If your estimation of the likelihood of God is negligible, then it may as well be zero.

This doesn't follow. For example, if you recite to me a 17 million digit number, my estimate that it is a prime is about 1 in a million by the prime number theorem. But, if I then find out that the number was in fact 2^57,885,161 -1, my estimate for it being prime goes up by a lot. So one can assign very small probabilities to things and still update strongly on evidence.