Thanks for posting here @Novita Olivera
We usually strongly recommend automatic exposure tracking. The reason for this is that the automatic exposure events triggered by our SDK do not count toward your monthly event limit. If you were to instrument it incorrectly and send a load of Experiment exposure and assignment events, they would not count towards your monthly limit. If you were to instrument it incorrectly on a custom exposure event, each custom exposure event would count towards your traffic, while automatic exposure tracking is "free". Automatic exposure tracking is less susceptible to instrumentation issues. It is also easier to implement than a custom exposure event.
We recommend custom exposure tracking if you truly want to implement your own exposure event. However, we only recommend this if you are very comfortable with Experiment and know what you are doing. We do recommend testing custom exposure tracking in a dummy project before implementing it in production if you were to choose to do so. It can also be that a user may be assigned but never fire the exposure event if there is an issue and this could cause further analysis issues.
Note that Custom Exposure events should be used if you are using Amplitude Experiment with your back-end and not the client side.