When an economy is at its "normal" or long-run level of production (i.e. potential GDP), there is an associated unemployment rate known as the "natural" rate of unemployment. This unemployment consists of frictional and structural unemployment but doesn't have any cyclical unemployment associated with business cycles. Therefore, it makes sense to think about how unemployment deviates from this natural rate when production goes above or below its normal level.
Okun originally stated that the economy experienced a 1 percentage point increase in unemployment for every 3 percentage point decrease GDP from its long-run level. Similarly, a 3 percentage point increase in GDP from its long-run level is associated with a 1 percentage point decrease in unemployment. In order to understand why the relationship between changes in output and changes in unemployment is not one-to-one, it's important to keep in mind that changes in output are also associated with changes in the labor force participation rate, changes in the number of hours worked per person, and changes in labor productivity. Okun estimated, for example, that a 3 percentage point increase in GDP from its long-run level corresponded to a 0.5 percentage point increase in the labor force participation rate, a 0.5 percentage point increase in the hours worked per employee, and a 1 percentage point increase in labor productivity (i.e. output per worker per hour), leaving the remaining 1 percentage point to be the change in the unemployment rate.
Since Okun's time, the relationship between changes in output and changes in unemployment has been estimated to be about 2 to 1 rather than the 3 to 1 that Okun originally proposed. (This ratio is also sensitive to both geography and time period.) In addition, economists have noted that the relationship between changes in output and changes in unemployment is not perfect, and Okun's Law should generally be taken as a rule of thumb as opposed to as an absolute governing principle since it is mainly a result found in the data rather than a conclusion derived from a theoretical prediction.