Thermowell!!!
I cut the curved end off a SS racking cane and sealed the bottom with food grade sealant. I then run my thermistor down the tube as far as I can get, fill the top end around the wire with paper towels or foam to keep the outside air from getting inside the tube.
That set up is run through the center part of my carboy cap, the other outlet on the cap has a blow off tube on it. The end of my thermistor is usually dead center, vertically and horizontally, in the beer and from the edges of the carboy.
I've left a separate thermometer taped to the side of the carboy to measure the difference and it was dramatic. The sensor on the outside of the carboy seemed to cool too quickly and if running my fridge, would have cut the cooling before a noticeable drop in temperature inside the carboy. The glass conducts the cool, outside air too well no matter how well you insulate the probe from air. In a matter of minutes the thermometer taped on the carboy showed a drop in temp of 5 degrees that would have shut off the fridge were it in control, however the wort temp as registered by the thermowell had not dropped a single degree at that point.
It works better than nothing, but without a thermowell you can't measure the wort temp, you're still just measuring the carboy temp.
It's not much of a jump to create a little thermowell and I'm convinced it maintains a MUCH more accurate reading of the actively fermenting wort than the other methods.
Of course, this is applicable only to my experience with GLASS carboys. I would think buckets may not introduce as much of a variable in the readings but that's speculation on my part with no experience to back it up.
EDIT: It appears other people have had different results. Perhaps I'm not insulating the thermistor as well as you guys, but the above reflects my experience.