If a feature class is clipped to a specific AOI (area of interest) and we utilize the original layer file (.lyr) to represent the newly clipped polygons (in this instance) there will be unused symbols displayed in the classification of the unique values and in the final legend (see screen capture below).
The current solution is to manually select one by one (or alt select) to remove the unused classifications/symbology.
Would like to see ESRI add a simple sort ascending or descending tool to the “Count” column to easily sort out all the unused symoblogy. In this instance it was easy to select the 22 geological classifications and manually remove them. Yet, there are instances with other data sets that may contain100+ unique values that need to be manipulated and cleaned before final representation in a legend.