Adaptive Intuitions Shape Susceptibility to Misinformation

Contributions

• Challenges truth-default and Spinozan theories by demonstrating that intuitive susceptibility to misinformation reflects ecological adaptation, not cognitive failure
• Provides experimental evidence that intuitions flexibly adapt to base rates of true vs. false content in simulated social media environments
• Uses drift-diffusion modeling (DDM) to decompose intuitive vs. deliberative components of judgment, showing the mechanism underlying the effect is a durable shift in intuitive starting point
• Demonstrates these adaptations occur even with accuracy incentives, suggesting they operate at an automatic level
• Offers Bayesian framing of intuition as functioning like prior beliefs about content accuracy in an information environment

Method

Pre-registered experiment (N=2,810 after exclusions) with exposure and evaluation phases:

Exposure phase: Participants randomly assigned to either:

• GoodFeed: 80% true, 20% false headlines (40 true, 10 false)
• BadFeed: 20% true, 80% false headlines (10 true, 40 false)

Participants rated 50 headlines as accurate or inaccurate; no feedback was given.

Evaluation phase: All participants evaluated 30 evenly split headlines (15 true, 15 false, unknown base rate change).

Headlines drawn from a set of 110 stimuli balanced on partisanship and pre-tested plausibility. True headlines rated as more plausible (4.01 vs. 3.16 on 6-point scale) but equally partisan to false ones. Participants maintained overall accuracy while spending less time per headline, suggesting adaptation was automatic.

Computational modeling: Drift-diffusion models (DDM) decomposed responses into intuitive (starting bias) and deliberative (drift rate) components. Initial bias was the key mechanism.

Results

Error rates:

• GoodFeed participants made 12.8pp more errors on false headlines than baseline (4.86pp increase, 95% CI [2.78, 6.95], t=4.67, p<.001)
• BadFeed participants misidentified true headlines at 7.7pp higher rates than GoodFeed participants (2.94pp, 95% CI [1.04, 4.84], t=3.10, p=.003)
• Differences were robust across multiple regression specifications

Learning during exposure phase:

• Error differences emerge early (by headlines 20–30) and grow throughout exposure
• Errors on common headline types decrease in both conditions (false errors in BadFeed: 2.9–4.1pp decrease; true errors in GoodFeed: 1.47–1.56pp decrease)
• Participants maintained overall accuracy despite reduced per-headline time, demonstrating efficiency

Evaluation phase persistence:

• Differences persist in 50/50 evaluation phase but begin to decline as participants gradually learn the new base rate
• The initial intuitive bias shows "stickiness"—only partial adjustment despite clear evidence of environmental change

DDM decomposition:

• Initial bias shifts in the direction of the base rate (more bias toward "true" in GoodFeed, toward "false" in BadFeed)
• No evidence of fixed truth-default; instead, bias is environmentally responsive
• Drift rate changes minimally by condition, indicating deliberative processing is relatively stable

Replication: These effects replicate in a follow-up experiment with accuracy incentives, supporting the ecological adaptation interpretation.

Notes

Strengths:

• Large, pre-registered sample (N=2,810) with clear experimental manipulation of base rate
• Novel theoretical insight: reframes intuitive misinformation vulnerability as ecological adaptation rather than cognitive failure
• Sophisticated computational modeling (DDM) identifying mechanism (shift in starting bias, not deliberative capacity)
• Evaluation phase design elegantly tests persistence of adaptations in changed environment
• Robustness checks include regression specifications, headline fixed effects, and follow-up with incentives

Limitations:

• Simulated news feed may not fully capture dynamic of real social media (no social context, explicit accuracy task)
• Environment change between phases not signaled to participants; unclear if same effects occur with explicit awareness of shift
• Individual differences (e.g., in analytic thinking, political sophistication) not examined; may moderate adaptive capacity
• Effect sizes are meaningful but moderate (4.86pp increase in errors on false headlines); unclear if this magnitude is concerning in real-world contexts where stakes are higher
• Generalizability to other content domains (e.g., conspiracy theories, health misinformation) untested

Interpretation: The paper challenges a decades-long assumption in misinformation research: that susceptibility to false content is primarily a failure of cognitive systems. Instead, the evidence suggests it reflects systems working as designed for typical information environments. In the American mediascape where most content is true, a truth-favoring intuition is rational. But this adaptation can be maladaptive when environments change (e.g., sudden influx of coordinated misinformation, algorithmic amplification of false content, or deliberate ecosystem degradation). The DDM evidence that intuitions are flexible (not immutable) offers both hope—people can adapt to better information environments—and caution: intentional degradation of information quality could systematically shift population-level susceptibility.