To accomplish the goal implied by the OP, we'd have to completely disable the Arrange Marriage diplo-action and handle it all magically with events. I think that would suck, on net.
Can marriage outcomes improve with the help of auto-marrying folks off to each other? Yes, although do not overestimate the difficulty and CPU consumption required for soft-code to do this, considering all the variables that we want it to consider-- including strategic planning re: line of succession and such, nor the amount of effort that would have to go into such a beast to completely cover the marriage problem for the AI.
In the end with as good of a complete implementation as soft-code could do, you'd still be able to find dozens upon dozens of cases where AIs married very poorly, because it is-- believe it or not-- a risky game with much luck involved in most of the factors requisite for its [subjective] success.
Could we do more limited stuff that tries to let the proper AI function (esp. since they *have* been putting a lot of work into this consistently for a couple patches now, which is a relief) but only intervenes when it seems that the AI has fallen on its face, with simpler logic? Sure, and we may at some point. But, again, none of those improvements would address players gaming the marriage AI (even if it's easy, at least when you get to pick the initial marriage proposal on day-1 before the AI's had a chance to do anything for itself).
These days, I have a lot more trouble than I once did securing land through marriages. However, offensive alliances are a thing now (for some reason, rather than defensive), so I imagine it'll take some tweaking before AI valuation of everything related to NAPs and alliances gets smarter.