The purpose of this paper is to examine when AI-assisted compassion signalling preserves credibility during consequential organisational change and when it instead weakens execution. Organisations increasingly have to communicate care while implementing change that imposes visible loss. Generative artificial intelligence (AI) makes compassion language fast, plentiful and inexpensive to produce.
This is a conceptual theory-building paper. It uses two illustrative vignettes to surface the phenomenon, integrates compassion research, signalling theory, strategic leadership, emotional labour, symbolic management, trust repair and human–AI communication, defines Synthetic Compassion as a distinct construct, develops a recipient inference mechanism, and derives propositions, a leadership playbook, governance requirements and a staged empirical pathway with illustrative measures.
Warm language is no longer sufficiently diagnostic once its production cost collapses. Synthetic Compassion supports cooperation when recipients can identify a named human owner, understand the process that follows the message, and expect workable recourse within a credible time frame. It backfires when delegation cues lower effort inference, blur ownership or reveal a remedy system that cannot absorb the volume created by scaled messaging.
The paper is conceptual and does not claim empirical validation. The vignettes are analytic illustrations rather than data. The manuscript therefore prioritises testable propositions, illustrative operationalisations, and a staged research agenda spanning vignette experiments, field audits and longitudinal designs.
The paper offers a streamlined leadership playbook consisting of a pre-release diagnostic, a three-state readiness screen, a stepwise implementation sequence, illustrative deployment heuristics for scale and recourse, and post-deployment monitoring indicators. It also specifies a rapid-cycle minimum bundle for crisis or time-compressed settings.
Better governance of AI-mediated care communication can reduce cynicism, preserve trust in institutions, and improve how employees, citizens, patients and service users experience disruptive change.
This paper introduces Synthetic Compassion as a bounded strategic leadership construct rather than a relabelling of generic impression management, organisational hypocrisy or broad AI trust. Its distinctive contribution is to explain an earlier-stage credibility mechanism: recipients may discount compassion language before the full action record is visible because scalable language shifts interpretive weight from warmth to ownership, procedural legibility and remedy capacity.
