I never found this to be a mind puzzle, more so generally an unclear puzzle. The crucial aspect of the problem is that the presenter will always open a goat door, and in my experience people fail to be clear about this. Whatever lies behind your door, at least one of the other two doors contains a goat, and the presenter knows.
Now, if the host can open either of the two unpicked doors, the probability change depends on how he goes about choosing an unpicked door. Suppose he chooses either door with equal probability.
Strategy $(1):$ Don't change unless host opens a car door
If at first you chose the car door, with probability $1/3$, then whatever door the host opens will be a goat door and you won't change.
If at first you chose a goat door, with probability $2/3$, then with probability $1/2$ the host will open a car door and you will change to the car; otherwise, with probability $1/2$, host opens a goat door and you stick to your goat door.
Your overall chance of getting the car is $1/3+2/3\cdot 1/2=2/3$.
Strategy $(2):$ Always change
If at first you chose the car door, with probability $1/3$, then whatever door the host opens will be a goat door and you will change to the other goat door.
If at first you chose a goat door, with probability $2/3$, then with probability $1/2$ the host will open a car door and you will change to the car; otherwise, with probability $1/2$, host opens a goat door and you change to the other door, which contains the car.
Once again, your overall chance of getting the car is $1/3\cdot 0+2/3\cdot(1/2+1/2)=2/3$.
Conclusion: If the host chooses an unpicked door uniformly at random (and you're allowed to change to the car door if it is revealed), then both strategies have a $2/3$ probability of success.
Weird, huh? To further drive the point home about 'how random' matters, let's generalize the problem. Suppose that when you pick a goat door, so that when choosing an unpicked door there is one goat door and one car door, the host picks the goat door with probability $0\leq p \leq 1$.
With Strategy $(1)$, we get a $1/3\cdot 1 + 2/3\cdot (1-p)=(3-2p)/3$ probability of success.
With Strategy $(2)$, we get a $1/3\cdot 0 + 2/3\cdot 1=2/3$ probability of success.
Hence, when $3-2p<2\iff p>1/2$, Strategy $(2)$ is best. BUT, when $3-2p>2$ $\iff p<1/2$, the scenario changes, and Strategy $(1)$ is best.
Our initial uniform case, with $p=1/2$, is the critical case in which both strategies are equivalent. The classical problem has $p=1$.