<p>Orthologs, evolutionarily related genes that diverged through speciation, are mutually the closest related sequences in different species. Consequently, they are ideal candidates for identifying functionally equivalent genes across taxa, a prerequisite for transferring gene function information from model to non-model organisms in silico. However, orthologs are not immune to functional divergence. Failing to recognize such divergent instances results in spurious functional annotation transfer. Here, we propose to treat the functional equivalence of orthologs as a null hypothesis that must be critically tested rather than assumed. This requires integrating several lines of evidence to evaluate both changes in an ortholog’s network of molecular interactions and alterations in its biochemical activity. We outline how such activity shifts can be assessed using increasingly fine-grained analyses, including comparisons of protein feature architectures and predicted 3D structures. While some orthology resources incorporate aspects of this evidence, such assessments are often manual and not scalable. We argue for a systematic, multi-level perspective to detect functional divergence prior to annotation transfer. To support the broader adoption of this approach, we offer methodological recommendations and practical examples that demonstrate the value of this framework in large-scale comparative genomics.</p>