Abstract This study examines the effect of anthropomorphic cues in AI-powered chat bots on consumer’ purchase intention, investigating the roles of trust (affective pathway) and perceived usefulness (cognitive pathway) in this relationship. A quantitative cross-sectional survey design was used, collecting data from 170 experienced chat bot users across various digital service contexts comprising of e-commerce, online banking and airline booking platforms. Empirically validated multi-item construct measured anthropomorphism, trust, perceived usefulness and purchase intention. Data were studied using descriptive statistics, Pearson correlation and mediation analysis applying both classical product-of-coefficients (a x b) and bootstrapping methods with 1000 re samples to establish reliable confidence intervals. Results show that anthropomorphic cues significantly influence both trust (a = 0.335, p < 0.001) and perceived usefulness (a = 0.383, p < 0.001), supporting H1 and H2. Both mediators positively affect purchase intention (trust: b = 0.665, p < 0.001; perceived usefulness: b = 0.754, p < 0.001), confirming H3 and H4. mediation analysis reveals significant indirect effects through trust (indirect effect = 0.223, 95% CI [0.095, 0.301]) and perceived usefulness (indirect effect = 0.296, 95% CI [0.157, 0.382]), with both demonstrating partial mediation. Perceived usefulness exhibited a marginally stronger mediating effect that trust, suggesting cognitive evaluations may exert slightly greater influence on purchase decisions in task-oriented contexts. This study advances understanding of human-AI interaction by empirically validating a dual-path mediation framework integrating affective (trust) and cognitive (perceived usefulness) mechanism, an integrated approach previously unexplored in chat bot research. It also extends Technology Acceptance Model (TAM) by incorporating anthropomorphism as a predictor and demonstrating that emotional engagement operates alongside functional evaluations. The rigorous methodological approach combining classical and bootstrapped estimates strengthens confidence in the findings.