Check out this new article by Gennaro, Derksen, Abdelrahman and colleagues in Nature Scientific Reports, which puts counterspeech strategies to the test in a real-world Twitter experiment. The goal? To see if encouraging users to adopt the perspective of minority groups can reduce the spread and amplification of xenophobic hate speech online.
The researchers ran a large-scale field experiment on Twitter using a sample of 2,441 tweets identified as xenophobic. These came from unverified US-based users with at least five followers and no bot-like or sarcastic behaviour. Each tweet received one of four randomly assigned counterspeech interventions or no intervention (control). The interventions included:
Traditional perspective-taking (“Imagine if you were…”),
Analogical perspective-taking (“Remember when someone attacked you… now imagine being on the receiving end”),
Perspective-getting (sharing a story from a friend/colleague),
and a disapproval message that simply stated the behaviour was not acceptable.
These messages were sent via controlled “sockpuppet” accounts—human-managed accounts made to look neutral and authentic. Each user’s tweet behaviour was then tracked over four weeks.
The results? The three perspective-based interventions had a modest but statistically significant impact on users’ behaviour. The main effects:
8.7% of users in the perspective conditions deleted their xenophobic tweet, compared to 5.1% in the control group.
Perspective-based counterspeech also reduced the likes on hate tweets by around 0.13 standard deviations, meaning fewer bystanders were endorsing or amplifying them.
However, the interventions didn’t significantly change how many future xenophobic tweets users posted.
Interestingly, analogical perspective-taking (getting users to reflect on their own negative experiences) slightly outperformed the other two strategies, though differences weren’t statistically significant.
From a policy perspective, this study shows that even subtle, non-coercive counterspeech—when framed around empathy and shared experience—can have measurable effects on both senders and bystanders. It’s especially relevant at a time when platforms like Twitter/X are cutting back moderation efforts. These findings suggest that user-driven strategies could fill some of that gap.
For researchers, the study offers a replicable model for testing interventions in the wild. It also highlights the importance of measuring both direct and indirect effects—especially amplification dynamics like likes and retweets, which shape visibility and spread. While the effects aren’t huge, they’re consistent, and point to a scalable strategy for reducing online hate.