A meta-analysis of artificial intelligence and the customer brand relationship in tourism and hospitality

This meta-analysis pools 23 independent samples ( N = 9,875) on the association between customer-facing artificial intelligence (AI) and the customer-brand relationship in tourism and hospitality. Earlier reviews of this question are narrative or stop at technology acceptance; this one quantifies the relational link and extends it into the generative-AI period. Each effect size is a zero-order, construct-level correlation taken from the primary study’s correlation matrix, and no standardised path coefficients were pooled. A random-effects model (restricted maximum likelihood with the Knapp-Hartung adjustment) gives a pooled correlation of r = 0.62, 95% CI [0.54, 0.68], with a 95% prediction interval of [0.18, 0.85]. The analyses found no statistically reliable differences by technology (service robots, chatbots, generative AI, general AI tools; Q-between = 4.1, p = 0.25), by anthropomorphism (Q-between = 1.6, p = 0.21), by type of AI predictor (Q-between = 1.8, p = 0.42), by outcome family (Q-between = 6.4, p = 0.17), by sub-sector, or by region, although several subgroups hold only a few samples and the tests have limited power. Heterogeneity is considerable (I-squared = 96.4%) and is not accounted for by the coded moderators, the correlation source, or publication year. There is no sign of a small-study effect (Egger p = 0.99; trim-and-fill adjusted r = 0.59). The certainty of evidence is low (GRADE), and because every correlation comes from a single-source survey, the pooled estimate is best read as an upper bound. Interpreted within a stimulus-organism-response framework, the evidence shows a positive association that the coded artefact-level categories do not differentiate, but the large between-study variation and the cross-sectional, self-report designs mean the average should not be treated as a universal value.