UTC is calculated in three steps. In the first step, the free atomic time scale (EAL) is calculated with a weighted average of about 420 atomic clocks located in 85 time laboratories around the world. International Atomic Time (TAI) is obtained in the second step by steering the frequency of EAL on the Primary and Secondary Frequency Standards (PSFS) and finally, UTC is obtained by adding the leap second. More than 10 Primary and 4 Secondary Frequency Standard (PSFS) regularly contribute to Circular T. Their frequency is evaluated against the frequency of TAI and reported in Section 3 of Circular T. With an apposite algorithm all the PSFS values are combined to determine d, defined as the opposite of y(TAI). As the frequency of the TAI must be in agreement with the frequency of the PSFS, the value of d must be as close as possible to zero. The current practice followed by the BIPM is based on the observation of the value of d against its uncertainty (indicated with u in Section 3 of Circular T) and only if the value is larger than twice its uncertainty (2 u), a correction, of typically 0.2 10-16, is planned to be applied in the future computation. To give the time to the laboratories contributing to UTC to apply the foreseen steering and due to the latency of UTC computation, this correction will have effect two months later. As the application of a frequency correction has a negative impact on the stability of TAI, a good balance needs to be found between the improvement of the accuracy of TAI and preserving its stability. In general, the BIPM avoids (if possible) repetitive steering and to apply strong corrections. In this paper, we analyse different strategies for optimal steering of EAL frequency. We start by characterizing EAL frequency stability over the last few years. After that different simulated steering based on UTC data from 2018-2022 will be presented. The different strategies differ in the frequency and amount of steering corrections ranging from more frequent small corrections to less frequent but bigger ones. The strategies will be evaluated in terms of the resulting TAI performances (accuracy and stability). The impact of the delay in the steering application (2 months) will also be taken into account in the analyses. The United States Naval Observatory (USNO) rubidium fountains will be used as reference to evaluate the obtained results. Finally, we will present some conclusions for the benefit of future Circular T computations.