Annotations are critical for training supervised learning models by providing labeled examples. This process includes identifying features or sentiments in data sets, which improves model accuracy and performance.