As a warmonger I used to settle them. I've come full circle - I often make them attached GG's now. I have 3 highly favored GG type units:
GG medic: Well, most people are familiar with this. I'll comment that mine usually winds up from a combat I axe origin. It gets medic, then a GG pushes it to medic III morale, giving it 2 moves. It will tend to stay an axe forever. Even regular combat axes defend before it so don't expect to lose it unless your stack goes squish.
The CR Sword ----> Mace---->Rifle---->Infantry route. GG gets CR III, then combat, then march after combat III, then more combat (which is a LOT of experience, just be happy with Combat III CR III). Swords are well-protected against counters by axes and spears, so this GG won't come up in defense to be lost very often. It also has very high odds when other units are around 60% or worse. Very handy.
The Combat VI Mounted unit: These are...easily countered by spearmen. But those are easily countered by axes. Mounted tends to have the highest base str in its time until infantry show up, and starting with knights or elephants this is a very nasty unit, especially with march. A combat VI march knight upgraded eventually to cavalry will win you a lot of battles your siege or other troops would lose, and at high odds, too.
Now that we have a means to keep these super units alive, why use them instead of a settled GG? Simply put, speed on returns. Yes, it's possible to settled 4-6 gg's into a city to make super units right out of the gate. You don't get these GG's until late usually though. A super CR sword is available very early, on the 1st or 2nd GG.
"but siege takes their hp down anyway"!
Yes, that's true, but how much siege do you want to give up? In my experience, it's best to keep as many units alive per city capture or stack fight as possible. What GG attachments do is allow you to reliably kill top defenders after no or minimal siege damage. This includes nasties like CG archers/longbows and so forth. A recent example comes from the current NC game, where I repeatedly did this with (breaking my above tendencies - wish I'd gone CR for one of them) a combination of a Combat VI maceman and a combat VI elephant (both supported by the axe medic). Having GG's of this nature in my stack allowed for ONE suicide CATAPULT, against LONGBOWS, then a successful capture of a city garrisoned by 5 units - 3 of them longbows but only 2 of those longbows promoted (AI had whipped the 3rd). Classical siege, and only a small handful of maces, the rest elephants, and I lost exactly ONE unit taking a city. The most I lost was vs the target AI (which was actually ahead in tech but pretty well on parity militarily) was 3, taking the capitol with about 8-10 units in it IIRC.
In other words, attaching GG's allow your few combat units to attack at the strength of an era ahead of the target. The scariest two medieval combat gg's are the 17.5 strength knight that heals super fast (and may even heal after attacking) and the mace that hits more like a knight with CR III (very high odds against cities). Both of these super units are well-covered by a couple shock maces or xbows - meaning they won't defend! This is important - you need them alive to get use from them. Even the mere elephant will have str 14 with this setup. What will elephants defend against? Horse archers and knights. At str 14, the AI will tend not to even touch that, especially when there are more elephants to protect it if it gets hurt

.
The fun part comes with 2-3 attached combat GG's with your medic GG (very reasonable to get 3 relatively quickly with early wars). Few cities have more than 2 defenders that are actually strong on defense like longbows with CG, meaning after that other troops (or even more siege, now that it will have a much better chance of survival) can move in to clean up.
When a GG has 70+ xp, you'd have to wonder if the benefit from settling it would outweigh the advantage it bought. In many cases a fight saved me a catapult in hammers. Also note that early vassalage or theo will buy you a 2nd promo, meaning it would take several gg's to make a difference after them. In many cases I'd rather conserve units.