ANALYZING DECEPTION THROUGH THE VERIFIABILITY APPROACH WITH AI ASSISTANCE

Can the details you share reveal truth or deception? This study tests the Verifiability Approach (VA), which predicts that liars downplay independently checkable facts, whereas truth-tellers offer them to build credibility. Participants wrote truthful or fabricated accounts of recent activities. Claude 3.5 Sonnet flagged perceptual, spatial, and temporal cues, which coders classified as verifiable or unverifiable. ANOVAs and discriminant analyses compared detail types and group membership. We predicted that truth-tellers would supply more verifiable information and that these counts would best discriminate veracity. We also evaluated whether Claude boosted coding speed without harming reliability. The project (a) replicated Nahari et al.’s (2014) findings and (b) extended them with AI assistance. Results showed that: (1) truthful statements contained more total, verifiable, and even unverifiable details; (2) raw counts of verifiable details, not their proportion, distinguished groups above chance; (3) AI-assisted coding retained excellent reliability; and (4) Claude cut coding time by roughly half. Thus, the study confirms the core VA pattern and highlights large-language models as practical tools for scaling deception research.