An experimental online study on the impact of negative social media comments on anxiety and mood

The study examines the specific emotional impact of negative social media comments on adults, addressing a gap in prior work that focused mainly on adolescents. Social media’s anonymity and online disinhibition can foster hostile interactions (e.g., sarcasm, personal attacks, discrimination), which have been linked to psychological distress and depressive symptoms. Building on evidence that social media use can harm mental wellbeing and that comment tone (e.g., uncivil vs. polite) influences emotional responses, the research tests whether exposure to negative comments elevates state anxiety and reduces mood among adults. The authors also predicted women would be more affected than men based on prior literature showing higher anxiety in women. The study’s purpose is to isolate the effects of comment valence (negative, neutral, positive) on adult mood and anxiety and to evaluate potential moderating roles of gender and, exploratorily, age, given differences in social media use across age groups.

The paper situates its research within work showing broad links between social media and mental health, including quasi-experimental and randomized trials reporting declines in wellbeing and increases in loneliness and depression with social media exposure or use. It highlights that comment tone matters: hostile, uncivil comments evoke stronger negative emotion than polite disagreements. Mechanisms discussed include anonymity reducing moral sensitivity and online disinhibition, increasing the likelihood of negative commenting and cyberbullying. Prior studies show adolescents with anxiety and low self-esteem are particularly sensitive to negative comments, potentially entering a vicious cycle of worsening anxiety. While emerging adults’ coping with cyberbullying has been examined, less is known about older adults. The authors therefore focus on adults, considering gender differences (women often report higher anxiety) and age-related differences in social media usage, motivations, and socioemotional responses (e.g., FOMO, social comparison), which may modulate comment impact.

Participants: A priori power analysis (G*Power) indicated a minimum N=128 for detecting a medium effect (f=0.25) in a 2 (Gender) × 3 (Comment Type) between-subjects ANOVA (α=0.05, power=0.80). A total of 142 English-speaking adults were recruited via Prolific; all provided informed consent and were compensated £7. Twelve withdrew at the end and one failed attention checks, leaving 129 participants (85 female, 43 male, 1 other), age M=37.0, SD=12.73, range 18–73. The study was approved by the University of Warwick Psychology Ethics Committee and preregistered on OSF. Design and stimuli: Between-subjects design with factors Gender (male, female) and Comment Type (negative, neutral, positive). Participants were randomly assigned to comment type; gender was quasi-experimental. To control content, AI-generated blogs and comments were produced using ChatGPT following prompts archived on OSF. Eight short blogs (75–100 words) across pairs of topics (shopping/building Lego; gaming/drinking; gardening/baking; upscale dinner/trip) were accompanied by theme-matched images (Unsplash). For each selected blog, participants viewed ten comments of one valence (positive, neutral, or negative). Across four choices, each participant saw 40 comments; the full corpus contained 240 comments (8×3×10). A minor error caused ~1/3 of neutral comments to include a fabricated username (e.g., “@roomie”), judged unlikely to affect results. Measures: Mood was measured with the Brief Mood Introspection Scale (BMIS; 16 adjectives, rated 1–4). Anxiety was measured with the STAI-S (Form X-1; 20 items, rated 1–4). Attention checks were embedded in both questionnaires. Procedure: After consent and demographics, participants imagined being a popular blogger and selected one of two topics (no actual posting required; choice was to increase engagement). After a brief delay, they viewed comments matching their assigned condition (negative, neutral, or positive). This choose-and-view cycle repeated four times. Participants then completed the BMIS and STAI-S. A 90-second humorous YouTube video was shown to alleviate potential negative affect. Debriefing followed and compensation via Prolific. Scoring and analysis: BMIS pleasant mood scores were computed by summing pleasant items and reverse-scoring unpleasant items, dividing by 16 to retain a 1–4 scale. STAI-S scores were computed by summing negative items and reverse-scoring positive items, dividing by 20 to retain a 1–4 scale. Analyses were conducted in JASP. Exclusions: those using "prefer not to say" for one or more responses were excluded in relevant analyses; the single "other" gender participant was excluded due to group size. Levene’s tests assessed variance equality; Kruskal–Wallis confirmed effects when variances were unequal. Exploratory analyses split age by the median (35 years) into younger (n=61, M=26.72, SD=4.23) and older (n=60, M=48.13, SD=9.37) groups for age-related effects not preregistered. BMIS sub-scores (arousal vs. calm, positive vs. tired, negative vs. relaxed) were examined post hoc per Mayer and Gaschke.

Topic choices did not affect anxiety or mood (all t-tests ns; anxiety p>.078; mood sub-scores p>.062). Anxiety (STAI-S): A 2×3 ANOVA (Gender × Comment Type) showed a significant main effect of Comment Type, F(2,115)=19.74, p<.001, ηp²=0.256. Post-hoc tests: negative comments produced higher anxiety than neutral or positive (means: 2.42 vs. 1.77 and 1.55; both p<.001), with neutral vs. positive not significantly different. Variance inequality (Levene’s p<.001) led to confirmation with Kruskal–Wallis, H(2)=26.83, p<.001. Gender main effect was marginal, F(1,115)=2.77, p=.099, ηp²=0.024; Gender × Comment Type interaction marginal, F(2,115)=2.37, p=.098, ηp²=0.040. Exploratory age analysis (Younger vs. Older; 2×2×3 ANOVA) revealed a significant Age Group main effect, F(1,109)=6.73, p=.011, ηp²=0.058, with younger participants reporting higher anxiety than older (2.07 vs. 1.77). No other age-related interactions were significant (all p>.200). Pleasant mood (BMIS): Significant main effect of Comment Type, F(2,117)=28.13, p<.001, ηp²=0.325; negative comments reduced pleasant mood relative to neutral and positive (means: 2.37 vs. 3.05 and 3.25). Levene’s test was nonsignificant (p=.103). Gender effects were nonsignificant (all p>.182). Adding Age Group showed a significant Age effect, F(2,111)=5.10, p=.026, ηp²=0.044, with younger participants reporting less pleasant mood than older (2.79 vs. 3.01). No Age Group interactions were significant (all p>.090). BMIS sub-scores: Arousal vs. calm showed main effects of Gender, F(1,117)=6.15, p=.015, ηp²=0.050 (males higher arousal: 2.39 vs. 2.26), and Comment Type, F(1,117)=4.69, p=.012, ηp²=0.073 (negative>neutral: 2.44 vs. 2.24). Adding Age Group revealed an Age Group × Comment Type interaction, F(1,111)=4.11, p=.019, ηp²=0.069; older participants showed minimal arousal changes to negative or positive comments compared to neutral (0.03 or −0.05) versus larger changes among younger participants (0.29 or 0.18). Positive vs. tired mood: main effects of Comment Type, F(2,112)=16.35, p<.001, η²=0.226, and Age Group, F(1,112)=6.01, p=.016, η²=0.051. Negative vs. relaxed mood: main effects of Comment Type, F(2,114)=27.24, p<.001, η²=0.323, and Age Group, F(1,114)=4.23, p=.042, η²=0.036; interactions with Age Group did not reach significance (p>.057).

The findings confirm the hypotheses that negative social media comments causally increase anxiety and reduce pleasant mood among adults, paralleling prior adolescent-focused work and extending it to a broad adult age range. Effects of gender were minimal: anxiety differences were only marginal and mood showed no significant gender effect, although men reported slightly higher arousal overall and appeared somewhat more anxious under negative or neutral comments. Exploratory analyses revealed younger adults are more susceptible to anxiety and lower mood than older adults when exposed to negative comments, suggesting age-related differences in sensitivity that may reflect identity development, social comparison tendencies, and heightened Fear of Missing Out. The arousal results align with the pattern in anxiety, consistent with theory linking anxiety to increased arousal. Potential mechanisms for the small gender differences include social pressures on men around emotional restraint and body image within online environments, which may amplify anxiety under negative evaluation; BMIS’s composite scoring could also mask differential emotional components (e.g., sadness vs. anxiety) between genders. The results underscore the broader psychological impact of online negativity and suggest the need for platform-level and clinical strategies to mitigate adverse effects, with tailored interventions: cognitive reappraisal training for younger adults and digital literacy enhancement for older adults.

This study demonstrates that exposure to negative social media comments significantly elevates state anxiety and reduces pleasant mood among adults, with particularly pronounced effects in younger adults. Gender effects were minimal overall, though men exhibited higher arousal. By using controlled, AI-generated comment stimuli and validated measures, the research isolates comment valence as a causal factor in adult emotional responses. The work contributes evidence for age-sensitive vulnerabilities to online negativity and highlights practical implications for mental health and platform design. Future research should increase ecological validity by using real platforms and diverse comment features (e.g., emojis, metaphors), incorporate physiological and behavioral indicators of affect, include a culturally diverse sample, and test non-lifestyle themes (e.g., social/political identity) to examine generalizability and mechanisms.

Potential observer effects may have influenced responses due to awareness of being in an experiment. Comments were AI-generated and may not capture the complexity of real social media interactions (e.g., emojis, metaphors, varied tone). The sample was primarily adult native English speakers from Western backgrounds, limiting cultural generalizability. The study focused on lifestyle-related themes, so effects may differ for social or political content. Individual differences (e.g., social support, self-efficacy, personality, emotional intelligence) were not measured and could moderate responses. Some neutral comments included fabricated usernames, though deemed unlikely to affect outcomes. Despite a realistic design, simulated environments may not reflect full real-world dynamics.