Contributions
- Unified framework integrating base-rate, fluency, and memory-based inferences into a single model of truth judgment
- Comprehensive review of illusory truth effect across decades of research, with mechanisms and boundary conditions
- Evidence that knowledge alone cannot overcome fluency-based illusions; people rely on fluency even when they "know better"
- Practical recommendations for misinformation correction targeting each of the three inference routes
Method
A narrative literature review integrating cognitive and social psychology. The paper identifies three classes of inference people rely on when judging claims:
Base rates: Following Bayesian cognition, people adopt a prior belief that most claims are true. This "bias to say true" is rational in everyday life but leaves people credulous when exposed to false claims. Repetition, distraction, or source ambiguity make unbelieving harder, as people must expend cognitive effort to reject information they initially accept.
Feelings as evidence: People interpret subjective experiences—particularly ease of processing (fluency)—as signals of truth. Repeated exposure increases fluency, leading to illusory truth: claims seem truer when repeated, even when they contradict known facts or are explicitly labeled false. Fluency effects are robust to individual differences in fluid intelligence and cognitive style. Even sadness, which reduces gullibility generally, cannot override fluency effects when illusions depend on mixed lists of old and new items.
Consistency with memory: People accept claims matching their stored knowledge and reject mismatches. This can be diagnostic, but people often misremember, forget source information, or fail to notice contradictions. When source information fades but fluency remains, people default to fluency-based judgments. Providing an accuracy focus at encoding can sometimes prevent illusory truth, but only when people possess relevant knowledge—without knowledge, people rely on fluency regardless.
Results
Illusory truth is medium-to-large in effect size. Meta-analyses show Cohen's ds of 0.39–0.49 for illusory truth across over 50 studies. The effect persists for days and withstands explicit warnings. Contrary to intuition, knowledge does not always protect: people show fluency-conditional effects, relying on fluency when they lack relevant knowledge, and knowledge-conditional effects, relying on knowledge only when they possess it.
Corrections are difficult and often backfire. Correcting a single misconception can inadvertently increase belief in related false claims. Debunking messages that repeat the misinformation can increase fluency of the false claim. Even when corrections succeed, people forget them faster than the original misconception.
Mismatches to base rates are hard to detect. People accept partial matches between claims and memory, allowing misattribution of false claims to reputable sources and acceptance of plausible—but false—elaborations.
Mood, visuals, and framing matter. Sad moods reduce gullibility but not fluency-based illusions. Accompanying claims with photos increases perceived truth. Deepfakes exploit both visual credibility and fluency. Difficult-to-read fonts reduce illusory truth, but only when people attend to processing difficulty.
Notes
Strengths: Integrates decades of fragmented research into a unified framework; practical focus on interventions; honest about where knowledge fails to protect (e.g., fluency-conditional model). The distinction between base-rate, fluency, and memory-based inferences clarifies why different interventions work (or don't) in different contexts. Strong coverage of illusory truth—a robust effect central to misinformation spread.
Limitations: No novel experiments; some recommendations (e.g., "use difficult fonts") rest on weak effect sizes or small samples. The paper does not model interactions between all three cues simultaneously, which may be important in real-world settings where visual, source, and fluency cues converge. Limited discussion of age differences in truth judgment, despite evidence that older adults share misinformation more. The generalization to high-stakes claims (political, medical) is assumed but not always tested; some findings rely on trivial claims or in-lab exposure.
Future directions: Understanding how affect, source credibility, and fluency interact in social-media environments. Modeling lifespan effects, especially in older adults who are over-represented among misinformation spreaders. Testing interventions in naturalistic social-media settings rather than labs.