The Not So Illusory Truth Effect: Source Credibility and the Effect of Repetition on Belief

Reed Orchinik; David Rand; Rahul Bhui

In brief

The illusory truth effect—belief increasing with repetition—is often framed as a fixed cognitive bias. This paper argues it may reflect rational, adaptive inference: when a credible source repeats a claim, that repetition is informative evidence the source meant to convey the claim and did not misspeak. Two preregistered experiments (N=4,947 and N=2,484) show the effect is substantially smaller when the source is not credible, with most participants in low-credibility conditions showing no illusory truth effect at all.

Contributions

A formal Bayesian model of the illusory truth effect based on credibility-dependent inference rather than automatic fluency processing.
Empirical demonstration that repetition effects hinge critically on source credibility, contradicting cognitive bias accounts that predict source-insensitive effects.
Evidence that adaptation to source quality operates via intuitive rather than deliberative processing.
Large-scale preregistered experiments testing the effect in both headline-evaluation and politically sensitive contexts.

Method

Study 1: Theoretical model. A Bayesian formulation in which a receiver infers statement veracity from three pieces of information: the base rate of truth from the source (credibility α), the statement's plausibility γ, and the number of repetitions r. Repetition adds evidence only insofar as the receiver interprets it as the source's intentional restatement. When α > 0.5 (credible source), each repetition adds logit(α) to the log-odds of belief. When α < 0.5 (non-credible source), repetition signals falsity instead.

Study 2: Large preregistered experiment (N=4,947). Participants rated accuracy of headlines from a newsfeed. Source credibility was manipulated by varying the true:false ratio in the familiarization phase—high-quality condition: 75% true headlines; low-quality: 75% false. Repeated headlines (seen 3 times during familiarization) were re-rated alongside novel ones in the judgment phase. Same 6 repeated headlines appeared in both conditions, controlling for content variation.

Study 3: Replication with politically sensitive claims from a trusted 2024 U.S. presidential candidate, testing whether adaptation occurs even with strong prior beliefs.

Results

In the high-quality condition, repeated headlines were believed 2.22 percentage points (pp) more than novel ones (t = 6.34, p < 0.001). This effect was 1.05pp smaller in the low-quality condition (t = −2.98, p = 0.004), a roughly two-thirds reduction.

Critically, the median participant in the low-quality condition showed a small, non-significant decrease in belief for repeated vs. novel headlines (median effect = −0.18, 95% CI: [−1.14, 0.69]), contrasting sharply with the high-quality median effect of +2.21 (95% CI: [1.43, 3.11]).

A Bayesian finite-mixture model identified two strategies: one following the standard illusory truth heuristic (66.4% of high-quality participants), another associating repetition with falsity (63.0% of low-quality participants). Highly intuitive participants (above-median CRT score) showed stronger adaptation: in the low-quality condition, they demonstrated a strong belief-reduction effect (β = −3.05, t = −3.63, p < 0.001), while low-CRT participants showed smaller adaptation.

Study 3 confirmed the pattern with political content: repetition increased belief only for the trusted candidate.

Notes

Strengths:

Rigorous preregistration and large sample (4,947 participants) provide strong statistical power.
Direct experimental manipulation of source credibility while holding repeated content constant isolates the role of source evaluation.
Formal Bayesian model offers clear, testable predictions and reconciles seemingly contradictory findings in the literature (e.g., why some studies find diminishing returns to repetition).
Intuitive-vs.-deliberative analysis (via CRT) provides process evidence: adaptation does not require effortful reasoning.

Limitations:

Source credibility manipulated via content ratio in a newsfeed may not generalize to real-world source reputation built over time.
Headlines used are pre-tested for plausibility; effects may differ for highly implausible or emotionally charged claims.
Study 3 uses only one candidate; generalizability to other political contexts unclear.
Does not directly measure whether participants are consciously inferring source intent from repetition vs. intuitively adjusting beliefs.

Questions for follow-up:

How robust is adaptation when source credibility is uncertain or changes over time?
Do these effects persist after a delay, or does fluency dominate when cognitive resources are limited?
How do participants' own prior beliefs interact with source credibility and repetition?