Empathy is broadly celebrated as a virtue: In our lab’s data, participants are almost uniformly favorable toward empathy, even when they think about empathy toward outgroup members or people who do not share their values. Similarly, religious teachings and contemporary discourse alike encourage people to empathize across social divides.

But what happens when people actually empathize with morally questionable others? Across seven well-powered (four of which were publicly pre-registered) experiments and an internal meta-analysis, we found that third-party observers’ evaluations of empathizers are not uniformly positive. Instead, these evaluations are attuned to the target of empathy and can be even more negative than evaluations of nonempathizers (Wang & Todd, 2021, JPSP).

Participants in a typical paradigm read about an interaction between two strangers, one of whom described an experience they had recently to the other person, who responded in an empathic or nonempathic way. Participants’ evaluations of the empathizer depended on the target of empathy: For example, participants liked and respected an empathizer more when she empathized with a hospital worker, but not when she empathized with a White supremacist. Moreover, although actively condemning a White supremacist—a decidedly not empathic response—lowered warmth, condemnation elicited greater respect and liking from participants.

These findings highlight that people’s abstract ideas about empathy do not always reflect their concrete, nuanced evaluations of empathy. More broadly, our lab’s research considers how empathy affects not just people who receive empathy (i.e., effects within the empathic dyad), but also third-party observers (i.e., effects beyond the dyad). Importantly, it suggests a conundrum: Although people are encouraged to empathize with disliked others, actually doing so could harm their reputation.