Domenico, G. D., Mangió, F., & Dineva, D. (2026). don’t you know that you’re toxic? how influencer‐driven misinformation fuels online toxicity. Psychology & Marketing. https://doi.org/10.1002/mar.70106
Summary
This paper investigates how the source of brand-related misinformation shapes toxic audience responses, arguing that social media influencers (SMIs) provoke more frequent and qualitatively distinct toxicity than regular users posting identical content. Drawing on a large multiplatform dataset (101 posts, 48,821 comments across six platforms, 2020–2023), the authors reframe online toxicity as a source-amplified, relational phenomenon rather than a property of message content or individual cognitive bias. They identify two influencer-specific mechanisms—brand-related misinformation legitimation and community enmeshment—through which influencer credibility and parasocial bonds are converted into sustained collective antagonism, sustaining toxic echo chambers.
Key Contributions
- Reframes online toxicity as source-amplified and relational, shifting focus away from content- and cognition-centric accounts of misinformation.
- Extends typologies of online incivility by identifying emergent influencer-specific patterns (flame-bait firestorms, toxic debunking).
- Integrates source credibility, parasocial interaction, and social influence theories into a unified framework in which influencer authority becomes infrastructural to misinformation-driven toxicity.
- Demonstrates that compliance, identification, and internalization act as collective alignment mechanisms sustaining toxicity.
- Delivers ecosystem-level managerial guidance (publishers, platforms, people) and introduces a validated, large-scale, cross-industry brand misinformation dataset and mixed-method protocol.
Methods
An empirics-first, three-stage sequential mixed-method design combining top-down automated text analysis, bottom-up topic modeling, and theory-building thematic analysis. Toxicity was detected with Google’s Perspective API (threshold 0.6), extensively validated and cross-checked against Detoxify and Hurtlex. Logistic regression predicted comment-level toxicity from source type with 15 controls plus robustness checks (propensity score matching, Gaussian copula for endogeneity). A Biterm Topic Model on the top quartile of toxic comments yielded 41 coherent topics grouped into five categories, with category-level OLS regressions (robust SEs, FDR adjustment). Hybrid thematic analysis of 1,800 comments and 45 SMI posts achieved high intercoder reliability (κ = 0.87–0.96).
Findings
- Regular-user posts have ~44% lower odds of eliciting toxic comments than influencer posts (OR = 0.56); predicted toxicity is 3.8% for SMIs vs. 2.2% for users.
- A toxicity–engagement spiral operates for influencers (more engagement → more toxicity) but reverses for regular users.
- Influencers produce significantly more toxicity specifically in sociopolitical misinformation (OR = 1.87), with no significant difference for commercial or health/safety content.
- Toxicity is highest on low-pseudonymity, identity-based platforms (e.g., Meta), contradicting the assumption that anonymity drives incivility.
- Five toxicity categories emerged: anti-brand reactions, C2C conflicts, flame-bait firestorms, toxic debunking, and trolling/flaming.
- Regular users elicit diverse toxicity across multiple targets, whereas influencers trigger homogeneous flame-bait firestorms aimed at the implicated brand.
- Two mechanisms sustain toxic echo chambers: misinformation legitimation (amplifying/sheltering → compliance) and community enmeshment (bonding/endearing → identification and internalization).
Connections
This work sits within information-disorder scholarship that scrutinizes the amplifying role of influential accounts and platform affordances, resonating with research on how prominent actors shape misinformation diffusion such as DeVerna2025-dl and Pierri2025-hm. Its argument that toxicity is source-driven and sustained through parasocial and community mechanisms connects to studies of online radicalization dynamics like Marwick2025-ov and Frischlich2025-vn. The paper’s platform-comparative and toxicity-focused empirical framing also complements incivility and hostile-discourse research such as Rossini2026-jn.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts