When Generative AI “Thinks Visibly”: How Exposure to Chain-of-Thought Reasoning Influences Users’ Perceived Credibility

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Chain-of-Thought (CoT) reasoning refers to intermediate natural-language reasoning steps generated by large language models (LLMs) to derive an answer. This study examines how exposure to displayed CoT reasoning influences users’ perceived credibility of AI-generated content from a human-AI communication perspective. Drawing on the Heuristic-Systematic Model (HSM), we examine how CoT exposure is associated with perceived credibility through systematic and heuristic processing-related perceptions. AI-related task complexity is also examined as a moderator. A 2 × 2 between-subjects experiment revealed that displayed CoT reasoning raises perceived credibility, with all four perceptual pathways showing indirect associations. While its effects on credibility, accuracy, and intelligence intensify under high task complexity, its impact on transparency and social presence remains stable regardless of complexity. These findings extend XAI research by highlighting displayed reasoning as a dual-function communicative cue that shapes users’ evaluations of AI-generated content and their interactional perceptions of AI.