Some elaboration on ENO's already good explanation:
The TGP knows where you are looking, but it does not know what you are looking at. By default, it computes your intended target as being the point on the ground at the center of your camera, not any object that happens to be in-frame. By lasing your target (no need to latch - just give it a squirt and you should be fine, though latching will help if you are forgetful) you are providing the targeting system with an invaluable piece of information - distance to target.
Low-angle attacks will naturally suffer from a greater chance of being off-target if the target is not lased.
A simple graphic: