My obligatory 13 keys post:
They "seem" to work because they're arbitrary, and because Lichtman changes his definition of what he's predicting depending on who wins. The Thirteen Keys are the equivalent of a confidence trick dressed up in the guise of academia.
The first six keys have explicit, testable definitions. They give the impression of a rigorous, mathematically definable process. The remaining seven keys are completely arbitrary, lacking any testable definition, but cloaked in "you can trust me, I'm a professional" rhetoric to disguise that fact. Seven keys are also the number needed to decide a winner.
Originally, Lichtman said his Keys predicted the popular vote winner. He dropped that for his 2000 predictions, simply saying Gore would "win". After Gore lost, Lichtman claimed he was, as per his original work, just predicting the popular vote. After Trump lost the popular popular vote, Lichtman reversed himself, claiming that he was now predicting the "elected" President.
The Keys real title should be "well-informed smart guy guesses who will be President". If he was using a crystal ball or watching the flights of birds as a prop, everyone would dismiss him. But because he has successfully disguised his prop, he gets outsize attention, as witnessed by regular threads about him here. "Professor 8 for 10 in calling Presidential elections" would be treated quite differently, yet that is exactly what the Keys are.