Imagine 100 examples, with five belonging to the positive class. A classifier finds one of those five, raises one false alarm, and correctly labels the other 94 negatives. It gets 95 predictions right, so its accuracy is 95%.
But it found only one of five actual positives: 20% recall. And only one of its two positive predictions was correct: 50% precision.
Both metrics count the same correctly identified positive. Recall compares it with all actual positives; precision compares it with all predicted positives. That change in denominator tells us whether we are measuring how many positives were found or how many positive predictions were correct.
This Aflora article makes those groups visible with green and red dots, showing what each metric counts and why choosing a metric depends on which errors matter in the application.