Toxic content and user engagement on social media: Evidence from a field experiment
Abstract
Most social media users have encountered harassment online, but there is scarce evidence of how this type of toxic content impacts engagement. In a pre-registered browser extension field experiment, we randomly hid toxic content for six weeks on Facebook, Twitter, and YouTube. Lowering exposure to toxicity reduced advertising impressions, time spent, and other measures of engagement, and reduced the toxicity of user-generated content. A survey experiment provides evidence that toxicity triggers curiosity and that engagement and welfare are not necessarily aligned. Taken together, our results suggest that platforms face a trade-off between curbing toxicity and increasing engagement.
Study specs
Pre-registered browser extension field experiment on Facebook, Twitter, and YouTube to randomly hide toxic content for six weeks; supplemented with a survey experiment.
- Institution
- Mozilla Foundation,Columbia University,Bocconi University,Stanford University,University of Warwick
- Discipline
- Social Science
- Study Type
- Experimental Study
- Year
- 2025
- Human Data Platform
- Prolific
- Source
- View Source Google Scholar
Measured Outcomes
Impact of reduced exposure to toxic content on advertising impressions, time spent, engagement, and user-generated content toxicity; explored curiosity and alignment between engagement and welfare.
Peer Review & Critical Discussion
Potential Selection Bias in 2023 Cohort
The participant pool shows a concerning overrepresentation of users from high-income demographics. Looking at Table 3, we can see that 78% of respondents had annual incomes above $75k, which significantly limits the generalizability of these findings to broader populations.
Non-naive Participants Issue
I've noticed a methodological concern regarding participant naivety. Given that Prolific users often complete multiple studies, there's a real risk that participants had prior exposure to similar experimental paradigms, which could confound the results.
RLHF Applicability to This Study Design
The implications for RLHF training pipelines are understated. If we accept the authors' conclusions about preference stability, this has direct consequences for how we should structure reward model training. The temporal decay effect described in Section 4.2 is particularly relevant.
Verify your expertise to join discussion
Create an account and verify your credentials to participate in peer discussions.