The Game Theory of Fake News

Contributions

• Develops an asymmetric, asynchronous coordination game between news publishers and readers, showing how transmitter strategies create engagement patterns that misrepresent receiver preferences
• Demonstrates theoretically that responsive misinformation publishers (who increase false-story production when engagement rises) can maintain high engagement with inattentive readers who nevertheless prefer accuracy
• Provides empirical evidence using Facebook engagement data and crowdsourced accuracy ratings that misinformation sites generate higher engagement with less accurate stories, contradicting the simple preference-for-novelty hypothesis
• Shows that reader preferences for accuracy are robust across general population and Twitter users with actual misinformation-site engagement
• Identifies two critical regions (F2 and T2) where transmitter strategies—not reader preferences—determine engagement patterns, with practical implications for content moderation

Method

A game-theoretic framework models the strategic interaction between transmitters (news outlets) who choose story accuracy rates and receivers (readers) who choose engagement based on perceived accuracy and past source behavior. The key insight is that transmitters can employ feedback strategies (e.g., increasing misinformation production when engagement rises) to induce readers with inattentive monitoring to engage more with inaccurate stories.

The model formalizes transmitter strategies with polynomial feedback functions: transmitters vary the probability of transmitting true vs. false stories based on recent engagement from prior stories. Receivers decide engagement probability based on perceived accuracy (with attention parameter $a_0$) and memory of past engagement outcomes (with attention parameter $a_1$). Simulations over $10^4$ rounds establish steady-state engagement correlations.

Empirically, three laboratory experiments test consumer preferences for accuracy:

• Experiment A: 1,000 articles from 40 outlets (20 misinformation, 20 mainstream), drawn from Crowdtangle. Participants rated 20 headlines' accuracy on a 7-point scale and willingness to click/share.
• Experiments B & C: 40 news cards (balanced on veracity and partisan lean) and 60 domains (20 mainstream, 20 hyper-partisan, 20 fake). Participants rated accuracy, sharing likelihood, and clicking likelihood for each story.

Accuracy for all headlines was independently rated by 1,000 Amazon Mechanical Turk workers, yielding 20,000 ratings validated against professional fact-checkers.

Results

Theoretical findings: Successful misinformation strategies enforce positive engagement-to-misinformation correlations (transmitters increase false stories as engagement rises), creating apparent preference for false stories even when receivers prefer accuracy and all readers value truth. Region F2 defines this critical dynamic: transmitter strategies determine engagement despite reader accuracy preferences. Conversely, mainstream publishers enforcing negative correlations (increasing true stories as engagement rises) produce opposite patterns (region T3).

Empirical findings:

• Misinformation sites show significant negative correlation between engagement and perceived accuracy ($p = 0.005$); mainstream sites show significant positive correlation ($p = 0.004$)
• Across all three experiments, participants strongly prefer accurate news and express willingness to click/share accurate stories at significantly higher rates ($p < 0.001$)
• The negative engagement-accuracy correlation for misinformation persists in Twitter users with prior misinformation engagement and general population samples
• Pattern is robust to partisanship controls and to using objective (professional fact-checker) vs. subjective (crowdsourced) accuracy ratings
• Effect replicates across different prominence metrics (shares, comments, reactions)

This pattern rules out the hypothesis that readers simply prefer novel or sensational false stories: the engagement gap emerges from responsive publisher strategies applied to inattentive but accuracy-preferring readers.

Notes

The paper reframes a persistent puzzle in misinformation research: why do false stories receive more engagement when people claim to prefer accuracy? Prior work has appealed to novelty preference, emotional appeal, or cognitive biases. This work provides a simpler mechanism—publisher strategy—that requires no assumption that people actually prefer falsehood. The model is elegant and the empirical validation is thorough, combining observational data (Facebook engagement, Crowdtangle), large-scale accuracy ratings, and experimental preference elicitation.

A significant implication is methodological: measuring engagement patterns to infer consumer preferences is unreliable when producers can adjust supply. The economics of information demand must account for the supply side. A second implication is practical: the pattern suggests that platforms or publishers promoting accuracy-aligned content, or that reduce microtargeting accuracy, should weaken the feedback loop driving misinformation spread.

The focus on receiver attention and memory as the mechanism for the F2/T2 effect is well-motivated and aligns with concurrent work on accuracy-nudge interventions showing that shifting attention to accuracy reduces misinformation sharing.