The Knuth-Morris-Pratt (KMP) algorithm efficiently searches for occurrences of a substring within a main string by preprocessing the pattern to determine the longest prefix which is also a suffix, thus avoiding unnecessary comparisons. This preprocessing step results in a time complexity of O(n + m), where n is the length of the text and m is the length of the pattern, making it highly efficient for large-scale text searching.