What counts as a prompt?
A prompt is whatever text you hand the model to respond to. At its smallest it is a single question. In practice it is usually more: instructions about the task, a few examples, and some background the model needs. A language model works by continuing from what it is given, so the prompt is not a formality wrapped around the real request. It is the request. Change the wording and you change the output.
That is also why the same model can seem sharp or useless depending on who is driving it. A vague prompt leaves the model guessing at context you never supplied, and it will fill the gap with something plausible whether or not it is right.
What makes a prompt work better?
The reliable moves are unglamorous. State the task plainly and say what format you want the answer in. Give the context the model cannot know on its own. When the output shape is specific, show one example rather than describing it. And prefer saying what to do over a list of things to avoid, since a positive instruction is easier to follow than a wall of prohibitions.
There is a cost to remember: every prompt is made of tokens, the chunks of text a model reads, and a longer prompt eats more of the context window, the fixed budget of tokens the model can hold at once. Padding a prompt with detail it does not need can crowd out the part that matters. Clear and complete beats long.