A hundred research teams each measure the oral temperature of their own sample of healthy adults and build an interval: average ± multiplier × SE, with SE = SD / √n. In a simulation we can do what real life never allows: we choose the true average (36.8 °C, the dashed line), so we can check every interval. Teal intervals caught the truth; pink ones missed.
The interval is a property of the method: used again and again, a 95% method catches the truth about 95 times in 100. Any single interval either caught it or not, and with real data we never see which.