Sharing of misinformation is habitual, not just lazy or biased

Contributions

• Demonstrates that social media platform reward structures shape habitual sharing behavior independent of individual cognitive or motivational factors
• Shows the 15% most habitual news sharers are responsible for 37–39% of false news spread across experimental conditions
• Provides evidence that habitual sharing overrides accuracy concerns (Study 2) and political bias (Study 3): habits persist despite conflicting attitudes and norms
• Demonstrates habits can be redirected toward accurate information through training with aligned rewards, without sacrificing user engagement

Method

Study 1 (N=200): Participants saw 8 true and 8 false headlines in Facebook format and chose whether to share each. Sharing habit strength was measured by past frequency and automaticity (Self-Report Behavioral Automaticity Index, SRBAI). Mixed-effects models predicted sharing decisions from headline veracity and habit strength.

Study 2 (N=839): Extended Study 1 with a manipulation: half the participants rated headline accuracy first, then made sharing decisions; half chose whether to share, then rated accuracy. This tested whether foregrounding accuracy would reduce habitual false-news sharing.

Study 3 (N=836): Participants chose whether to share headlines differing in political orientation (liberal vs. conservative), with political concordance/discordance coded relative to participants' stated political identity. Same design as Study 2 but varying political alignment instead of veracity.

Study 4 (N=601): Simulated habit formation via instrumental learning. Participants completed 80 training trials sharing/not sharing headlines. In the accuracy condition, they received rewards (+5 points) for sharing true information or not sharing false information. In the misinformation condition, rewards were reversed. A control group received no rewards. After training, all participants made 16 sharing decisions without rewards ("as they normally would"). Tested whether training contingencies persisted when rewards ended.

Results

Study 1: Strong-habit users (±1 SD) shared 43% of true and 38% of false headlines (not significantly different); weak-habit users shared 15% of true and 6% of false (3.9× more discerning). The 15% most habitual sharers accounted for 37% of all false news shared.

Study 2: Accuracy rating first reduced false-news sharing overall (from 13% to 9%) but did NOT interact with habit strength. Habitual users continued to share 42% of true and 26% of false headlines regardless of condition; weak-habit users were 1.9× more discerning. The 15% most habitual participants shared 39% of misinformation across experimental conditions.

Study 3: Weak-habit users showed strong partisan bias (9.22× more concordant than discordant sharing); strong-habit users showed weaker bias (4.27× ratio). Habitual users shared politically discordant information at higher rates (16% vs. 3% for weak-habit users), consistent with outcome-insensitive habits.

Study 4—Training phase: Accuracy training increased discernment (72% true vs. 26% false, d=1.08); misinformation training reduced it (48% true vs. 43% false, d=0.10); control (45% true vs. 19% false, d=0.70).
Study 4—Test phase (after rewards ended): Accuracy-trained users maintained high discernment (66% true vs. 24% false, d=1.40); control and misinformation groups converged (40% vs. 24% and 54% vs. 33%, d=0.76 and d=0.69). Trained habits persisted despite changed reward contingencies, demonstrating both the malleability of sharing habits and the possibility of redirecting them toward accuracy.

Notes

Strengths:

• Rigorous experimental design across four studies with large samples and preregistered hypotheses (except Study 1)
• Key insight: shifts attribution of misinformation from individual user deficits (laziness, bias, poor reasoning) to system-level reward structures—a systems-frame approach
• Study 4 is the strongest contribution: demonstrates habits can be formed and redirected, proving false-news spread via habits is not inevitable
• Findings robust across two measures of habit strength (frequency and automaticity)

Methodological notes:

• Participants were MTurk/Prolific users with Facebook accounts; generalization to less digitally native populations unclear
• Headlines presented in simplified Facebook format (photo + source + headline); real feed algorithms and recommendation systems may amplify or diminish effects
• Political-concordance study (Study 3) used moderate headlines, so effects may not generalize to highly polarized or extreme content

Open questions:

• Do habit-redirecting interventions (Study 4 paradigm) scale to real social media platforms?
• How do platform-level interventions (timeline disruption, algorithmic de-prioritization) interact with individual habits?
• Do habits differ across platforms (Twitter, TikTok, etc.) with different reward structures and user bases?
• What role do algorithmic recommendations play in reinforcing or disrupting sharing habits post-training?