I’m learning and improving! The impact of chatbot learning evidence on customer tolerance of chatbot service failure

Purpose The rapid adoption of AI chatbots in diverse service contexts has been accompanied by increased service failures, making effective self-recovery strategies critical. This research examines the effect of linguistic forms of chatbot learning evidence on customer tolerance for service failures, aiming to provide actionable insights for effective chatbot service recovery. Design/methodology/approach We conducted four experiments with 688 participants, including students and professionals. Each experiment simulated distinct service contexts and manipulated learning evidence types (quantitative vs. qualitative) and disclosure timings (ex ante vs. ex post). Participants were randomly assigned to chatbot interactions with varying service failure scenarios to test the proposed model. Findings The results indicate that customers have a higher tolerance to chatbot service failures when chatbot learning evidence is provided, particularly with quantitative (vs. qualitative) evidence and that these effects are driven by perceived frustration. The boundary effect of disclosure timing of chatbot learning evidence indicates that the effect of linguistic form on customer tolerance of chatbot service failure is more effective in ex post disclosure. Originality/value This research pioneers the exploration of learning evidence’s role in chatbot service recovery, uncovering mechanisms (perceived frustration) and boundary conditions (disclosure timing).