It's worse than just 5% because p hacking. Every analysis has many parameters, and you can choose the most significant out of thousands of possibilities. On to

It's worse than just 5% because p hacking. Every analysis has many parameters, and you can choose the most significant out of thousands of possibilities. On top of this, the more interesting results are the most surprising. And therefore the least likely to be true a priori. So the chance of an interesting study being true is even lower.