The social media context interferes with truth discernment

Contributions

• Identification of the social media context as a cognitive interference: Demonstrates that the simple act of considering whether to share content on social media interferes with truth discernment, establishing a novel mechanism by which platform affordances shape misinformation vulnerability.
• Evidence for spillover over consistency mechanism: Shows that the effect operates primarily through cognitive distraction (spillover: thinking about sharing occupies cognitive resources needed for accuracy assessment) rather than motivated consistency-seeking (people adjusting beliefs to align sharing and accuracy judgments).
• Order-dependent effects: Reveals that the magnitude of accuracy-discernment reduction depends on whether sharing or accuracy is asked first, with a stronger effect when sharing is foregrounded (35% reduction vs. 18%).
• Cross-domain generalization: Replicates effects across two distinct domains (COVID-19 health statements and political headlines with images) with large sample sizes (N=3,157 total across waves).
• Implications for platform design: Suggests that centralizing sharing as a core affordance may have unintended negative consequences for users' ability to evaluate content truthfulness.

Method

The study employed two experimental waves with multiple conditions to isolate the effect of sharing considerations on accuracy judgment.

Participants and Design: Two waves with N=1,656 (Wave 1: COVID-19) and N=1,501 (Wave 2: Politics) American participants recruited via Prolific Academic. Participants were randomized to one of four conditions:

• Accuracy-Only: Rate the accuracy of true and false headlines
• Sharing-Only: Indicate whether they would share each headline on social media
• Accuracy-Sharing: Rate accuracy, then indicate sharing intention
• Sharing-Accuracy: Indicate sharing, then rate accuracy

Wave 1 (COVID-19): 24 headlines about COVID-19 (12 true, 12 false) presented as text without images or source information. Participants rate each headline's accuracy on a continuous scale (0 = definitely false, 1 = definitely true) and/or indicate willingness to share.

Wave 2 (Politics): Political headlines paired with images, presented with source information, likeable content, and commenting capability (in Sharing conditions). Same accuracy-rating and sharing-intention measures, with additional measurement of liking and commenting in wave 2's Sharing conditions.

Key manipulations: The critical independent variable is whether participants assess sharing, accuracy, or both, and in what order. This design allows comparing: (1) the marginal effect of asking about sharing on accuracy ratings, and (2) whether order matters (testing spillover vs. consistency mechanisms).

Results

Primary finding—Accuracy discernment: Participants showed significantly lower accuracy discernment (perceived accuracy of true minus false headlines) when sharing was also assessed, compared to accuracy-only baseline.

Wave 1 (COVID-19) results:

• Accuracy-Only baseline: d = 0.30 (accuracy discernment)
• Accuracy-Sharing: d = 0.25 (18% decrease, b = -0.070, p < 0.001)
• Sharing-Accuracy: d = 0.19 (35% decrease, b = -0.136, p < 0.001)
• Interaction effect: asking about sharing before accuracy led to significantly larger decrements (order X veracity, b = -0.066, p = 0.001)

Wave 2 (Politics) results:

• Baseline accuracy discernment lower overall but the pattern holds
• Accuracy-Sharing: 8% decrease in discernment, non-significant order interaction (b = -0.021, p = 0.093)
• Main effect of sharing: b = -0.035, p = 0.002

Mechanism testing:

• Spillover hypothesis (asking about sharing creates cognitive load/social media mindset unrelated to headline verity): Headline-level regression analysis found that the effect of sharing on accuracy was independent of how shareable each headline was in the Sharing-Only condition (p = 0.754). This disconfirms the consistency account (which would predict higher sharing likelihood for headlines people rated as more accurate).
• Consistency hypothesis (people adjust accuracy beliefs to justify sharing choices): The consistency model predicted a positive correlation between shareability and accuracy-judgment change, but the spillover model predicted a negative correlation based on baseline accuracy. Results significantly supported the spillover account (p < 0.001 for baseline perceived accuracy predicting effect size).

Partisanship moderation: In Wave 1, the effect was moderated by political partisan preference: Republicans showed larger accuracy-discernment decrements when sharing was asked (3-way interaction, b = -0.036, p = 0.047). In Wave 2 (political headlines), no significant moderation by partisanship (b = 0.010, p = 0.358).

Sharing intention effects: Interestingly, asking about accuracy increases sharing discernment (i.e., people become more selective about what they say they would share; b = 0.021, p = 0.017)—a reversal of the asymmetry seen for accuracy. Asking about accuracy pushes people into an accuracy-focused mindset that increases selectivity, while asking about sharing creates a sharing-focused mindset that decreases accuracy focus.

Notes

Strengths:

• Large and well-powered samples (N=3,157 total) with adequate statistical power to detect interaction effects.
• Two distinct experimental waves (health and politics) demonstrate generalizability beyond a single domain.
• Strong experimental logic isolating the spillover mechanism through order manipulations and headline-level mediation analyses.
• Preregistered hypotheses and transparent acknowledgment of which predictions were and were not supported.
• Addresses a theoretically important and practically salient problem: the role of platform affordances in shaping cognitive processes.

Weaknesses:

• Laboratory task may not fully capture real social media contexts where sharing decisions are made amid dozens of competing stimuli, network cues, and emotional arousal.
• The effect size, while statistically significant and meaningful, is moderate (18–35% reductions in a modeled context). Real-world impact magnitude remains uncertain.
• Order effects in Wave 2 (politics) are weaker and less robust than in Wave 1, suggesting domain-dependence that is not fully explained.
• The paper does not test interventions to mitigate the interference (e.g., separating accuracy and sharing interfaces, temporal delays, or redesigned layouts).
• Measurement of "sharing intention" is self-reported and hypothetical; actual sharing behavior on platforms may differ.

Follow-ups:

• Investigate design interventions—can interface redesigns that separate or reorder accuracy and sharing assessment reduce the interference?
• Test real sharing behavior on platforms with modified affordances to assess generalizability beyond stated intentions.
• Examine persistence: does the interference effect decay if people are repeatedly exposed to accuracy-salience cues?
• Explore individual-difference moderators: do cognitive-reflection ability, media literacy, or prior fact-checking experience buffer against the effect?