I have to disagree here. You're right in that they manage to guard settlers OK. And yes it's therefore a natural question to ask, "If A, why not B?" And if that were the only issue, I suggest they would already have done it.
The problem lies in the difference between the strategic movement of a settler versus a GG. The settler needs to get from town of birth to a rough area of the map, that's all. The AI probably knows where it's going. It doesn't need to synchronize with any other troop movements. It just needs to be guarded. Easy. The GG must synchronize with the ebb and flow of its other troops, and it must do so while predicting the flow of opposing troops. I would suggest this is a task that is three orders of magnitude harder for the AI, and one that I expect we will never see satisfactorily solved by any AI in a game produced by a small company outside the realm of full military simulation and billion dollar budgets.