in the world" doesn't require the phrase 'I believe...'.
The last case is even more curious. Lets assume that John (from our first example) says "I believe there is a bridge at grid reference 234567" but means it in the third case. I should add a note to my map saying "John has the following note on his map: 'I believe there is a bridge at grid reference 234567'". You would hope that the reason he has that note is because there is actually a bridge on his map. Unfortunately people are not that rational. You can have a cached belief that says "I believe X" even if you do not have "X" as a belief. By querying why they have that belief you should be able to work out if you should believe it, or even if they should.
To use the example from religion you can have the belief "I believe god exists" even if you do not have the belief "god exists".
Recommendations
I'm going to put myself on the line and give some recommendations: