Normally, I watch TF2 gameplay, which allows me to look at the maps without the distraction of being shot at.
This tends to get me excited about making something that someone would want to play, and looking at the maps makes me think about what makes it fun, which leads me to want to make a fun map that uses similar fun mechanics.
On top of that, since I'm quite early in the mapping journey, I frequently realise new things about TF2's design that good maps support and bad maps don't, and that makes me want to surpass my earlier maps by making a new map that incorporates these new design principles.
As a side note, lurking in voice chat and watching imps satisfies both of these, because I get to watch TF2 gameplay and hear people think/give feedback on design elements.
On top of this, I find that I have a 'store' of cool ideas - I find that inspiration strikes randomly, with some small dependence on the factors I talked about above, so I amass new ideas at a fairly decent rate - I then evaluate these ideas, and if they're too few in number and/or not good enough to make into a full map, I 'store them away'. This tends to leave me with new knowledge that I can make fun mechanics out of later, and with knowledge you gain even later, you can sometimes make good mechanics out of your stored mechanics.
For instance, in the map I released half an hour ago, I had wanted for a long time to make "SD but it's good" - you could call that a 'stored cool idea'. Previous analysis of KOTH and present analysis of class balance and choice led me to make mechanics which suited the gameplay of that map and turned it into "KOTH with extra steps", thus turning stored cool ideas and knowledge into a map.