Arthropod-borne viruses pose a major threat to global public health. Thus, innovative strategies for their control and prevention are urgently needed. Here, we exploit the natural capacity of viruses to generate defective viral genomes (DVGs) to their detriment. While DVGs have been described for most viruses, identifying which, if any, can be used as therapeutic agents remains a challenge. We present a combined experimental evolution and computational approach to triage DVG sequence space and pinpoint the fittest deletions, using Zika virus as an arbovirus model. This approach identifies fit DVGs that optimally interfere with wild-type virus infection. We show that the most fit DVGs conserve the open reading frame to maintain the translation of the remaining non-structural proteins, a characteristic that is fundamental across the flavivirus genus. Finally, we demonstrate that the high fitness DVG is antiviral in vivo both in the mammalian host and the mosquito vector, reducing transmission in the latter by up to 90%. Our approach establishes the method to interrogate the DVG fitness landscape, and enables the systematic identification of DVGs that show promise as human therapeutics and vector control strategies to mitigate arbovirus transmission and disease.