The answer is: it depends. I am a big advocate of using ref to help your illustrations be more specific (and not just in terms of how things appear, but with research as well), but I am also very cautious about letting one’s ref dictate the image. It’s easy to simply “copy” reference, which depending on the source could possibly be plagiarism, but more often than not leads to deadened illustrations that don’t execute your compositions as well as they could have.
So yes, I’ll do a quick little photoshoot (most of the time just using Photobooth in OSX) to get a pose, or hands or something down, and yes, I’ll stockpile relevant images when I’m working on projects, but only AFTER the thumbnailing stage, and my composition is kind of locked in.
Increasingly, I’ve started to just draw without any planning or reference as a way to bring back some immediacy into my drawing, outside of my usual illustration process. I think good drawings should have elements of both control and spontaneity, so it’s important to practice both.
In general: the more realistically rendered your work is, the more reference you need. In my classes, I talk about Laplace’s Demon: the theoretical omnipotent creature who could supposedly know all of the information in the universe, and therefore see into the future. Modern physics tells us that this is impossible; there is literally not enough computational power in the universe to know the positioning of everything. Same goes with rendering and light: you cannot possibly know exactly how light will interact with all the different elements in a scene without help.
I could go on for days about using reference. But in the meantime, check out (of course) Gurney Journey, the Muddy Colors blog, and how Norman Rockwell would exaggerate his photo ref.