We provide causal evidence that the context in which harmful conduct occurs shapes how it is evaluated. We study this using a vignette-based survey experiment that holds behaviour constant while randomising whether an incident of sexual harassment occurs online or in person. Online settings generate a systematic discount in perceived seriousness (7% of the mean), with larger effects on willingness to report (13% of the mean) and preferred sanctions. The discount is concentrated in image-based harassment and larger among male respondents. In addition, it is not moderated by either direct or vicarious experience of harassment, suggesting that it may reflect normative perceptions of online harm as less serious rather than lack of exposure alone. These context-dependent distortions have implications for the enforcement of emerging legal protections, victim support, and the design of public communication around digital abuse.