Advances in Research

Artificial intelligence is becoming embedded in organisational communication through message generation, conversational agents, monitoring systems, decision support, and automated service recovery. During crises, these functions offer speed, scale, consistency, and continuous availability, but they also change how stakeholders infer agency, competence, care, responsibility, and authenticity. This critical narrative review integrates crisis communication, organisational trust, human–automation trust, artificial intelligence-mediated communication, public relations, and service-recovery research to examine when AI-mediated communication is likely to sustain or erode stakeholder trust. The evidence indicates that no uniform ‘AI effect’ exists. Stakeholder responses depend on whether AI assists a human communicator or appears to replace one, whether the task is informational or relational, whether the crisis demands empathy and moral accountability, whether AI involvement is disclosed, and whether human responsibility remains visible. Technical or process transparency can strengthen perceived accountability and system trust, while simple disclosure that a message was produced with AI may reduce perceived legitimacy, authenticity, or credibility in some contexts. Anthropomorphic cues can increase social presence and perceived empathy, yet they may also raise expectations that become costly when the system fails. Evidence from crisis-response experiments, organisational studies, chatbot research, and service recovery therefore favours calibrated human–AI orchestration over indiscriminate automation. The review proposes that trustworthy AI-mediated crisis communication requires fit among task, crisis responsibility, stakeholder vulnerability, disclosure purpose, system reliability, and accountable human oversight. Current confidence is constrained by a literature dominated by short-term experiments, consumer and convenience samples, hypothetical crises, and limited longitudinal or field evidence. Research should move towards preregistered factorial studies, live crisis simulations, cross-cultural and multi-stakeholder designs, longitudinal trust measurement, and explicit testing of governance and escalation arrangements.