A cliffhanger works on the same mechanism as an open loop, applied at a structural boundary: the end of an episode, a chapter, a part one. The boundary is the point at which stopping is easiest, which is exactly why it is the point that should be least comfortable.
The distinction between a good and a cheap one is whether the tension is real. Cutting away mid-sentence is a cheap cliffhanger and audiences resent it because the resolution turns out to have been nothing. A real one ends on a decision made and not yet acted on, which is unresolved without being withheld.
It carries an obligation: the opening of the next unit has to pay it, quickly and without a recap that dissolves the tension it depended on.
What makes a cliffhanger work?
Real unresolved tension rather than a withheld reveal. Ending on a decision made and not yet acted on works; cutting away mid-sentence is resented because the resolution turns out to be nothing.
Should short-form video use cliffhangers?
Only with a genuine part two that is easy to reach. A cliffhanger with no accessible payoff is a broken promise, and the cost lands on every subsequent piece.