A parent says: 'If you finish your homework, then you can play video games.' When has the parent broken the promise? Only if you finish your homework and are still not allowed to play. If you finish and play, the promise is kept. If you don't finish, the promise says nothing about what happens, so whether you play or not, it hasn't been broken. That is the truth table of P→Q: false in one row (P true, Q false) and true in the other three.
The two true rows with a false P surprise people, but they matter. 'If n is divisible by 4, then n is even' should be true for every whole number. Try n=6: not divisible by 4, but even. Try n=7: neither. If those rows counted as false, a correct theorem would fail on numbers it never talked about. A conditional only makes a claim about the cases where its hypothesis holds.
Reversing the arrow gives the converse Q→P, which is a different claim. 'If a shape is a square, then it has four sides' is true; 'if a shape has four sides, then it is a square' is false, and any rectangle that is not a square shows it. Watch the phrase 'only if' too: 'P only if Q' means P→Q. 'You may vote only if you are 18 or older' says that voting requires being 18, not that every 18-year-old votes.
Think of it like this. A conditional is a vending machine's promise: put in the right coins and you get a snack. The machine has failed only if you pay and get nothing. If you never pay, getting nothing (or a lucky free snack) proves nothing against it. The picture has one limit: a conditional is about truth values, not about one thing causing another.
Core idea. A conditional if P then Q is false only when P is true but Q is false.